CN111428127B - Personalized event recommendation method and system integrating theme matching and bidirectional preference - Google Patents

Abstract

The invention discloses a personalized event recommendation method and system integrating theme matching and bidirectional preference. Firstly, extracting topic information of an event and a historical event participated by a user by using a document topic generation model LDA, and calculating topic matching degree of the user and the event; secondly, considering the social network recommendation based on the event from the two-way angles of the user and the event, constructing preference models of the user and the event, respectively obtaining user preference scores and event preference scores, and more completely mining preference relations from the two angles of the user and the event; finally, the user-event pair matching degree is combined with the user event bi-directional preference linear weighted combination to obtain the final user-event pair comprehensive score, and the ordered TOP-K user-event pairs are used as recommendation results. The performance of the recommendation algorithm of the scheme is superior to that of the traditional recommendation scheme, and the personalized preference of the user can be well predicted, so that the purpose of personalized recommendation is achieved.

Description

Personalized event recommendation method and system integrating topic matching and two-way preference

Technical Field

The present invention relates to the technical field of information recommendation, and in particular to a personalized event recommendation method and system integrating topic matching and two-way preference.

Background Art

With the rapid development of the Internet and computer technology, traditional social networks have also developed in different innovative directions in recent years, and some special types of new social networks have been formed, such as location-based social networks (LBSN), which mainly form social relationships based on users' geographic check-in information, and another complex heterogeneous social network that combines online and offline - event-based social networks (EBSN). Different from the friendship established between acquaintances in traditional social networks, users in event-based social networks establish interpersonal relationships through social activities. Users join online interest groups and offline collective social activities based on their own interests or common points.

Event-based social networks are in the process of rapid development, and more and more users choose to participate in social activities in event-based social networks. On event-based social network platforms, users can join various online groups. Organizers or users in the group can initiate and participate in any offline social activities, such as parties, hiking, sports activities, concerts, etc., and share information with other users.

Event-based social networks can provide users with social services that combine online and offline, helping users initiate and develop personalized event participation plans. Users form online group relationships through common interests and initiate offline gatherings online. Event-based social networks have more extensive social attributes than location-based social networks. Existing work has shown that event social networks have better recommendation features than traditional social networks in recommendation systems.

Most current event-based social network recommendations are mainly based on user-based feature preferences. Although they take into account the social influence of the event organizer, they are not representative enough of the potential attractiveness of the event. On the other hand, the influence of thematic factors only takes the event theme as one of the recommendation factors, and rarely considers the user theme factor and its matching degree with the event theme.

Summary of the invention

In view of this, it is necessary to provide a personalized event recommendation method and system that combines several main types of contextual information to calculate user preferences and event potential preferences, and finally integrates topic matching and user-event two-way preferences.

A personalized event recommendation method integrating topic matching and bidirectional preference includes the following steps:

Step 1: Use the document topic generation model LDA to extract the topic information of the event, and obtain the user topic information based on the historical event records in which the user participated, calculate the topics of the new event and the user's historical events, and use the JS divergence algorithm to calculate the topic matching score of the user-event pair;

Step 2: construct a user preference model and an event preference model respectively, and calculate the user preference score and the event preference score respectively;

Step three, use the Bayesian personalized ranking algorithm BPR to learn the weight parameters of user preference scores and event preference scores, obtain the user-event bidirectional preference score, linearly weighted combine the topic matching score and the bidirectional preference score to obtain the final recommendation score of the user-event pair, and recommend the top K events after sorting to the user.

Furthermore, the document topic generation model LDA in step one has a three-layer generative Bayesian network structure, including documents, topics and words, wherein both document-topic and topic-word obey multinomial distribution; each document selects a topic with a certain probability, and selects a word from this topic with a certain probability, and the topic in any document conforms to the Dirichlet distribution, through which the relationship between texts is discovered.

Furthermore, the calculation of the topics of the new event and the user's historical events in step 1 uses the JS divergence algorithm to calculate the topic matching score of the user-event pair, and the specific steps include:

Step 1-1, all event description contents are grouped into a document set D and stop words are removed, the document set D is input into a document topic generation model LDA, and the topic distribution of each event is obtained respectively;

Remove stop words and punctuation marks from all event contents, and regard the document contents after removing noise interference words as the set D of all documents, which is input into the LDA topic model to generate document Joint distribution of topics and words , as shown in formula (1):

(1);

Then the Gibbs sampling method is used to estimate the two unknown parameters in the model: event topic distribution and keyword distribution ;

Step 1-2, calculate the topic distribution similarity between the target user's historical events and new events according to the JS divergence algorithm;

According to formula (1), the topic distribution of all events has been generated , given an event and They have topic distribution , first calculate the JS divergence between the two through the JS divergence method , as shown in formula (2):

(2);

in, , Represents KL divergence, which is used to describe two probability distributions and The difference between them is calculated as shown in formula (3):

(3);

Combining equation (2) and equation (3), we can get the event and The topic similarity is , as shown in formula (4):

(4);

Among them, the topic similarity of the event The value of lies in [0,1], and the closer the value is to 1, the higher the event similarity;

Steps 1-3: average the similarities of all historical events of the target user to obtain the topic matching score between the user and the new event;

by Represents the number of historical events of the target user, taking the average value of all similarities of the target user As the topic matching score between the user and the new event, as shown in formula (5):

(5);

According to the constructed topic matching model, the final To measure the topic matching relationship between target users and new events.

Furthermore, the user preference model constructed in step 2 constructs the user's single factor preference from three aspects: geographical location, social relationship, and time factor, specifically including:

Step 2-1-1, build a geographic location preference model:

The location preference model calculates the probability that the target user will participate in an event held at that location. The kernel density estimation (KDE) method is used to model the two-dimensional location distribution of the events that the user participates in. The normalized event participation probability is used to represent the user's preference for the location. The longitude and latitude coordinates of the event location are represented by ( Lx, Ly ), and the set of locations where the user has historically participated in events is represented by L ( u ). Then the KDE function for user u is As shown in formula (6):

(6);

Where, l _i =( Lx _i ,Ly _i ) ^T represents the two-dimensional vector of the longitude and latitude coordinates of the event location, m _l ( u,l _i ) represents the frequency of user u participating in activities held at geographical location l _i , σ represents the size of the neighborhood window (bandwidth), N represents the number of location samples, and K ( • ) represents the Gaussian kernel function, which is defined as shown in formula (7):

(7);

Combining equations (6) and (7), we can define the probability that user u will participate in an event held at location l , as shown in equation (8):

(8);

Normalize the probability to get the user's preference score for geographic location , as shown in formula (9):

(9);

The denominator represents the target user’s maximum event participation probability;

Step 2-1-2, build a social relationship preference model:

In the user social relationship network, users will join at least one or more interest groups online and choose to participate in events and activities released by different groups. The user's social relationship preferences are judged by the user's online group relationships, which mainly include two types of interactive relationships;

The first type, user-group relevance, is defined as the interaction relationship between users and all the groups they belong to and between users and events created within the groups. G ( u ) represents the set of groups to which events user u participates. Then the user-group relevance is It can be expressed as shown in formula (10):

(10);

Where, m _p ( u,g ) represents the set of events that user u has participated in in the group to which he belongs;

The second type is the intra-group user relevance. The intra-group user relevance is defined by the similarity of the friends in the target user’s group. The similarity between the target user and the users in the group is calculated. , as shown in formula (11):

(11);

Among them, sim ( u _i ,u _j ) represents the similarity between user u _i and user u _j in the same group, as shown in formula (12);

(12);

Normalize s ( u,g ) to , as shown in formula (13):

(13);

Combining the above two interactive relationships, users belonging to the same group tend to participate in events created by other users in these groups. Combining the correlation between users and groups and the correlation between users in groups, the social preference score of user u about online group g is obtained. , as shown in formula (14):

(14);

in, As a weight control parameter, in social relationship networks, setting the preference association between the target user and the group is as important as the association between users in the group. The value of is set to 0.5;

Step 2-1-3, construct a time factor preference model:

The time factor of the event is an important preference factor that needs to be considered when calculating user preferences; the new event e that the user can choose to participate in is represented as a 7*24-dimensional event time vector , when a new event occurs in a specific time period of a week, the vector component value of the time period is set to 1, otherwise it is 0; in the time preference model, users are represented as user time vectors according to the historical event records of the users. , as shown in formula (15):

(15);

Among them, Eu represents the set of historical events that the target user has participated in, and then calculates the cosine similarity between the _user time vector and the new event time vector , as shown in formula (16):

(16);

For new events ,user The similarity can be obtained according to formula (16): , normalize the similarity to get the user's time preference score for the event , as shown in formula (17):

(17);

Furthermore, the calculation of the user preference score in step 2 specifically includes:

For the geographic location preference model, the geographic location preference score is expressed by predicting the probability of the user participating in events held at the location; for the social relationship preference model, the social preference score of the target user is calculated from two aspects: the relationship between the target user and the group and the correlation with the users in the group; for the time factor preference model, a unified vector representation of the two granularities of date and hour is constructed, and the similarity of the user-event pair is calculated based on this as the time preference score of the target user; these three single-factor preferences are combined to form a user preference perception model, and the three single-factor preferences are linearly combined to obtain the overall preference score of user u for event e , as shown in formula (18):

(18);

in, They represent the user's preference scores on three single factors: geographic location, social relationship, and time.

Furthermore, the event preference model constructed in step 2 constructs single factor preference of events from two aspects: event location popularity and event organizer influence, specifically including:

Step 2-2-1, construct event location popularity preference model:

The popularity of a geographic location is calculated based on the frequency of visits to the location by user u and the users in the online group g to which he joins;

First, define the popularity of the event location l _e with respect to user u , as shown in formula (19):

(19);

Among them, the molecule is the frequency of user u participating in activities held at geographic location l _e , and the denominator is the maximum frequency of locations that user u has visited historically; similarly, define the popularity of geographic location l _e with respect to group g to which user u belongs , as shown in formula (20):

(20);

The numerator represents the frequency of each user in group g participating in practical activities at location l , and the denominator is the maximum frequency of locations that group members have visited historically. This allows us to calculate the popularity of geographic location l _e with respect to users in group g . and Define the total popularity of the location of the recommended event to the target user u as , as shown in formula (21):

(twenty one);

Step 2-2-2, build the influence preference model of event organizers:

First, the influence of the event organizer on the target user. The organizer's reputation or influence is used to express the implicit preference of the event. Define the influence of the event on user u , as shown in formula (22):

(twenty two);

in, represents the set of events held by organizer u _h that user u has participated in, E _h is the set of all events held by organizer u _h ;

Second, the influence of the event organizer in the group. For the online group where the target user is located, the influence of the event in the group is expressed by the frequency ratio of users participating in the event. The influence of the user in the group is expressed by It is expressed as shown in formula (23):

(twenty three);

Among them, U _g represents the group The collection of users in Indicates user Participants are organized by A collection of events held, express In the group The event collection held in the group; the comprehensive influence score of the event organizer is obtained by combining the influence of the event organizer on the target users and the users in the group. , as shown in formula (24):

(twenty four);

Furthermore, the calculation of the event preference score in step 2 specifically includes:

For new events that have not occurred, the preference for the event is expressed by calculating the event location popularity and event organizer influence of the new event; for the constructed event location popularity and event organizer influence Linear combination, calculate the preference score of event e for user u , as shown in formula (25):

(25);

Furthermore, the user-event bidirectional preference score is obtained in step 3, and the topic matching score and the bidirectional preference score are linearly weighted combined to obtain the final recommendation score of the user-event pair. The specific steps include:

Step 3-1, find the bidirectional preference for the user-event pair:

Assume that the preference score weights of users and events are and , weighted fusion of the two to obtain the user event bidirectional preference score ; Convert the problem of bidirectional preference scoring into finding the weight vector of two preference scores, and choose to use implicit feedback as training data to learn the weight vector;

We select the learning algorithm BPR based on Bayesian maximum likelihood estimation to rank the weights and learn the correct ranking order of user-event pairs based on the implicit feedback data of users on events, so that the events in which users participate are ranked before new events or other events. First, we define the maximum posterior probability , as shown in formula (26):

(26);

Among them, θ represents the weight vector, R represents the set of all user-event pairs, The definition is as shown in formula (27);

(27);

Among them, Indicates user of user-event pairs, and For users event Ranked The previous probability is shown in formula (28):

(28);

in, Bidirectional preference score , ; In order to facilitate optimization, assume Obeying the normal distribution with a mean of 0, the final optimization objective function is derived , as shown in formula (29):

(29);

in, represents the regularization term coefficient. The objective function is optimized by maximizing the implicit interactive feedback data of user events to obtain the optimal weight parameter vector. The stochastic gradient descent algorithm SGD is used to solve the optimization problem. In the iterative process, the user-event pairs of the target user are randomly extracted from the training set to update the weight vector. , the update process is shown in formula (30):

(30);

in, is the learning rate, ; Through the above learning process, the training set and hyperparameters can be automatically scored according to user event preferences and Find the weight vector , thus obtaining a two-way preference score ;

Step 3-2, combine topic matching and bidirectional preference to obtain the final recommendation score for the user-event pair:

Firstly, the LDA topic model is used to extract event topics and obtain the topic matching scores of users and events. Secondly, the preference models of users and events are constructed according to the user event context information in EBSN, and the bidirectional preference scores of users and events are obtained through the BPR learning algorithm. Finally, the topic matching scores are calculated. Bidirectional preference scoring with user events Linear weighted summation is used to obtain the final user-event recommendation score , as shown in formula (31):

(31);

in, is a weight parameter, which is usually set manually based on experience, and the optimal setting will be determined through experiments.

And, a system for implementing personalized event recommendation integrating topic matching and two-way preference, which is used to implement the personalized event recommendation method integrating topic matching and two-way preference as described in any one of the above items, and the implementation system includes:

The document topic generation module is used to extract the topics of the user's historical events and new events, and calculate the topic distribution and word distribution of the events. The topic similarity between the user's historical events and new events is used to represent the topic matching degree, which is integrated into the recommendation model as one of the key factors for recommendation to perform event recommendation.

Build a user preference module to construct the user's single factor preference from three aspects: geographic location, social relationship, and time factor, and weightedly integrate the three single factor preferences to obtain the user's overall preference;

Construct an event preference module, which uses the social influence of the event organizer in the group and the popularity of the geographical location where the event is held in the group to express the preference for the event;

The user event bidirectional preference scoring module uses a ranking learning algorithm to solve the weight parameters of the user preference score and the event preference score to obtain the user event bidirectional preference score;

The final recommendation score module for the user-event pair is used to linearly weight the topic matching score and the two-way preference score to obtain the final recommendation score of the user-event pair.

Furthermore, the user preference module includes a geographic location preference module, a social relationship preference module and a time factor preference module, and the event preference module includes an event location popularity preference module and an event organizer influence preference module, wherein:

The geographic location preference module is used to express the geographic location preference score by predicting the probability of a user participating in an event held in a certain geographic location;

The social relationship preference module is used to calculate the social preference score of the target user from two aspects: the relationship between the target user and the group, and the correlation with the users in the group;

The time factor preference module is used to construct a unified vector representation of two granularities, date and hour, and calculate the similarity of the user-event pair as the time preference score of the target user;

The event location popularity preference module is used to recommend new events. The location of the event is an important selection basis for interested users, which is called the popularity of the geographical location among the user group. Considering the popularity of the geographical location of the event can more accurately calculate the attractiveness of the event to the user;

The event organizer influence preference module is used to improve the accuracy of recommendations based on the influence of the event organizer in the target user's group, and calculates the influence of the event organizer from two aspects: the influence of the event organizer on the target user and the influence of the event organizer in the group.

In the above-mentioned personalized event recommendation method and system based on the fusion of topic matching and bidirectional preference, firstly, the topic information of the event is extracted by using the document topic generation model LDA, and the user topic information is obtained according to the historical event records of the user's participation, and the topic matching degree between the user and the event is calculated as an important recommendation factor in the recommendation model. The topic factor can better represent the feature preference; secondly, for the event-based social network recommendation, from the bidirectional perspective of users and events, the preference model of users and events is constructed, and the user preference score and event preference score are obtained respectively, and the preference relationship is more completely mined from the two perspectives of users and events; finally, the user-event pair matching degree is fused with the user-event bidirectional preference linear weighted combination to obtain the final user-event pair comprehensive score, and the top K (i.e., TOP-K) user-event pairs after sorting are used as the recommendation results. This scheme has been extensively experimented on the real Meetup dataset and compared with other event recommendation algorithms, which shows that the performance of this software recommendation algorithm is better than the traditional recommendation scheme, and can well predict the personalized preferences of users, thereby achieving the purpose of personalized recommendation.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a diagram of an overall recommendation fusion framework structure of a personalized event recommendation method and system integrating topic matching and two-way preference according to an embodiment of the present invention.

FIG. 2 is a structural block diagram of a document topic generation model LDA of a personalized event recommendation method and system integrating topic matching and bidirectional preference according to an embodiment of the present invention.

DETAILED DESCRIPTION

This embodiment takes the personalized event recommendation method integrating topic matching and two-way preference as an example, and the present invention will be described in detail below in conjunction with specific embodiments and drawings.

Please refer to FIG. 1 and FIG. 2 , which illustrate a personalized event recommendation method and system integrating topic matching and two-way preference provided by an embodiment of the present invention.

Here we will explain the technical details of the personalized event recommendation system of this software that integrates topic matching and two-way preference. The main idea is that, first, the topics of new events and user historical events are calculated through the LDA topic model, the topic matching degree of user-event pairs is calculated using cosine similarity, and the user preference model and event preference model are constructed respectively. Among them, the user preference model calculates the user's comprehensive preference score from three aspects: time, geography, and social relationship. The event preference model represents the potential preference score of the event according to the geographical location popularity of the new event in the target user group and the social influence of the organizer within the group. Then, the Bayesian Personalized Ranking (BPR) algorithm is used to learn the weight parameters of the user preference score and the event preference score to obtain the user event two-way preference score. Finally, the final recommendation score of the user-event pair is obtained by linear weighted fusion with the topic matching degree, and the top-k events after sorting are recommended to the user. That is, this software matches users and event topics, calculates user preferences and event potential preferences in combination with several major contextual information, and finally integrates the topic matching degree and the user-event two-way preference to recommend events.

1. Recommendation framework integrating LDA topic matching and user event bidirectional preference

On the basis of existing work, based on the geographic location information, time information, social relations and other relevant user event context information in EBSN, an event recommendation scheme combining user-event pair topic matching and user-event pair bidirectional preference is proposed. In this scheme, the influence of topic matching, user preference and event preference on event recommendation is considered respectively, and these factors are integrated to effectively recommend events of interest to users. The overall framework of the recommendation model is shown in Figure 1, and its specific recommendation process is as follows:

1) Based on the description documents of events in EBSN, the LDA topic model is used to calculate the topics of new events and historical events of the target user. The topics of the user's historical events are used to represent the user's topics. Then, the semantic similarity between the event and user topic distribution is calculated to obtain the matching score of the user-event topic.

2) Calculate user preference scores and event preference scores. For user preference, the preference scores are calculated from three aspects: geographic location, social relationship, and time, and linearly integrated. Event preference is represented by the popularity of the geographic location of the event and the social influence of the event organizer. Linear integration is also performed to obtain the event preference score. It should be noted that when calculating the geographic location popularity and organizer influence of the event, only the group and users in the group to which the target user belongs are calculated, and all associations with other users and groups are ignored to improve recommendation performance and reduce computational complexity.

3) Through the above calculations, we get the user-event topic matching score, the user's preference score for the event, and the event's preference score for the user. We first use the Bayesian personalized ranking algorithm to learn the weights of the user preference score and the event preference score, and then fuse the user and event preference scores according to the weights to get the two-way preference score. Finally, we linearly combine the topic matching score and the two-way preference score information to get the final user-event recommendation score, and recommend the top-k events with the highest scores to the user.

2. LDA-based topic matching model

In event social networks, there is an obvious topic semantic similarity relationship between users and events. Users usually choose to participate in a certain type of event of interest, which generally has similar attributes and topics. Applying the topic of events in recommendations can better capture the preferences of users and events. We represent user topics with the topics of historical events that users have participated in, and calculate the topic distribution and word distribution of new events. We represent the topic matching degree with the topic similarity between user historical events and new events, and integrate it into the recommendation model as one of the key factors for recommendation to make event recommendations.

When two documents have the same topic and other features, it is difficult to distinguish the two objects using the TF-IDF (Term Frequency–Inverse Document Frequency) algorithm. Therefore, the Bayesian-based LDA topic model is selected to calculate the document topic distribution and word distribution. The LDA topic model is a Bayesian probability model for calculating the document topic distribution. It is used to cluster potential topics for documents and generate document topics. The core idea is that each document selects a topic with a certain probability, and selects a word from this topic with a certain probability. It is believed that the topic in any document conforms to the Dirichlet distribution, through which the relationship between texts can be discovered. LDA consists of a three-layer generative Bayesian network structure, including documents, topics, and words. Both document-topic and topic-word obey multinomial distribution. The LDA topic model generation process is shown in Figure 2.

Given a set of documents , Figure 2 and Respectively represent documents The prior Dirichlet distribution of topic distribution and word distribution, are the hyperparameters of the topic prior distribution and word prior distribution given by experience, k is the number of topics in the document set specified in advance, _and Nm represents the document The total number of words in the document set. For each word in, LDA uses prior knowledge Determine the topic distribution of documents , and then from the topic distribution Extract a topic z from the Determine the word distribution of the current topic , and then from the word distribution corresponding to the topic z Extract a word from Repeat the above process N _m times to generate the document In this process, the Gibbs sampling method can be used to solve the document The topic distribution.

According to LDA, the topic similarity between users and events is calculated. The text content is first converted into semantic features, and the topic distribution is calculated for each event using the LDA topic model. Event content is mainly composed of titles and description documents, as well as information such as time and venue. Event topics can be extracted through event content. Relatively speaking, users can choose to set interest tags to express their preferences. However, many users do not set interest tags or self-introductions. User content lacks document information and faces the problem of extremely sparse data. At this time, there are no available features to represent user topics, so it is more accurate to choose the topic of historical events in which the user participated to express the user topic, and avoid the problems of data sparsity and label blanks. Stop words and punctuation marks are removed from all event content, and the document content after removing noise interference words is regarded as the set D of all documents, which is input into the LDA topic model to generate documents according to the generation process described above. Joint distribution of topics and words , as shown in formula (1). Then the Gibbs sampling method is used to estimate the two unknown parameters in the model, namely the event topic distribution and keyword distribution .

(1)

After obtaining the topic distribution and word distribution of the event document through the LDA process, the JS divergence (JensenShannon divergence) method is then used to calculate the similarity between events based on the topic distribution of the events. JS divergence is a variant based on KL divergence (Kullback-Leibler divergence). It is symmetric and solves the asymmetric problem of KL divergence. It can better measure the similarity of two probability distributions. According to formula (1), the topic distribution of all events has been generated , given an event and They have topic distribution , first calculate the JS divergence between the two through the JS divergence method , as shown in formula (2).

(2)

in, , Represents KL divergence, which is used to describe two probability distributions and The difference between them is calculated as shown in formula (3).

(3)

Combining equation (2) and equation (3), we can get the event and The topic similarity is , as shown in formula (4).

(4)

Thematic similarity of events The value of lies in [0,1]. The closer the value is to 1, the higher the event similarity is. As mentioned above, the topic similarity between the new event and the user's historical events is used as the topic similarity between the user and the event. Users often participate in multiple events and have multiple topic similarities with the new event. Represents the number of historical events of the target user, taking the average value of all similarities of the target user As the topic matching score between the user and the new event, as shown in formula (5).

(5)

Algorithm 1 describes the process of calculating the topic matching degree of user-event pairs through the LDA topic model, where The distribution of words representing the topic, represents document topic distribution, Dir () represents Dirichlet distribution, Mult () represents multinomial distribution, and Poiss () represents Poisson distribution.

Algorithm 1 gives the process of solving the user-event topic matching score using the LDA topic model and the JS divergence algorithm. First, all event descriptions are organized into a document set and stop words are removed as the input of the LDA model, and the topic distribution of each event is obtained respectively (lines 2 to 11); then the topic distribution similarity between the target user's historical events and new events is calculated according to the JS divergence algorithm (lines 12 to 14); finally, the similarity of all historical events of the target user is averaged to obtain the topic matching score between the user and the new event (lines 15 to 16).

3. User-based preference model

For user preferences, feature learning is generally performed from the user's relevant context information, and the learned feature information is expressed as user preferences. The following constructs the user's single factor preferences from three aspects: geographical factors, social relationships, and time factors, and then weights and fuses the three single factor preferences to obtain the user's overall preference.

3.1 Geographic location preference

The location preference model calculates the probability that the target user will participate in an event held at that location. It uses the KDE (Kernel Density Estimation) method to model the two-dimensional location distribution of the events that the user participates in, and uses the normalized event participation probability to represent the user's preference for the location. The longitude and latitude coordinates of the event location are represented by ( Lx, Ly ), and the set of locations where the user has historically participated in events is represented by L(u). The KDE function for user u is As shown in formula (6).

(6)

Among them, l _i =( Lx _i ,Ly _i ) ^T represents the two-dimensional vector of the longitude and latitude coordinates of the event location, m _l ( u,l _i ) represents the frequency of user u participating in activities held at geographic location l _i , σ represents the size of the neighborhood window (bandwidth), N represents the number of location samples, K ( • ) represents the Gaussian kernel function, and its definition is shown in formula (7).

(7)

Combining equations (6) and (7), we can define the probability that user u will participate in an event held at location l , as shown in equation (8).

(8)

Normalize the probability to get the user's preference score for geographic location , as shown in formula (9).

(9)

3.2 Social relationship preference

In the user social relationship network, users generally join at least one or more interest groups online and can choose to participate in events published by different groups. In these group relationships, users usually choose to participate in the preferred group that they are most interested in. Members in the same group generally have the same interests. Therefore, the user's social relationship preferences can be considered through the user's online group relationships, which mainly include two types of interactive relationships.

1) The correlation between users and groups. That is, the interactive relationship between users and all the groups they belong to, and between users and events created in the groups. Let G ( u ) represent the set of groups to which the events that user u participates belong. Then the correlation between users and groups is It can be expressed as shown in formula (10).

(10)

Among them, m _p ( u,g ) represents the set of events and activities that user u has participated in in the user group.

2) Intra-group user relevance. Intra-group user relevance is defined by the similarity of friends in the target user’s group. The similarity between the target user and the users in the group is calculated. , as shown in formula (11).

(11)

Among them, sim ( u _i ,u _j ) represents the similarity between user u _i and user u _j in the same group, as shown in formula (12).

(12)

Finally, s ( u,g ) is normalized to , as shown in formula (13).

(13)

Combining these two interactive relationships, users belonging to the same or similar groups tend to participate in events created by these groups. Combining the correlation between users and groups and the correlation between users within groups, we can derive the social preference score of user u regarding online group g . , as shown in formula (14).

(14)

in, As a weight control parameter, in social relationship networks, it is generally believed that the preference association between the target user and the group is as important as the association between users in the group. The value of is set to 0.5.

3.3 Time Preference

The time factor of the event is another important preference factor that needs to be considered when calculating user preferences. Different users have different preferences when choosing to participate in events. Some users may prefer to participate in events in the evening, while others may prefer to participate in events in the morning, or they may prefer different time points on weekdays or weekends. In reality, time is cyclical, mainly based on 7 days a week and 24 hours a day. For users to choose to participate in events on a certain day of the week and a certain hour of the day, two different levels of user time preferences will be formed. We represent the user's time preference by combining the user's choices at two levels of granularity.

If a user chooses a certain time period on a certain day of the week to participate in an event, this may indicate an implicit time preference of the user, and the user may choose to participate in the event again in the same time period next time. In order to uniformly and intuitively represent this implicit preference, we represent the new event e that the user can choose to participate in as a 7*24-dimensional event time vector When a new event occurs in a specific time period of a week, the vector component value of that time period is set to 1, otherwise it is set to 0. Therefore, in the time preference model, users can be represented as user time vectors based on the historical event records that the users participated in. , as shown in formula (15).

(15)

Among them, Eu represents the set of historical events that the target user has participated in, and then calculates the cosine similarity between the _user time vector and the new event time vector , as shown in formula (16).

(16)

For new events ,user The similarity can be obtained according to formula (16): , normalize the similarity to get the user's time preference score for the event , as shown in formula (17).

(17)

3.4 User Fusion Preference Scoring

Based on the previous modeling of the user's single-factor preference model from three aspects, the user's preference scores for geographic location, social relationships, and time are calculated respectively. For geographic location, the geographic location preference score is expressed by predicting the probability of the user participating in events held at that location; for social relationships, the target user's social preference score is calculated from two aspects: the relationship between the target user and the group, and the correlation with the users in the group; for time preference, it is represented by constructing a unified vector with two granularities of date and hour, and based on this, the similarity of the user-event pair is calculated as the target user's time preference score. These three single-factor preferences are combined to form a user preference perception model, and the three single-factor preferences are linearly combined to obtain the overall preference score of user u for event e. , as shown in formula (18).

(18)

in, The distribution represents the user's preference scores on the three factors of geographic location, social relationship, and time. Algorithm 2 describes the calculation process of user preference scores.

Algorithm 2 gives the process of solving the user's comprehensive preference score by combining the user's preferences in geographic location, social relationship, and time. The kernel density estimation algorithm is used to predict the probability that the user may attend an event held in a specific location, and the probability is normalized to represent the user's geographic preference (line 3); the social association between the user and the online group and the members in the group is calculated according to equations (10) and (13) to represent the social preference (lines 5 to 11); new events and user historical events are represented as time vectors, and the cosine similarity between the two is calculated to represent the user's time preference (line 4); finally, the three preference values are linearly combined to obtain the user's total preference score (lines 13 to 14).

4. Event-based Preference Model

For event preferences, we consider learning from event organizers and event ontology information. Since events lack active personalized context information compared to users, new events do not have historical records, personalized tags, and other information. Therefore, we use the social influence of the event organizer in the group and the popularity of the geographical location where the event is held in the group to represent event preferences.

4.1 Event Location Popularity

The geographical location of an event is a factor that users consider when choosing whether to attend an event. For an online group that a user joins, it is generally a group of users with the same interests. There may be multiple users who choose to attend the same event. Therefore, for new event recommendations, the location of the event can be an important basis for interested users to choose. This relationship is called the popularity of the geographical location in the user group. Considering the popularity of the geographical location of an event in the model for calculating event preferences can more accurately calculate the attractiveness of the event to users. The popularity of the geographical location is calculated based on the frequency of visits to the location by user u and the users in the online group g that he joined.

First, define the popularity of the event location l _e with respect to user u , as shown in formula (19).

(19)

Among them, the molecule is the frequency of user u participating in activities held at location l _e , and the denominator is the maximum frequency of locations that user u has visited in the past. Similarly, the popularity of location l _e with respect to group g to which user u belongs can be defined as , as shown in formula (20).

(20)

The numerator represents the frequency of each user in group g participating in practical activities at location l , and the denominator is the maximum frequency of locations that group members have visited historically. This allows us to calculate the popularity of geographic location l _e with respect to users in group g . and The total popularity of the location of the recommended event to the target user u can be defined as , as shown in formula (21).

(twenty one)

4.2 Influence of event organizers

In event social networks, the initiator of each event is also an ordinary user on the network. Generally, if the organizer initiates an event and gets a good response, then the next time the organizer initiates other new events, the users who participated before are likely to choose to participate in the event again. Although the event to be recommended is a brand new event that has not yet occurred for each user, the organizer of the event may be an active organizer of this type of event and may have hosted many events in the past. This provides more auxiliary recommendation information for solving the cold start problem in event recommendation. It can be seen that the influence of the event organizer in the user group within the group is an important feature of event preference. This software improves the accuracy of recommendation based on the influence of the event organizer in the group where the target user is located. Its influence can be considered from the following two aspects.

1) The influence of the event organizer on the target user. In the event social network, there is no user rating information on the event, so it is impossible to intuitively represent the influence of the organizer and the event. Moreover, it is meaningless to rate the event at the end of its life cycle because it will not affect new events held later. Therefore, the organizer's reputation or influence is used to represent the implicit preference for the event. First, define the influence of the event on user u , as shown in formula (22).

(twenty two)

in, represents the set of events organized by organizer u _h that user u has participated in, and E _h is the set of all events organized by organizer u _h .

2) The influence of the event organizer in the group. For the online group where the target user is located, the influence of the event in the group can be similarly expressed by the frequency ratio of users participating. The influence of the user in the group is expressed as It is represented as shown in formula (23).

(twenty three)

Among them, U _g represents the group The collection of users in Indicates user Participants are organized by A collection of events held, express In the group The event collection held in the group. The comprehensive influence score of the event organizer can be obtained by combining the influence of the event organizer on the target users and the users in the group. , as shown in formula (24).

(twenty four)

4.3 Event Potential Preference Scoring

For new events that have not yet occurred, this software sets the two key factors to attract users to participate as geographic location and the influence of the organizer. The preference for the event is expressed by calculating the geographic location popularity of the new event and the social influence of its organizer. In order to reduce the complexity of calculation and avoid the interference and influence of weakly correlated data, the geographic location popularity of the event and the social influence of the organizer are limited to the group where the target user belongs. It is assumed here that the correlation of the remaining users or groups is zero, which has no effect on the event preference. For the above constructed event geographic location popularity and the influence of the organizers Linear combination to find the preference score of event e for user u , as shown in formula (25).

(25)

Algorithm 3 describes in detail the process of calculating the event potential preference score based on the event location popularity and organizer influence.

Algorithm 3 gives the process of solving the potential preference score of an event based on the popularity of the event's geographic location and the influence of the organizer. For the group to which the target user belongs, the popularity of the event's geographic location to the user and the group is calculated according to equations (19) and (20), and the combination of the two represents the total popularity of the event's geographic location (rows 3 to 8); similarly, the influence of the event organizer on the user and the group is obtained by equations (22) and (23) (rows 9 to 13), and the combination of the two represents the influence of the event organizer; finally, the location popularity and the organizer's influence are linearly combined to obtain the potential preference score of the event (row 17).

5. Recommendation algorithm integrating topic matching and user event bidirectional preference

Previously, we used the LDA topic model to solve the topic distribution of users and events respectively, and calculated the topic matching of user-event pairs based on the topic distribution; then we built a feature preference scoring model for users and events to obtain the user preference score and event preference score respectively. Now we will combine the topic matching and user event preference to solve the final recommendation score. In the first step, we use the ranking learning algorithm to solve the weight parameters of the user preference score and event preference score to obtain the user event bidirectional preference score; in the second step, we linearly weight the topic matching score and the bidirectional preference score to obtain the final recommendation score of the user-event pair. The following is a detailed introduction.

1) Calculate the bidirectional preference score for the user-event pair. Assume that the preference score weights of the user and event are and , weighted fusion of the two to obtain the user event bidirectional preference score Therefore, the key problem of two-way preference scoring is to find the weight vector of the two preference scores. We choose to use implicit feedback as training data to learn the weight vector. Different from the explicit feedback of users rating items, implicit feedback in event social networks can only be represented by the interaction information between users and events, that is, if the user participates in the event, the feedback is 1, otherwise the feedback is 0. Obviously, for all new events, the user's feedback is 0.

Here, we choose the learning algorithm BPR based on Bayesian maximum likelihood estimation to rank the weights and learn the correct ranking order of user-event pairs based on the implicit feedback data of users on events, so that the events in which users participate are ranked before new events or other events. First, we define the maximum posterior probability , as shown in formula (26).

(26)

Among them, θ represents the weight vector, R represents the set of all user-event pairs, The definition is shown in formula (27).

(27)

in, Indicates user of user-event pairs, and For users event Ranked The previous probability is shown in formula (28).

(28)

in, Bidirectional preference score , To facilitate optimization, assume Obeying the normal distribution with a mean of 0, the final optimization objective function is derived , as shown in formula (29).

(29)

in, Represents the regularization term coefficient. By maximizing the objective function through the implicit interactive feedback data of user events, the optimal weight parameter vector can be obtained. The Stochastic Gradient Descent (SGD) algorithm is used to solve the optimization problem. In the iterative process, the user-event pairs of the target user are randomly extracted from the training set to update the weight vector , the updating process is shown in formula (30).

(30)

in, is the learning rate, Through the above learning process, the training set and hyperparameters can be automatically scored according to user event preferences. and Find the weight vector , thus obtaining a two-way preference score .

2) Combine topic matching and bidirectional preference to obtain the final recommendation score of the user-event pair. Based on the above discussion on topic matching and preference calculation of users and events, first, the event topic is extracted through the LDA topic model and the topic matching score of users and events is obtained; secondly, the preference models of users and events are constructed respectively according to the user event context information in EBSN, and the bidirectional preference score of users and events is obtained through the BPR learning algorithm; finally, the topic matching score is calculated. Bidirectional preference scoring with user events Linear weighted summation is used to obtain the final user-event recommendation score , as shown in formula (31).

(31)

in, is a weight parameter, which is usually set manually based on experience. The optimal setting will be determined through experiments. Algorithm 4 describes the process of integrating topic matching and bidirectional preference to solve the final recommendation score of user-event pairs.

Algorithm 4 gives the final fusion process of topic matching score and user event bidirectional preference score. First, the training set generated by the user preference score set and the event preference score set is sorted and learned by the Bayesian personalized sorting algorithm to obtain the optimal weight vector , and according to Calculate the target user's user-event pair bidirectional preference score (rows 2 to 10); secondly, linearly combine the user-event pair's topic matching score and bidirectional preference score to obtain the final recommendation score (rows 11 to 13), and then recommend TOP-K events to the user according to the final recommendation score ranking.

So far, we have combined topic matching and user-event bidirectional preferences to propose a personalized event recommendation solution, and introduced its specific content in detail in the above section.

And, a system for implementing personalized event recommendation integrating topic matching and two-way preference, which is used to implement the personalized event recommendation method integrating topic matching and two-way preference as described in any one of the above items, and the implementation system includes:

The document topic generation module is used to extract the topics of the user's historical events and new events, and calculate the topic distribution and word distribution of the events. The topic similarity between the user's historical events and new events is used to represent the topic matching degree, which is integrated into the recommendation model as one of the key factors for recommendation to perform event recommendation.

Build a user preference module to construct the user's single factor preference from three aspects: geographic location, social relationship, and time factor, and weightedly integrate the three single factor preferences to obtain the user's overall preference;

Construct an event preference module, which uses the social influence of the event organizer in the group and the popularity of the geographical location where the event is held in the group to express the preference for the event;

The user event bidirectional preference scoring module uses a ranking learning algorithm to solve the weight parameters of the user preference score and the event preference score to obtain the user event bidirectional preference score;

The final recommendation score module for the user-event pair is used to linearly weight the topic matching score and the two-way preference score to obtain the final recommendation score of the user-event pair.

Furthermore, the user preference module includes a geographic location preference module, a social relationship preference module and a time factor preference module, and the event preference module includes an event location popularity preference module and an event organizer influence preference module, wherein:

The geographic location preference module is used to express the geographic location preference score by predicting the probability of a user participating in an event held in a certain geographic location;

The social relationship preference module is used to calculate the social preference score of the target user from two aspects: the relationship between the target user and the group, and the correlation with the users in the group;

The time factor preference module is used to construct a unified vector representation of two granularities, date and hour, and calculate the similarity of the user-event pair as the time preference score of the target user;

The event location popularity preference module is used to recommend new events. The location of the event is an important selection basis for interested users, which is called the popularity of the geographical location among the user group. Considering the popularity of the geographical location of the event can more accurately calculate the attractiveness of the event to the user;

The event organizer influence preference module is used to improve the accuracy of recommendations based on the influence of the event organizer in the target user's group, and calculates the influence of the event organizer from two aspects: the influence of the event organizer on the target user and the influence of the event organizer in the group.

In the above-mentioned personalized event recommendation method and system based on the fusion of topic matching and bidirectional preference, firstly, the topic information of the event is extracted by using the document topic generation model LDA, and the user topic information is obtained according to the historical event records of the user's participation, and the topic matching degree between the user and the event is calculated as an important recommendation factor in the recommendation model. The topic factor can better represent the feature preference; secondly, for the event-based social network recommendation, from the bidirectional perspective of users and events, the preference model of users and events is constructed, and the user preference score and event preference score are obtained respectively, and the preference relationship is more completely mined from the two perspectives of users and events; finally, the user-event pair matching degree is fused with the user-event bidirectional preference linear weighted combination to obtain the final user-event pair comprehensive score, and the sorted TOP-K user-event pairs are used as the recommendation results. This scheme has been extensively experimented on the real Meetup dataset and compared with other event recommendation algorithms, which shows that the performance of this software recommendation algorithm is better than the traditional recommendation scheme, and can well predict the user's personalized preferences, thereby achieving the purpose of personalized recommendation.

It should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims (9)

1. A personalized event recommendation method integrating topic matching and bidirectional preference, characterized by comprising the following steps: Step 1: Use the document topic generation model LDA to extract the topic information of the event, and obtain the user topic information based on the historical event records in which the user participated, calculate the topics of the new event and the user's historical events, and use the JS divergence algorithm to calculate the topic matching score of the user-event pair; Step 2: construct a user preference model and an event preference model respectively, and calculate the user preference score and the event preference score respectively; Step 3: Use the Bayesian personalized ranking algorithm BPR to learn the weight parameters of user preference scores and event preference scores to obtain the user-event bidirectional preference score, perform a linear weighted combination of the topic matching score and the bidirectional preference score to obtain the final recommendation score of the user-event pair, and recommend the top K events after ranking to the user; The calculation of the topics of the new event and the user's historical events in step 1 uses the JS divergence algorithm to calculate the topic matching score of the user-event pair, and the specific steps include: Step 1-1, all event description contents are grouped into a document set D and stop words are removed, the document set D is input into a document topic generation model LDA, and the topic distribution of each event is obtained respectively; Remove stop words and punctuation marks from all event contents, and regard the document contents after removing noise interference words as the set D of all documents, which is input into the LDA topic model to generate document Joint distribution of topics and words , as shown in formula (1): (1) Given a set of documents , and Respectively represent documents The prior Dirichlet distribution of topic distribution and word distribution, are the hyperparameters of the topic prior distribution and word prior distribution given by experience, k is the number of topics in the document set specified in advance, _and Nm represents the document The total number of words, M is the number of documents in the document set; for the document For each word in, LDA uses prior knowledge Determine the topic distribution of documents , and then from the topic distribution Extract a topic z from the Determine the word distribution of the current topic , and then from the word distribution corresponding to the topic z Extract a word from Repeat the above process N _m times to generate the document ; Then the Gibbs sampling method is used to estimate the two unknown parameters in the model: event topic distribution and keyword distribution ; Step 1-2, calculate the topic distribution similarity between the target user's historical events and new events according to the JS divergence algorithm; According to formula (1), the topic distribution of all events has been generated , given an event and They have topic distribution , first calculate the JS divergence between the two through the JS divergence method , as shown in formula (2): (2) in, , Represents KL divergence, which is used to describe two probability distributions and The difference between them is calculated as shown in formula (3): (3) Combining equation (2) and equation (3), we can get the event and The topic similarity is , as shown in formula (4): (4) Among them, the topic similarity of the event The value of lies in [0,1], and the closer the value is to 1, the higher the event similarity; Steps 1-3: average the similarities of all historical events of the target user to obtain the topic matching score between the user and the new event; by Indicates the number of historical events of the target user, taking the average value of all similarities of the target user As the topic matching score between the user and the new event, as shown in formula (5): (5) According to the constructed topic matching model, the final To measure the topic matching relationship between target users and new events. 2. The personalized event recommendation method integrating topic matching and two-way preference as described in claim 1 is characterized in that the document topic generation model LDA in step one has a three-layer generative Bayesian network structure, including documents, topics and words, wherein both document-topic and topic-word obey multinomial distribution; each document selects a topic with a certain probability, and selects a word from this topic with a certain probability, and the topic in any document conforms to the Dirichlet distribution, through which the relationship between texts is discovered. 3. The personalized event recommendation method integrating topic matching and two-way preference according to claim 2 is characterized in that the user preference model constructed in step 2 constructs the user's single factor preference from three aspects: geographical location, social relationship, and time factor, and specifically includes: Step 2-1-1, build a geographic location preference model: The location preference model calculates the probability that the target user will participate in an event held at that location. The kernel density estimation (KDE) method is used to model the two-dimensional location distribution of the events that the user participates in. The normalized event participation probability is used to represent the user's preference for the location. The longitude and latitude coordinates of the event location are represented by ( Lx, Ly ), and the set of locations where the user has historically participated in events is represented by L ( u ). The KDE function for user u is As shown in formula (6): (6) Where, l _i =( Lxi _, Ly _i ) ^T represents the two-dimensional vector of the longitude and latitude coordinates of the event location, m _l ( u, l _i ) represents the frequency of user u participating in activities held at geographic location l _i , σ represents the size of the neighborhood window (bandwidth), N represents the number of location samples, and K ( • ) represents the Gaussian kernel function, which is defined as shown in formula (7): (7) Combining equations (6) and (7), we can define the probability that user u will participate in an event held at location l , as shown in equation (8): (8) Normalize the probability to get the user's preference score for geographic location , as shown in formula (9): (9) The denominator represents the target user’s maximum event participation probability; Step 2-1-2, build a social relationship preference model: In the user social relationship network, users will join at least one or more interest groups online and choose to participate in events and activities released by different groups. The user's social relationship preferences are judged by the user's online group relationships, which mainly include two types of interactive relationships; The first type, user-group relevance, is defined as the interaction relationship between users and all the groups they belong to and between users and events created within the groups. G ( u ) represents the set of groups to which events user u participates. Then the user-group relevance is It can be expressed as shown in formula (10): (10) Where, m _p ( u,g ) represents the set of events that user u has participated in in the group to which he belongs; The second type is the intra-group user relevance. The intra-group user relevance is defined by the similarity of the friends in the target user’s group. The similarity between the target user and the users in the group is calculated. , as shown in formula (11): (11) Among them, sim ( u _i ,u _j ) represents the similarity between user u _i and user u _j in the same group, as shown in formula (12); (12) Normalize s ( u,g ) to , as shown in formula (13): (13) Combining the above two interactive relationships, users belonging to the same group tend to participate in events created by other users in these groups. Combining the correlation between users and groups and the correlation between users in groups, the social preference score of user u about online group g is obtained. , as shown in formula (14): (14) in, As a weight control parameter, in social relationship networks, setting the preference association between the target user and the group is as important as the association between users in the group. The value of is set to 0.5; Step 2-1-3, construct a time factor preference model: The time factor of the event is an important preference factor that needs to be considered when calculating user preferences; the new event e that the user can choose to participate in is represented as a 7*24-dimensional event time vector , when a new event occurs in a specific time period of a week, the vector component value of the time period is set to 1, otherwise it is 0; in the time preference model, users are represented as user time vectors according to the historical event records of the users. , as shown in formula (15): (15) Among them, Eu represents the set of historical events that the target user has participated in, and then calculates the cosine similarity between the _user time vector and the new event time vector , as shown in formula (16): (16) For new events ,user The similarity can be obtained according to formula (16): , normalize the similarity to get the user's time preference score for the event , as shown in formula (17): (17). 4. The personalized event recommendation method integrating topic matching and bidirectional preference according to claim 3, wherein the step of calculating the user preference score in step 2 specifically includes: For the geographic location preference model, the geographic location preference score is expressed by predicting the probability of the user participating in events held at the location; for the social relationship preference model, the social preference score of the target user is calculated from two aspects: the relationship between the target user and the group and the correlation with the users in the group; for the time factor preference model, a unified vector representation of the two granularities of date and hour is constructed, and the similarity of the user-event pair is calculated based on this as the time preference score of the target user; these three single-factor preferences are combined to form a user preference perception model, and the three single-factor preferences are linearly combined to obtain the overall preference score of user u for event e , as shown in formula (18): (18) in, They represent the user's preference scores on three single factors: geographic location, social relationship, and time. 5. The personalized event recommendation method integrating topic matching and two-way preference according to claim 4 is characterized in that the event preference model constructed in step 2 constructs single-factor preferences of events from two aspects: event location popularity and event organizer influence, specifically including: Step 2-2-1, construct event location popularity preference model: The popularity of a geographic location is calculated based on the frequency of visits to the location by user u and the users in the online group g to which he joins; First, define the popularity of the event location l _e with respect to user u , as shown in formula (19): (19) Among them, the molecule is the frequency of user u participating in activities held at geographic location l _e , and the denominator is the maximum frequency of locations that user u has visited historically; similarly, define the popularity of geographic location l _e with respect to group g to which user u belongs , as shown in formula (20): (20) The numerator represents the frequency of each user in group g participating in practical activities at location l , and the denominator is the maximum frequency of locations that group members have visited historically. This allows us to calculate the popularity of geographic location l _e with respect to users in group g . and Define the total popularity of the location of the recommended event to the target user u as , as shown in formula (21): (twenty one) Step 2-2-2, build the influence preference model of event organizers: First, the influence of the event organizer on the target user. The organizer's reputation or influence is used to express the implicit preference of the event. Define the influence of the event on user u , as shown in formula (22): (twenty two) in, represents the set of events held by organizer u _h that user u has participated in, E _h is the set of all events held by organizer u _h ; Second, the influence of the event organizer in the group. For the online group where the target user is located, the influence of the event in the group is expressed by the frequency ratio of users participating in the event. The influence of the user in the group is expressed by It is expressed as shown in formula (23): (twenty three) Among them, U _g represents the group The collection of users in Indicates user Participants are organized by A collection of events held, express In the group The event collection held in the group; the comprehensive influence score of the event organizer is obtained by combining the influence of the event organizer on the target users and the users in the group. , as shown in formula (24): (twenty four). 6. The personalized event recommendation method integrating topic matching and two-way preference according to claim 5, characterized in that the calculation of event preference score in step 2 specifically includes: For new events that have not occurred, the preference for the event is expressed by calculating the event location popularity and event organizer influence of the new event; for the constructed event location popularity and event organizer influence Linear combination, calculate the preference score of event e for user u , as shown in formula (25): (25). 7. The personalized event recommendation method integrating topic matching and two-way preference according to claim 6 is characterized in that the step of obtaining the two-way preference score of the user event in step 3, linearly weighting the topic matching score and the two-way preference score to obtain the final recommendation score of the user-event pair, specifically comprises the following steps: Step 3-1, find the bidirectional preference for the user-event pair: Assume that the preference score weights of users and events are and , weighted fusion of the two to obtain the user event bidirectional preference score ; Convert the problem of bidirectional preference scoring into finding the weight vector of two preference scores, and choose to use implicit feedback as training data to learn the weight vector; We select the learning algorithm BPR based on Bayesian maximum likelihood estimation to rank the weights and learn the correct ranking order of user-event pairs based on the implicit feedback data of users on events, so that the events in which users participate are ranked before new events or other events. First, we define the maximum posterior probability , as shown in formula (26): (26) Among them, θ represents the weight vector, R represents the set of all user-event pairs, The definition is as shown in formula (27); (27) Among them, Indicates user of user-event pairs, and For users event Ranked The previous probability is shown in formula (28): (28) in, Bidirectional preference score , ; In order to facilitate optimization, assume Obeying the normal distribution with a mean of 0, the final optimization objective function is derived , as shown in formula (29): (29) in, represents the regularization term coefficient. The objective function is optimized by maximizing the implicit interactive feedback data of user events to obtain the optimal weight parameter vector. The stochastic gradient descent algorithm SGD is used to solve the optimization problem. In the iterative process, the user-event pairs of the target user are randomly extracted from the training set to update the weight vector. , the update process is shown in formula (30): (30) in, is the learning rate, ; Through the above learning process, the training set and hyperparameters can be automatically scored according to user event preferences and Find the weight vector , thus obtaining a two-way preference score ; Step 3-2, combine topic matching and bidirectional preference to obtain the final recommendation score for the user-event pair: Firstly, the LDA topic model is used to extract event topics and obtain the topic matching scores of users and events. Secondly, the preference models of users and events are constructed according to the user event context information in EBSN, and the bidirectional preference scores of users and events are obtained through the BPR learning algorithm. Finally, the topic matching scores are calculated. Bidirectional preference scoring with user events Linear weighted summation is used to obtain the final user-event recommendation score , as shown in formula (31): (31) in, is a weight parameter, which is usually set manually based on experience, and the optimal setting will be determined through experiments. 8. A system for implementing personalized event recommendation integrating topic matching and two-way preference, which is used to implement the personalized event recommendation method integrating topic matching and two-way preference as claimed in any one of claims 1 to 7, characterized in that the implementation system comprises: The document topic generation module is used to extract the topics of the user's historical events and new events, and calculate the topic distribution and word distribution of the events. The topic similarity between the user's historical events and new events is used to represent the topic matching degree, which is integrated into the recommendation model as one of the key factors for recommendation to perform event recommendation. Build a user preference module to construct the user's single factor preference from three aspects: geographic location, social relationship, and time factor, and weightedly integrate the three single factor preferences to obtain the user's overall preference; Construct an event preference module, which uses the social influence of the event organizer in the group and the popularity of the geographical location where the event is held in the group to express the preference for the event; The user event bidirectional preference scoring module uses a ranking learning algorithm to solve the weight parameters of the user preference score and the event preference score to obtain the user event bidirectional preference score; The final recommendation score module for the user-event pair is used to linearly weight the topic matching score and the two-way preference score to obtain the final recommendation score of the user-event pair. 9. The system for implementing personalized event recommendation integrating theme matching and two-way preference according to claim 8, characterized in that the user preference module includes a geographic location preference module, a social relationship preference module and a time factor preference module, and the event preference module includes an event location popularity preference module and an event organizer influence preference module, wherein: The geographic location preference module is used to express the geographic location preference score by predicting the probability of a user participating in an event held in a certain geographic location; The social relationship preference module is used to calculate the social preference score of the target user from two aspects: the relationship between the target user and the group, and the correlation with the users in the group; The time factor preference module is used to construct a unified vector representation of two granularities, date and hour, and calculate the similarity of the user-event pair as the time preference score of the target user; The event location popularity preference module is used to recommend new events. The location of the event is an important selection basis for interested users, which is called the popularity of the geographical location among the user group. Considering the popularity of the geographical location of the event can more accurately calculate the attractiveness of the event to the user; The event organizer influence preference module is used to improve the accuracy of recommendations based on the influence of the event organizer in the target user's group, and calculates the influence of the event organizer from two aspects: the influence of the event organizer on the target user and the influence of the event organizer in the group.