CN116303983A

CN116303983A - Keyword recommendation method and device and electronic equipment

Info

Publication number: CN116303983A
Application number: CN202111569802.0A
Authority: CN
Inventors: 武文雅
Original assignee: China Mobile Communications Group Co Ltd; China Mobile Communications Ltd Research Institute
Current assignee: China Mobile Communications Group Co Ltd; China Mobile Communications Ltd Research Institute
Priority date: 2021-12-21
Filing date: 2021-12-21
Publication date: 2023-06-23

Abstract

The application discloses a keyword recommendation method, a keyword recommendation device and electronic equipment, relates to the technical field of Internet, and aims to solve the problem that recommendation accuracy is poor because the conventional keyword recommendation technology cannot be well matched with real demands of users. The method comprises the following steps: acquiring an input search word and acquiring time information of the input search word; determining a target time category to which the time information belongs; determining a target user search record set corresponding to the target time category from a pre-established user search log database, wherein a plurality of user search record sets respectively corresponding to different time categories are stored in the user search log database; determining a first set of candidate words associated with the search term from the set of target user search records; and determining a recommended word set based on the first candidate word set and recommending. According to the method and the device for recommending the information, the information which is more in line with the expectations of the user can be recommended to the user according to the search records corresponding to the search time of the user, and the recommendation accuracy is improved.

Description

Keyword recommendation method and device and electronic equipment

Technical Field

The application relates to the technical field of internet, in particular to a keyword recommendation method and device and electronic equipment.

Background

Keyword recommendation is a technique often used in search engines, also known as query recommendation. The keyword recommendation technology is used in the query process, so that a user can be quickly and accurately helped to locate specific information to be queried, the user query experience is improved, and the search time is saved.

In the prior keyword recommendation technology, recommended content is usually preferred according to the historical search behavior of multiple users, however, the problem that the recommendation accuracy is poor because the recommendation result can not be well matched with the real requirement of the users still exists in the mode.

Disclosure of Invention

The embodiment of the application provides a keyword recommendation method, a keyword recommendation device and electronic equipment, and aims to solve the problem that the conventional keyword recommendation technology cannot be matched with the actual requirements of users well, so that recommendation accuracy is poor.

In a first aspect, an embodiment of the present application provides a keyword recommendation method, including:

acquiring an input search word and acquiring time information of the input search word;

determining a target time category to which the time information belongs;

determining a target user search record set corresponding to the target time category from a pre-established user search log database, wherein a plurality of user search record sets respectively corresponding to different time categories are stored in the user search log database;

Determining a first set of candidate words associated with the search term from the set of target user search records;

and determining a recommended word set based on the first candidate word set and recommending.

Optionally, before determining the target user search record set corresponding to the target time category from the pre-established user search log database, the method further includes:

dividing the time into a plurality of time categories according to the date attribute and/or the time period attribute;

acquiring user search log data;

classifying each user search record in the user search log data according to the time category to which the historical search time belongs to obtain a plurality of user search record sets respectively belonging to different time categories;

and storing the plurality of user search record sets to the user search log database according to time category labels.

Optionally, the dividing the time into a plurality of time categories according to date attribute and/or time period attribute includes:

dividing time into legal holidays, other holidays and non-holidays according to date attributes, and dividing working time and non-working time of the other holidays according to time period attributes to obtain five time categories including the legal holidays, the working time of the other holidays, the non-working time of the other holidays, the working time of the non-holidays and the non-working time of the non-holidays, wherein the other holidays are the non-legal holidays.

Optionally, the acquiring the input search term includes:

acquiring a first search term which is initially input;

displaying a plurality of meanings of the first search term if the first search term is identified as ambiguous;

based on the meaning selected by the user, a search term that the user confirms the input is determined.

Optionally, after the input search term is acquired, before determining a recommended term set based on the first candidate term set and recommending, the method further includes:

determining a word vector of the search word;

determining a similar word vector set similar to the word vector of the search word from a pre-constructed word vector set to obtain a similar word set corresponding to the similar word vector set;

performing replacement processing on the search word by using the words in the similar word set to obtain a second candidate word set;

the determining and recommending the recommended word set based on the first candidate word set comprises the following steps:

and determining a recommended word set based on the first candidate word set and the second candidate word set, and recommending the recommended word set.

Optionally, the determining a recommended word set based on the first candidate word set and the second candidate word set and recommending includes:

Comparing the candidate words in the first candidate word set with the candidate words in the second candidate word set in a similar way;

determining a first recommended word set in the first candidate word set and a second recommended word set in the second candidate word set based on a similarity comparison result;

and recommending the first recommended word set and the second recommended word set.

Optionally, the determining, based on the similarity comparison result, the first recommended word set in the first candidate word set and the second recommended word set in the second candidate word set includes:

determining a first group of candidate words and a second group of candidate words in the first candidate word set as the first recommended word set, and determining a third group of candidate words in the second candidate word set as the second recommended word set;

wherein the first set of candidate words includes candidate words in the first set of candidate words that are similar to any candidate word in the second set of candidate words; the second group of candidate words comprises candidate words in the first candidate word set, which are dissimilar to any candidate word in the second candidate word set; the third group of candidate words comprises candidate words in the second candidate word set, which are dissimilar to any candidate word in the first candidate word set;

The recommending the first recommended word set and the second recommended word set includes:

and sequencing the first recommended word set and the second recommended word set according to the sequence of the first group of candidate words, the second group of candidate words and the third group of candidate words, and recommending.

In a second aspect, an embodiment of the present application further provides a keyword recommendation apparatus, including:

the first acquisition module is used for acquiring an input search word and acquiring time information of the input search word;

the first determining module is used for determining a target time category to which the time information belongs;

the second determining module is used for determining a target user searching record set corresponding to the target time category from a pre-established user searching log database, wherein a plurality of user searching record sets respectively corresponding to different time categories are stored in the user searching log database;

a third determining module configured to determine a first candidate word set associated with the search word from the target user search record set;

and the recommending module is used for determining a recommending word set based on the first candidate word set and recommending.

Optionally, the keyword recommendation device further includes:

The dividing module is used for dividing the time into a plurality of time categories according to the date attribute and/or the time period attribute;

the second acquisition module is used for acquiring user search log data;

the classification module is used for classifying each user search record in the user search log data according to the time category to which the historical search time belongs to obtain a plurality of user search record sets respectively belonging to different time categories;

and the storage module is used for storing the plurality of user search record sets to the user search log database according to the time category labels.

Optionally, the dividing module is configured to divide the time into a legal holiday, other holidays and non-holidays according to the date attribute, and divide the other holidays and the non-holidays into working hours and non-working hours according to the time period attribute, so as to obtain five time categories including the legal holiday, the working hours in other holidays, the non-working hours in other holidays, the working hours in the non-holidays and the non-working hours in the non-holidays, where the other holidays are the non-legal holidays.

Optionally, the first acquisition module includes:

the acquisition unit is used for acquiring the initially input first search word;

A display unit for displaying a plurality of meanings of the first search word in case that the first search word is recognized to have ambiguity;

and a first determining unit for determining a search term input by the user confirmation based on the meaning selected by the user.

Optionally, the keyword recommendation device further includes:

a fourth determining module, configured to determine a word vector of the search word;

a fifth determining module, configured to determine a set of similar word vectors similar to the word vector of the search word from a set of pre-constructed word vectors, to obtain a set of similar words corresponding to the set of similar word vectors;

the processing module is used for carrying out replacement processing on the search word by using the words in the similar word set to obtain a second candidate word set;

the recommendation module is used for determining a recommendation word set and recommending the recommendation word set based on the first candidate word set and the second candidate word set.

Optionally, the recommendation module includes:

a similarity comparison unit, configured to perform similarity comparison on the candidate words in the first candidate word set and the candidate words in the second candidate word set;

a second determining unit configured to determine a first recommended word set in the first candidate word set and a second recommended word set in the second candidate word set based on a similarity comparison result;

And the recommending unit is used for recommending the first recommending word set and the second recommending word set.

Optionally, the second determining unit is configured to determine that a first set of candidate words and a second set of candidate words in the first candidate word set are the first recommended word set, and determine that a third set of candidate words in the second candidate word set are the second recommended word set;

the recommending unit is used for sequencing the first recommended word set and the second recommended word set according to the sequence of the first group of candidate words, the second group of candidate words and the third group of candidate words and recommending the first recommended word set and the second recommended word set.

In a third aspect, embodiments of the present application further provide an electronic device, including: memory, a processor and a computer program stored on the memory and executable on the processor, which processor when executing the computer program implements the steps in the keyword recommendation method as described above.

In a fourth aspect, embodiments of the present application further provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps in the keyword recommendation method as described above.

In the embodiment of the application, the input search word is acquired, and the time information of the input search word is acquired; determining a target time category to which the time information belongs; determining a target user search record set corresponding to the target time category from a pre-established user search log database, wherein a plurality of user search record sets respectively corresponding to different time categories are stored in the user search log database; determining a first set of candidate words associated with the search term from the set of target user search records; and determining a recommended word set based on the first candidate word set and recommending. Therefore, the search records of the user are classified according to time by considering the search preference of the user at different times, so that the search records corresponding to the search time of the user can be recommended to the user according to the information which is more in line with the expectations of the user, and the recommendation accuracy is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments of the present application will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is one of the flowcharts of the keyword recommendation method provided in the embodiments of the present application;

FIG. 2 is a second flowchart of a keyword recommendation method according to an embodiment of the present disclosure;

FIG. 3 is a flowchart of a method for determining a first candidate word set provided by an embodiment of the present application;

FIG. 4 is a flowchart of a method for determining a second candidate word set provided by an embodiment of the present application;

FIG. 5 is a flowchart of a method for determining a final recommended vocabulary provided by an embodiment of the present application;

fig. 6 is a block diagram of a keyword recommendation apparatus provided in an embodiment of the present application;

fig. 7 is a block diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, are intended to be within the scope of the present application.

Referring to fig. 1, fig. 1 is a flowchart of a keyword recommendation method provided in an embodiment of the present application, as shown in fig. 1, including the following steps:

step 101, acquiring an input search word and acquiring time information of the input search word.

In the embodiment of the application, the recommended word set can be constructed according to the preference information of the user searching behavior in different dates and times.

The above-described acquisition-input search term may be a search keyword input by a user and representing the search intention thereof, for example, a keyword "corn", "today's weather", or "contract template on house rental" input by the user in a search engine.

The time information of inputting the search term is obtained, that is, the time information when the user inputs the search term is obtained, for example, the input timestamp of the search term may be obtained.

Optionally, the acquiring the input search term includes:

acquiring a first search term which is initially input;

In one embodiment, considering the possible problems of word ambiguity and language ambiguity in the search word, the first search word initially input by the user can be subjected to semantic recognition to identify whether the first search word has ambiguity, if so, multiple meanings exist, for example, for the search word "corn", it may refer to plant corn or a fan group name.

And under the condition that the first search word initially input by the user is recognized to have ambiguity, acquiring a plurality of meanings of the first search word, and displaying the plurality of meanings so as to enable the user to select and confirm the meaning really wanted to be searched. For example, all meanings of the first search term may be retrieved and presented in the form of options for selection by the user.

And determining that the user confirms the input search term having a definite meaning based on a meaning selected by the user from the plurality of meanings that can represent the user's actual search intention. For example, for the search term "corn" initially entered by the user, "plant corn" and "fan-population" may be provided: corn "has two meanings for the user to choose from, and when the user selects" plant corn "it can be used as a search term after confirmation. Therefore, the search intention can be more definite, and recommendation of other ambiguous related recommended words is avoided.

In this way, in this embodiment, by providing a novel search input strategy, when the content input by the user is ambiguous, a user selection link is added, all meanings covered by the input term are pushed for the user to select, the user is guided to further clearly search for intention, so that more accurate related keywords can be recommended to the user, and more accurate search results are provided.

Step 102, determining the target time category to which the time information belongs.

In this embodiment of the present application, the time may be classified, where different dates and different time periods are classified into different time categories according to a specific rule, for example, classification of holidays, weekdays, weekends, working hours, non-working hours, and the like. And the historical search records of the user can be classified and stored according to time categories so as to be matched with the historical search records of corresponding time categories according to the search time of the user when the search behavior occurs.

In this step, the time category to which the time information belongs may be determined according to the time attribute corresponding to the time information of the search term input by the user, so as to obtain the target time category. For example, the user inputs "source of noon" at 8 points 18 of 25 months of 06 months of 2020, then based on the input time information: 8 points of 25 days in 06 month in 2020 are 18 points, and 25 days in 06 month in 2020 are known as the noon festival, so that the time category to which the time belongs can be determined as the holiday.

And 103, determining a target user search record set corresponding to the target time category from a pre-established user search log database, wherein a plurality of user search record sets respectively corresponding to different time categories are stored in the user search log database.

The user search log database may be a database for storing user search log data, where the user search log data may refer to log data generated during any user search, for example, a user search log database of a certain search website may include search log data generated when various users perform a search action on the search website.

In this embodiment of the present application, the user search log database may be established, by collecting each user search log data (including a search timestamp and a search record), analyzing a search time of each user search log data, and classifying each user search log data according to a search time class, to obtain user search record sets respectively corresponding to different time classes, where the classified user search record sets are stored in the user search log database. Thus, each set of user search records in the user search log database has a respective temporal category.

In this step, a target user search record set corresponding to the target time category may be determined from the user search log database, for example, a user search record set whose time category is the target time category may be searched from the user search log database, and the searched user search record set of the target time category is the target user search record set.

Optionally, before the step 103, the method further includes:

acquiring user search log data;

That is, in one embodiment, time may be classified according to date attribute, time zone attribute, or the like, for example, holidays, anniversaries, non-holidays, or the like may be classified according to date attribute, and time zone attribute may be classified into working hours, non-working hours, or daytime, nighttime, or the like. Thus, by time division, a plurality of different time categories can be obtained.

The user search log data may refer to search log data generated when a user performs a search action, that is, a history search record, and the obtaining user search log data may be obtaining user search log data in a period of time, for example, obtaining user search log data of a year, a half year, a three month, a month or a week, and a specific obtaining period may be flexibly set according to actual needs.

In this embodiment, in order to perform targeted recommendation according to the search preference of the user at different times, the user search records in the user search log data may be classified according to the historical search times, specifically, based on the classified time categories, it may be determined which time category the historical search time of each user search record in the user search log data belongs to, and the user search records of the historical search time belonging to the same time category are classified and integrated into a user search record set. In this way, the user search records in the user search log data can be divided into a plurality of user search record sets according to the historical search time categories, and each user search record set corresponds to one time category.

Finally, the plurality of user search record sets may be stored to the user search log database, and a respective time category label may be added to each user search record set for tag discrimination.

In this way, in this embodiment, by classifying the time according to the date attribute and/or the time period attribute, and classifying and storing the search records of each user in the user search log data according to the time category to which the historical search time belongs, the search preference of the user in different dates and different time periods can be classified, so that in the subsequent search, the search keyword recommendation of the real search requirement of the user can be more accurately and more understood according to the search preference of the user search time.

Further, the dividing the time into a plurality of time categories according to date attributes and/or time period attributes includes:

That is, in one embodiment, the content of the search may have a tendency to be considered by the user before and after different holidays, and during working hours and non-working hours, for example, the probability of searching for "dropoff", "zongzi" and "dragon boat" before and after the noon is higher than usual. Searching for "dropoff" during the noon, then the user has a high probability of wanting to know the calendar of the noon, and then the search engine recommends "the calendar of the noon" for the user; if the user searches for "yield" at ordinary times, the user may want to know the literature of the yield better, and then it is more appropriate to recommend "yield poetry" for the user. As another example, content searched during working hours is more prone to be relevant to working properties, while content searched during non-working hours is more prone to life and leisure.

Therefore, in this embodiment, the time is divided into a legal holiday (such as a primordial year, a spring festival, and a Qing dynasty festival … …), a non-legal holiday (such as an lover's festival, a tree planting festival, and a green festival … …), and a non-festival (i.e., a date outside the division fixed holiday and the non-legal holiday), and the other holiday and the non-festival are further divided into an active time (such as 9:00-18:00) and a non-active time (such as 18:00-next day 9:00) according to the time period attribute, so that five types of time can be obtained, namely, the legal holiday, the active time in the other festival, the non-active time in the other festival, the active time in the non-festival, and the non-active time in the non-festival.

Therefore, the time is reasonably classified by considering search habits and preferences of users in different holidays and different working hours, so that the user search records can be classified according to reasonable time categories, and further the reasonability and accuracy of recommendation according to the search preferences of the users in each time category are ensured.

Step 104, determining a first candidate word set associated with the search word from the target user search record set.

In this step, a search record associated with the search word may be found from the target user search record set as a candidate word set to be recommended, that is, the first candidate word set, where the association with the search word may refer to correlation or similarity. For example, each search record in the target user search record set (i.e., a historical search term) may be compared to the search term for similarity, e.g., calculating a similarity, where the similarity may be greater than a threshold, e.g., greater than 0.8, and search records in the target user search record set that are similar to the search term may be added as candidate terms to the first candidate term set.

And 105, determining a recommended word set based on the first candidate word set and recommending.

In this step, the recommended word set may be determined based on the candidate words in the first candidate word set, specifically, a part of candidate words may be selected from the first candidate word set, for example, a plurality of candidate words with higher search frequency or higher similarity with the search word may be selected and added as recommended words to the recommended word set, or all candidate words in the first candidate word set may be added as recommended words to the recommended word set, and thus the first candidate word set may be directly used as the recommended word set. And finally, recommending the recommended words in the recommended word set to the user, for example, displaying the recommended words in the recommended word set one by one near the position where the user inputs the search word.

A specific embodiment of determining the first candidate word set is described below, by way of example, with reference to fig. 3:

the user log data is analyzed, and date and time are extracted, namely, the structured data of the search record. And classifying the time according to a certain rule. For example, different holidays are distinguished by year, month and day, the working time and the non-working time are distinguished by time minutes and seconds. For different moments of the day, it is distinguished into working time and non-working time. For example, 9:00-18:00 are working time intervals, and 18:00-9:00 of the next day are non-working time intervals. For each time (year, month, day, time, and second), whether the time belongs to a certain festival is firstly divided according to the year, month, day, and in one embodiment, 3 days before a certain festival can be divided into the range of the festival; for example: the national festival is 10 months 1 day to 10 months 7 days, and the search content in the period of 9 months 28 days to 10 months 7 days can be all in the national festival range in the division of the application, because before festival, the user usually prepares for spending the festival by searching the related content of the festival. If the time is determined to belong to legal holidays by the classification of the year, month and day, whether the time belongs to working time or not is further judged according to time division seconds. If the time is not legal holidays, it is further determined whether the time is working time based on time-division seconds. After the search keyword is input, the current time information can be obtained, the search keyword is classified according to the time, and the word with the similarity higher than 0.8 with the search keyword in the user search record in the classifying time period is found to be used as the candidate recommended word of the search, so that the candidate recommended word set 1, namely the first candidate word set, is obtained.

The following is illustrative:

processing log data to obtain structured data:

2020-05-07 10:34:44) — "operational steps of relational database"

2020-05-07:20:30:03- "high-Bean cotyledon score movie"

2020-06-25:08:18:52-custom of England "

……

Time is classified (for example in 2020):

legal holidays: { the primary denier 2019.12.29-2020.01.01,

spring festival 2020.01.21-2020.02.02,

the Qingming festival 2020.04.01-2020.04.06,

……}

other holidays:

{ scenario 2020.02.11-2020.02.14: { on time 9:00-18:00, off time 18:00-next day 9:00},

tree planting sections 2020.03.09-2020.03.12: { on time 9:00-18:00, off time 18:00-next day 9:00},

……}

non-holiday: { on time 9:00-18:00, off time 18:00-next day 9:00}

And sorting the user log records according to time categories, wherein each time category obtains a user search record set.

The user enters search keywords: contract templates for house rentals (search time: 2020-03-2110:34:22)

Classifying the time for searching for the keyword: belonging to working time periods in non-holidays

Searching candidate words with high similarity (similarity > 0.8) with the keywords in the user search record set corresponding to the time classification: [ Jiangxi House lease contract, house lease contract template, … … ]

Obtaining a candidate recommended word set 1: [ Jiangxi House lease contract, house lease contract template, … … ].

Optionally, after the obtaining the input search term, before the step 105, the method further includes:

determining a word vector of the search word;

the step 105 includes:

In one embodiment, the recommended word set may also be constructed according to the semantic relationship, so that the recommended word set can be finally constructed by combining the semantic relationship and preference information of the user searching behavior in different dates and times.

In this embodiment, a large-scale corpus may be utilized in advance to construct a word vector set of each word, and another candidate word set related to the search word may be determined based on a similarity relationship between word vectors.

Specifically, for a search term entered by a user, it may be first pre-processed, including removing special characters, such as "/,? The following is carried out "etc.; in addition, the content of the query input by the user can be not only a word or phrase but also a sentence, so that the search word input by the user can be segmented, and the segmented imaginary word or meaningless part, such as ' or ' and ' or ' yes ', and the like, can be removed. Then, candidate word sets can be searched for real words obtained after word segmentation according to semantic relation similarity, and the similarity calculation method can be to convert words into word vectors, and the semantic similarity among the words can be measured by calculating cosine distances among the word vectors.

For the word vector acquisition method, a bi-directional multi-head attention mechanism language model pre-trained on a large-scale unsupervised corpus can be adopted, and the word vector set can be obtained by a bi-directional coding representation (Bidirectional Encoder Representations from Transformers, bert) model based on a converter. The word vector obtained in this way can solve the problem of word ambiguity in Chinese to a certain extent. For example, for the phrases "apple growth happiness in this orchard" and "i get more familiar with the apple system than android", the words "apple" appear in both sentences, and it is obvious that they have distinct meanings, and the word vectors constructed according to the conventional one-hot (one-hot) vector or word2vec word vector method cannot distinguish between different meanings, whereas the word vectors obtained by the bidirectional multi-headed attention mechanism model can distinguish between different meanings of the same vocabulary.

For the search term, a bi-directional multi-headed attentiveness-mechanism language model may also be employed to represent the search term as a word vector. The real words obtained after the word segmentation of the search words need to be found out in sequence, and particularly whether the real words are added into a candidate word set can be judged by calculating the similarity between the word vector of each real word and the word vector in the word vector set. For example, the similarity threshold may be set to 0.8, and if the similarity between a real word and a word in the word vector set is greater than the threshold, the word is added to the candidate word set of the real word. After the candidate word sets of all the real words are found, the candidate words can be used for replacing at the corresponding positions of the real words, and the real words at any positions can be replaced. Therefore, more candidate words to be recommended can be obtained, but a large amount of noise exists in the candidate words, and the situation that the candidate words to be recommended are not smooth exists, so that words with high smoothness in the candidate words to be recommended can be screened by means of a language model, and a second candidate word set is obtained.

In this way, a first candidate word set can be constructed according to preference information of user search behaviors in different dates and times, and a second candidate word set can be constructed according to semantic relations among words, so that a recommended word set can be determined and recommended based on the first candidate word set and the second candidate word set, and particularly, part of candidate word sets can be screened from the first candidate word set and the second candidate word set to recommend, and similar candidate words in the first candidate word set and the second candidate word set can be fused to recommend, so that high-quality recommended keywords which can meet the requirements of users can be finally obtained.

Therefore, by further combining the semantic relation to construct the recommended word set, more matched recommended words can be found in the aspect of semantic similarity, and the word multi-meaning problem of the words can be solved to a certain extent.

That is, in one embodiment, the keyword recommendation method shown in fig. 2 may be used, including: acquiring search keywords input by a user; providing all meaning displays of the search content for the user to select a real search intention; after the search content is determined, constructing a recommended word set 1 according to the semantic relation; constructing a recommended word set 2 according to preference information of the user searching behavior in different dates and times; post-processing the recommended word set; and obtaining the final recommended keywords.

A specific embodiment of determining the second candidate word set is described below, by way of example, with reference to fig. 4:

the user enters search keywords: contract template for house rentals

Word segmentation: related/house/lease/contract/template

Removing the imaginary word and nonsensical word: [ related, house, rent, contract, template ]

Word vector set: [ House: word vector representations (0.663896 0.921862-1.805689 … …); contract: word vector representations (0.669016 2.295912-0.409784 … …);

house: word vector representations (0.456529-0.898323-1.437611 … …);

……]

a set of similar words of the keyword real words (the similarity between the candidate similar words of each word and the original real word is greater than a set threshold value of 0.8):

{ about: [ about, related ],

the house comprises the following steps: [ House, house ],

leasing: [ lease, lease ],

contract: [ agreement, contract, convention ],

and (3) a template: mode, model, stamp }

The replaced candidate word set: contract template for House rentals, contract template for House rentals … … ]

Candidate recommended word set 2, namely a second candidate word set, screened by means of the language model: contract templates for house rentals, contract templates for house rentals … ….

Further, the determining a recommended word set based on the first candidate word set and the second candidate word set and recommending includes:

That is, in one embodiment, the candidate words in the first candidate word set may be compared with the candidate words in the second candidate word set in a similar manner, specifically, for each candidate word in the first candidate word set, the candidate words in the second candidate word set need to be compared with each candidate word in the first candidate word set separately, so as to determine whether the candidate words in the first candidate word set are similar to the candidate words in the second candidate word set, so that similar candidate words in the two candidate word sets and dissimilar candidate words, that is, a similar comparison result, may be obtained, where the similarity may refer to that the similarity is greater than a predetermined threshold.

Then, based on the similarity comparison result, the candidate words in the first candidate word set as recommended words may be determined, a first recommended word set may be obtained, and the candidate words in the second candidate word set as recommended words may be determined, so as to obtain a second recommended word set, for example, candidate words in the first candidate word set similar to the second candidate word set may be determined as recommended words, and for candidate words in the first candidate word set dissimilar to the second candidate word set, whether to be recommended words may be further determined according to other factors, such as a search frequency, a similarity with the search words, and the like.

Finally, the first set of recommended words and the second set of recommended words may be recommended to the user, e.g., the recommended words in the first set of recommended words and the recommended words in the second set of recommended words may be displayed at search locations, respectively.

In this way, the recommendation word set is determined by comparing the candidate words in the first candidate word set with the candidate words in the second candidate word set in a similar manner, so that the accuracy of keyword recommendation can be further ensured.

Further, the determining, based on the similarity comparison result, the first recommended word set in the first candidate word set and the second recommended word set in the second candidate word set includes:

That is, after comparing the similarity of the candidate words in the first candidate word set and the candidate words in the second candidate word set, both the similar candidate words in the first candidate word set (i.e., the first group of candidate words) and the dissimilar candidate words in the second candidate word set (i.e., the second group of candidate words) may be recommended as recommended words, while the similar candidate words in the second candidate word set are removed, and only the dissimilar candidate words in the second candidate word set (i.e., the third group of candidate words) may be recommended as recommended words. The similar candidate words are candidate words similar to one candidate word in the other candidate word set, and the dissimilar candidate words are candidate words dissimilar to any one candidate word in the other candidate word set.

For example, if the ith candidate word in the first candidate word set is similar to the jth candidate word in the second candidate word set, the ith candidate word is a similar candidate word in the first candidate word set, and the first recommended word set may be added, and the jth candidate word is a similar candidate word in the second candidate word set and is deleted, without adding the second recommended word set.

For another example, the kth candidate word in the first candidate word set is dissimilar to each candidate word in the second candidate word set, and the kth candidate word in the second candidate word set is dissimilar to each candidate word in the first candidate word set, and the kth candidate word is a dissimilar candidate word in the first candidate word set and may be added to the first recommended word set, and the kth candidate word is a dissimilar candidate word in the second candidate word set and may be added to the second recommended word set.

When the first group of candidate words, the second group of candidate words and the third group of candidate words are determined to be recommended word sets, the recommendation sequence of the candidate words of each group can be determined, specifically, in order to ensure the accuracy and efficiency of recommendation, the first group of candidate words can be recommended in the forefront, the second group of candidate words can be recommended in the middle, the third group of candidate words can be recommended in the last, namely, the recommended words are displayed according to the front-back position sequence of the first group of candidate words, the second group of candidate words and the third group of candidate words.

Therefore, the candidate words in the first candidate word set are more in accordance with the search preference of the user in the current search period, and the candidate words in the second candidate word set are more relevant to the current search word, so that the recommendation of the non-similar candidate words in the second candidate word set can be performed by recommending both the similar candidate words and the non-similar candidate words in the first candidate word set, the accuracy of keyword recommendation can be ensured, and the comprehensiveness of keyword recommendation can be ensured. In addition, by recommending the first group of candidate words, the second group of candidate words and the third group of candidate words after sequencing according to the sequence of the first group of candidate words, the recommended words which most probably meet the requirements of the user can be displayed and recommended to the user in front, and the user can notice the search words which meet the requirements of the user quickly.

The specific embodiment of determining the final recommended word set described above is described below by way of example with reference to fig. 5:

as shown in fig. 5, after the candidate recommended word set 1 (first candidate word set) and the candidate recommended word set 2 (second candidate word set) are obtained, the final recommended words are ranked by post-processing, and the recommended words that can more satisfy the user's expectations are displayed in the front. Specifically, the similarity between the words { a, d, e } in the candidate recommended word set 1 and the words { A, B, C } in the candidate recommended word set 2 can be calculated, if a is similar to a, a is reserved, a is deleted, and a is arranged at the forefront. The candidate recommended word set 1 and the candidate recommended word set 2 are arranged according to the sequence of the words in the candidate recommended word set 1, wherein the words with the similarity exceeding the threshold value are arranged in the middle, and the non-similar words in the candidate recommended word set 1 are arranged behind the non-similar words in the candidate recommended word set 2. Finally, a recommended word set is obtained: { a, d, e, B, C }.

Existing keyword recommendation techniques typically acquire words similar to or related to search keywords based on search related documents, and cannot clearly push keywords that meet user expectations when user input information is ambiguous. According to the method and the device, ambiguous search contents are displayed for further selection by the user, the search intention of the user is clear, and accurate recommendation is achieved. According to the method, the word vectors are trained on a large-scale corpus by using a bidirectional multi-head attention mechanism model, the effect of fitting word semantic information is better than that of a traditional feature vector and word2vec method, a basis is provided for recommending keywords based on a semantic similarity method, and the method can solve the word multi-meaning problem of Chinese to a certain extent. In addition, the application fully considers the preference of the user search behavior on different dates and times. Because the content searched by the user during different holidays and working hours or non-working hours can be greatly different, for example, the searching obvious preference can appear during working hours and working hours before and after the festival, the historical user logs are analyzed and classified according to time (holidays, other holiday working hours, other holiday non-working hours, non-holiday working hours and non-holiday non-working hours). The recommendation of the user search keywords and the related recommendation are combined with the time category to better meet the requirements of the user.

According to the keyword recommendation method, input search words are obtained, and time information for inputting the search words is obtained; determining a target time category to which the time information belongs; determining a target user search record set corresponding to the target time category from a pre-established user search log database, wherein a plurality of user search record sets respectively corresponding to different time categories are stored in the user search log database; determining a first set of candidate words associated with the search term from the set of target user search records; and determining a recommended word set based on the first candidate word set and recommending. Therefore, the search records of the user are classified according to time by considering the search preference of the user at different times, so that the search records corresponding to the search time of the user can be recommended to the user according to the information which is more in line with the expectations of the user, and the recommendation accuracy is improved.

The embodiment of the application also provides a keyword recommendation device. Referring to fig. 6, fig. 6 is a block diagram of a keyword recommendation apparatus provided in an embodiment of the present application. Since the principle of the keyword recommendation device for solving the problem is similar to that of the keyword recommendation method in the embodiment of the present application, the implementation of the keyword recommendation device can refer to the implementation of the method, and the repetition is omitted.

As shown in fig. 6, the keyword recommendation apparatus 600 includes:

a first obtaining module 601, configured to obtain an input search term, and obtain time information of inputting the search term;

a first determining module 602, configured to determine a target time category to which the time information belongs;

a second determining module 603, configured to determine, from a pre-established user search log database, a target user search record set corresponding to the target time category, where the user search log database stores a plurality of user search record sets corresponding to different time categories respectively;

a third determining module 604 for determining a first set of candidate words associated with the search term from the set of target user search records;

and a recommendation module 605, configured to determine a recommended word set based on the first candidate word set, and perform recommendation.

Optionally, the keyword recommendation apparatus 600 further includes:

the second acquisition module is used for acquiring user search log data;

Optionally, the first acquisition module 601 includes:

Optionally, the keyword recommendation apparatus 600 further includes:

the recommendation module 605 is configured to determine a recommended word set and recommend the recommended word set based on the first candidate word set and the second candidate word set.

Optionally, the recommendation module 605 includes:

The keyword recommendation apparatus 600 provided in the embodiment of the present application may execute the above method embodiment, and its implementation principle and technical effects are similar, and this embodiment will not be described herein again.

The keyword recommendation apparatus 600 of the embodiment of the present application obtains an input search term, and obtains time information for inputting the search term; determining a target time category to which the time information belongs; determining a target user search record set corresponding to the target time category from a pre-established user search log database, wherein a plurality of user search record sets respectively corresponding to different time categories are stored in the user search log database; determining a first set of candidate words associated with the search term from the set of target user search records; and determining a recommended word set based on the first candidate word set and recommending. Therefore, the search records of the user are classified according to time by considering the search preference of the user at different times, so that the search records corresponding to the search time of the user can be recommended to the user according to the information which is more in line with the expectations of the user, and the recommendation accuracy is improved.

The embodiment of the application also provides electronic equipment. Because the principle of solving the problem of the electronic device is similar to that of the keyword recommendation method in the embodiment of the present application, the implementation of the electronic device may refer to the implementation of the method, and the repetition is omitted. As shown in fig. 7, an electronic device according to an embodiment of the present application includes:

the processor 700 is configured to read the program in the memory 720, and execute the following procedures:

determining a target time category to which the time information belongs;

Wherein in fig. 7, a bus architecture may comprise any number of interconnected buses and bridges, and in particular one or more processors represented by processor 700 and various circuits of memory represented by memory 720, linked together. The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, power management circuits, etc., which are well known in the art and, therefore, will not be described further herein. The bus interface provides an interface. The processor 700 is responsible for managing the bus architecture and general processing, and the memory 720 may store data used by the processor 700 in performing operations.

Optionally, the processor 700 is further configured to read the program in the memory 720, and perform the following steps:

acquiring user search log data;

Acquiring a first search term which is initially input;

determining a word vector of the search word;

The electronic device provided in the embodiment of the present application may execute the above method embodiment, and its implementation principle and technical effects are similar, and this embodiment will not be described herein again.

Furthermore, a computer readable storage medium of an embodiment of the present application is configured to store a computer program, where the computer program is executable by a processor to implement the steps of the method embodiment shown in fig. 1.

In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

In addition, each functional unit in each embodiment of the present application may be integrated in one processing unit, or each unit may be physically included separately, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in hardware plus software functional units.

The integrated units implemented in the form of software functional units described above may be stored in a computer readable storage medium. The software functional unit is stored in a storage medium, and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform part of the steps of the transceiving method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

While the foregoing is directed to the preferred embodiments of the present application, it should be noted that modifications and adaptations to those embodiments may occur to one skilled in the art and that such modifications and adaptations are intended to be comprehended within the scope of the present application without departing from the principles set forth herein.

Claims

1. A keyword recommendation method, comprising:

determining a target time category to which the time information belongs;

2. The method of claim 1, wherein prior to determining the set of target user search records corresponding to the target time category from a pre-established user search log database, the method further comprises:

acquiring user search log data;

3. The method according to claim 2, wherein the dividing the time into a plurality of time categories according to date attributes and/or time period attributes comprises:

4. A method according to any one of claims 1 to 3, wherein the obtaining the entered search term comprises:

acquiring a first search term which is initially input;

5. The method of claim 1, wherein after the obtaining the input search term, the determining a recommended set of terms based on the first candidate set of terms and before recommending, the method further comprises:

Determining a word vector of the search word;

6. The method of claim 5, wherein the determining and recommending a recommended set of words based on the first candidate set of words and the second candidate set of words comprises:

7. The method of claim 6, wherein the determining a first set of recommended words in the first set of candidate words and a second set of recommended words in the second set of candidate words based on the similarity comparison result comprises:

8. A keyword recommendation apparatus, comprising:

9. An electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor; the method according to any one of claims 1 to 7, characterized in that the processor is adapted to read a program in a memory to implement the steps in the keyword recommendation method.

10. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the keyword recommendation method of any one of claims 1 to 7.