CN113886539A

CN113886539A - Method and device for recommending dialect, customer service equipment and storage medium

Info

Publication number: CN113886539A
Application number: CN202111149117.2A
Authority: CN
Inventors: 刘明
Original assignee: Weikun Shanghai Technology Service Co Ltd
Current assignee: Weikun Shanghai Technology Service Co Ltd
Priority date: 2021-09-29
Filing date: 2021-09-29
Publication date: 2022-01-04

Abstract

The embodiment of the application provides a conversation recommendation method, a conversation recommendation device, customer service equipment and a storage medium, which relate to the technical field of artificial intelligence and comprise the following steps: acquiring first dialogue information, acquiring second dialogue information related to the first dialogue information based on a knowledge graph, calling a service classification model to process the first dialogue information and the second dialogue information, and determining a target service scene; and acquiring a target node model corresponding to the target service scene, calling the target node model to process the first dialogue information and the second dialogue information, determining a target node in the target service scene, and outputting a dialogue associated with the target node. The accuracy of the target node association conversation can be effectively improved. The present application may relate to blockchain techniques, such as target products under a target node may be written into a blockchain. The application also relates to the technical field of digital medical treatment, for example, the first dialogue information is used for describing information of the technical field of digital medical treatment.

Description

Method and device for recommending dialect, customer service equipment and storage medium

Technical Field

The application relates to the technical field of artificial intelligence, in particular to a conversation recommendation method and device, customer service equipment and a storage medium.

Background

With the rapid development of internet technology, people have an increasing demand for customer service staff. Due to the factors of long training period of manual customer service or high labor cost and the like, the customer service robot gradually enters the lives of people. The customer service robot can search corresponding reply contents according to questions provided by the user. However, as the integration degree of the business system is higher, the accuracy of the reply content of the existing customer service robot is lower. Therefore, it is an important research question to improve the accuracy of the response content of the customer service robot.

Disclosure of Invention

The embodiment of the application provides a conversation recommendation method and device, customer service equipment and a storage medium, and can effectively improve the accuracy of the customer service robot in responding to contents.

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

acquiring first dialogue information, and acquiring second dialogue information related to the first dialogue information based on a knowledge graph;

calling a service classification model to process the first dialogue information and the second dialogue information and determine a target service scene;

acquiring a target node model corresponding to a target service scene, calling the target node model to process the first dialogue information and the second dialogue information, and determining a target node in the target service scene;

and outputting the target node association language, wherein the target node association language is response information corresponding to the first dialogue information.

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

an acquisition unit configured to acquire first dialogue information and acquire second dialogue information associated with the first dialogue information based on a knowledge graph;

the processing unit is used for calling the service classification model to process the first dialogue information and the second dialogue information and determine a target service scene;

the processing unit is further configured to obtain a target node model corresponding to the target service scene, and call the target node model to process the first session information and the second session information, so as to determine a target node in the target service scene;

and the output unit is used for outputting the target node association language, and the target node association language is response information corresponding to the first dialogue information.

In a third aspect, an embodiment of the present application provides a customer service device, where the customer service device includes an input interface and an output interface, and the customer service device further includes:

a processor adapted to implement one or more instructions; and the number of the first and second groups,

a computer storage medium having stored thereon one or more instructions adapted to be loaded by a processor and to perform the method of the first aspect.

In a fourth aspect, the present application further provides a computer-readable storage medium, which stores computer program instructions, and when the computer program instructions are executed by a processor, the computer program instructions are configured to perform the method of the first aspect.

In the embodiment of the application, the customer service equipment can acquire first dialogue information and acquire second dialogue information related to the first dialogue information based on the knowledge graph; calling a service classification model to process the first dialogue information and the second dialogue information, determining a target service scene, calling a target node model corresponding to the target service scene to process the first dialogue information and the second dialogue information, determining a target node under the target service scene, and outputting a dialogue associated with the target node, wherein the dialogue associated with the target node is response information corresponding to the first dialogue information. Due to the fact that the business classification model and the node model are combined, the target business scene is determined according to the business classification model, and then the target node is determined according to the target node model under the target business scene, accuracy of the target node can be effectively improved, and accuracy of word operation related to the target node is further improved. In addition, the first dialogue information and the second dialogue information are combined, and the target service scene determined based on the first dialogue information and the second dialogue information is more accurate, so that the target node determined based on the target node model in the target service scene is more accurate, and the accuracy of the dialogue associated with the target node is effectively improved.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a schematic diagram of an architecture of a customer service system according to an embodiment of the present application;

fig. 2 is a schematic flowchart of a conversation recommendation method provided in an embodiment of the present application;

FIG. 3 is a flow chart of another conversational recommendation method provided by an embodiment of the application;

fig. 4 is a schematic structural diagram of a speech recommendation device according to an embodiment of the present application;

fig. 5 is a schematic structural diagram of a customer service device according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

With the rapid development of internet technology, electronic commerce gradually enters the daily life of people, and the demand of people for customer service personnel is increasing. Due to the factors of long training period of manual customer service or high labor cost, the customer service robot gradually becomes a research focus. However, as the integration level of the service system is higher and higher, the customer service robot cannot accurately identify the problem of the user during the conversation process between the user and the customer service robot, so that the customer service robot cannot accurately respond. Therefore, it is an important research question to improve the accuracy of the response content of the customer service robot.

Based on this, the embodiment of the application provides a conversation recommendation method, in which a customer service device may obtain first conversation information, and obtain second conversation information associated with the first conversation information based on a knowledge graph; calling a service classification model to process the first dialogue information and the second dialogue information, determining a target service scene, calling a target node model corresponding to the target service scene to process the first dialogue information and the second dialogue information, determining a target node under the target service scene, and outputting a dialogue associated with the target node, wherein the dialogue associated with the target node is response information corresponding to the first dialogue information. The dialect associated with the target node can accurately reply the first dialogue information, and the accuracy of the reply content of the customer service robot is effectively improved.

It should be noted that, the service classification model and the target node model may be constructed based on a machine learning algorithm in an Artificial Intelligence (AI) technology. The artificial intelligence is a theory, a method, a technology and an application system which simulate, extend and expand human intelligence by using a digital computer or a machine controlled by the digital computer, sense the environment, acquire knowledge and obtain the best result by using the knowledge. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a robot technology, a biological recognition technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

In one embodiment, the conversational recommendation method mentioned in the present application can be applied to a customer service system. As shown in fig. 1, the customer service system may include at least: a customer service device 11 and a terminal device 12, wherein the customer service device 11 is a device running a service classification model and a node model, and the customer service device 11 can be a customer service robot. Optionally, the customer service device 11 may be a server as shown in fig. 1, where the server may be an independent physical server, may also be a server cluster or a distributed system formed by a plurality of physical servers, and may also be a cloud server that provides basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a Network service, cloud communication, a Content Delivery Network (CDN), a middleware service, a domain name service, a security service, a big data and artificial intelligence platform, and the like. The customer service device 11 may also be a terminal device, wherein the terminal device may include but is not limited to: smart phones, tablets, laptops, wearable devices, desktop computers, and the like. Terminal device 12 may be, among other things, a terminal device associated with a user. A user may interact with the customer service device through terminal device 12.

Please refer to fig. 2, which is a flowchart illustrating a conversation recommendation method according to an embodiment of the present application. As shown in fig. 2, the conversational recommendation method includes S201-S204:

s201: first dialogue information is obtained, and second dialogue information related to the first dialogue information is obtained based on the knowledge graph.

The dialog information mentioned in the embodiments of the present application includes a dialog text for describing an arbitrary field. For example, the session information may include session text in the medical field, such as personal health profile, prescription, examination report, and the like.

In one embodiment, the customer service device may obtain initial information of any modality and obtain dialog text (i.e., dialog information) based on the initial information. The initial information may include, but is not limited to, one or more of text information, voice information, and image information. When the initial information is text information, then the initial information may be directly used as the dialogue information. When the initial information is voice information, the voice information can be converted into text information based on an offline or cloud voice recognition technology. When the initial information is image information, text information in the image information may be extracted using an Optical Character Recognition technique (OCR).

Wherein the first dialog information may be a dialog actively initiated by a terminal device associated with the user. The second session information may refer to a session between the customer service device and the terminal device associated with the user, and the second session information may be a session actively initiated by the customer service device or a session actively initiated by the terminal device associated with the user.

The knowledge graph is a graph organization form which associates various entities or concepts existing in the real world through semantics, and a graph structure is mainly formed by nodes, edges and node attributes. Wherein each entity or concept is identified by a globally unique determined code, each attribute-value pair is used to characterize an intrinsic attribute of the entity or concept, and an edge is used to connect two entities or concepts together to characterize an association between them.

In one embodiment, the customer service device may acquire the dialogue data including the dialogue information, the attribute of the dialogue information, and the association between the dialogue information, and construct the knowledge graph according to the dialogue information, the attribute of the dialogue information, and the association between the dialogue information. When the first dialogue information is acquired, second dialogue information having an association relationship with the first dialogue information may be acquired from the knowledge map. Specifically, the obtaining of the second session information having an association relationship with the first session information from the knowledge graph may include: and taking the first dialogue information as a starting point, and acquiring dialogue information which has an incidence relation with the first dialogue information in the knowledge graph by using a breadth-first search algorithm or a depth-first search algorithm to obtain second dialogue information. The breadth-first search algorithm or the depth-first search algorithm is an existing search algorithm and is not described herein again.

In one embodiment, the knowledge graph referred to herein can be a domain knowledge graph. The association relationship is a domain relationship. Optionally, the second session information is the same as the domain to which the first session information belongs, or a correlation degree between the second session information and the domain to which the first session information belongs is greater than a preset threshold. For example, the first session information is "what the exchange rate of rmb to dollar is", and the second session information is "store one ten thousand yuan", both of which belong to the financial field.

In another embodiment, the knowledge graph referred to herein may be a temporal knowledge graph. The association relationship is a temporal relationship. Optionally, a duration between the time of generating the second dialog information and the time of generating the first dialog information is less than a preset duration.

S202: and calling a service classification model to process the first dialogue information and the second dialogue information and determine a target service scene.

In one embodiment, the customer service device may perform feature extraction on the first dialogue information and the second dialogue information to obtain a feature vector, call the service classification model to process the feature vector, determine the prediction probability of the feature vector in each candidate service scenario, and determine the target service scenario according to the prediction probability in each candidate service scenario.

Optionally, the step of performing, by the customer service device, feature extraction on the first dialog information and the second dialog information to obtain a feature vector includes: determining target dialogue information based on the first dialogue information and the second dialogue information, performing word segmentation on the target dialogue information to obtain segmented target dialogue information, mapping the segmented target dialogue information to obtain word vectors, and calling a feature extraction model to perform feature extraction on the word vectors to obtain the feature vectors.

In one embodiment, the first dialog information and the second dialog information may be processed based on an attention mechanism to determine the target dialog information. Specifically, the customer service device may obtain the attention weight of the first session information and the attention weight of the second session information, respectively, and process the first session information and the second session information based on the attention weight of the first session information and the attention weight of the second session information to determine the target session information. Wherein attention mechanism means that attention can be focused on the actually important feature by attention weight. For example, when the customer service device focuses more on the first dialogue information, the attention weight of the first dialogue information may be set to be greater than that of the second dialogue information. For another example, when the second dialog information is more focused, the attention weight of the second dialog information may be set to be greater than that of the first dialog information.

In one embodiment, the customer service device may perform word segmentation processing on the target dialog information in multiple ways to obtain the segmented target dialog information. Optionally, the customer service device may perform word segmentation processing on the target dialogue information based on a dictionary word segmentation algorithm (such as a forward maximum matching method, a reverse maximum matching method, a bidirectional matching word segmentation method, and the like), or may also perform word segmentation processing on the target dialogue information based on a statistical machine learning algorithm (such as a hidden markov model, a conditional random field model, an SVM algorithm, a deep learning algorithm, and the like). The embodiment of the present application does not limit this.

In one embodiment, the customer service device may map the segmented target dialog information through a coding algorithm to obtain a word vector. Optionally, the customer service device may use a common Word2vec network to perform high-dimensional mapping on the segmented target dialogue information. The Word2vec network uses a Huffman tree as a data structure to replace a traditional Deep Neural Network (DNN), leaf nodes of the Huffman tree are used for serving as neuron output, and the number of the leaf nodes is set according to the size of a dictionary.

In one embodiment, the feature extraction model may include a convolutional neural network and an attention layer. Optionally, the customer service device may obtain the local feature from the mapped word vector by using a convolutional neural network. Specifically, the word vectors may be divided into two-dimensional matrices of the same shape, and input into a convolutional neural network, and the convolutional neural network may perform sliding convolution to obtain the local features. The convolutional neural network may comprise a plurality of convolutional layers, for example, the convolutional neural network comprises four convolutional layers of 1 x 3, 1 x 4, 1 x 5 and 1 x 6. Optionally, considering that the word vector may lack some information, the local feature vector derived from the word vector may also lack some information. Therefore, local features can be extracted by using the attention mechanism in the attention layer, and a global-based feature vector can be obtained. Optionally, a BilSTM network based on an attention mechanism may be used to perform feature extraction on the local features to obtain a global-based feature vector. Due to the serialization characteristic of the target dialogue information, the features are extracted by using the BilSTM network, so that the word order features can be reserved to the maximum extent.

The business classification model may be a classification model trained based on a machine learning algorithm. Among other things, the machine learning algorithm may include, but is not limited to, one or more of the following: decision Tree (DT) algorithm, Rocchio algorithm, extreme Gradient Boosting (xgboost) algorithm, Naive Bayes (Naive Bayes, NB) algorithm, Linear Discriminant Analysis (LDA), Support Vector Machine (SVM) algorithm, Random Forest (RF) algorithm, and Logistic Regression (LR) algorithm. For example, the service classification model may be one or more of a decision tree model obtained by training based on a decision tree algorithm, a rocchiao model obtained by training based on a rocchiao algorithm, an xgboost model obtained by training based on an extreme gradient lifting algorithm, an NB model obtained by training based on a naive bayesian algorithm, an LDA model obtained by training based on a linear discriminant analysis, a support vector machine model obtained by training based on a support vector machine algorithm, a random forest model obtained by training based on a random forest, and a logistic regression model obtained by training based on a logistic regression algorithm.

Due to the high integration level of the existing service system, a plurality of candidate service scenarios may be involved in one service system. For example, in business systems in the financial field, one business system may relate to a credit card business scenario and a savings business scenario. For another example, in a business system in the medical field, one business system may relate to a visit business scenario and a rehabilitation business scenario. And so on.

In one embodiment, the customer service device may first perform processing according to the feature vector corresponding to the target session information, and determine the target service scenario from a plurality of candidate service scenarios of the service system. Specifically, the customer service device may invoke the service classification model to process the feature vector, determine the prediction probability of the feature vector in each candidate service scenario, determine the maximum prediction probability from the prediction probabilities in each candidate service scenario, and determine the candidate service scenario corresponding to the maximum prediction probability as the target service scenario. The higher the prediction probability is, the higher the confidence coefficient representing that the feature vector belongs to the candidate service scenario is, and the lower the prediction probability is, the lower the confidence coefficient representing that the feature vector belongs to the candidate service scenario is.

In one embodiment, before calling the traffic classification model to process the feature vectors and determine the target traffic scene, the traffic classification model needs to be trained. Specifically, the training process of the business classification model may include s11-s12:

s 11: a set of dialog samples is obtained and a feature vector for each dialog sample in the set of dialog samples is determined.

It should be noted that the embodiment of determining the feature vector of the dialog sample is similar to the manner of determining the feature vector of the target dialog information, and reference may be specifically made to the foregoing description, which is not described herein again.

s12: classifying the conversation sample set according to the feature vector corresponding to each conversation sample to obtain conversation sample subsets under different categories, wherein the feature vector corresponding to the conversation sample in the conversation sample subset under one category corresponds to one candidate service scene.

Specifically, a business classification model can be constructed based on a machine learning algorithm, and parameter tuning is continuously performed in the construction process to construct an optimal business classification model. Specifically, the customer service device may classify the feature vector corresponding to each dialog sample in the dialog sample set by using the initial classification model to obtain dialog sample subsets of different categories, and update the initial parameters of the initial classification model according to the dialog sample subsets of different categories. And after multiple updates, the service classification model can be obtained through training. The feature vector corresponding to the dialog sample included in the dialog sample subset of one category corresponds to one candidate service scenario.

S203: and acquiring a target node model corresponding to the target service scene, calling the target node model to process the first dialogue information and the second dialogue information, and determining a target node in the target service scene.

In one embodiment, to improve the accuracy of the target node, different node models may be set for different candidate service scenarios. For example, it is assumed that the service system includes 3 candidate service scenes, which are respectively a candidate service scene 1, a candidate service scene 2, and a candidate service scene 3, and the node models corresponding to the 3 candidate service scenes are respectively a node model 1 corresponding to the candidate service scene 1, a node model 2 corresponding to the candidate service scene 2, and a node model 3 corresponding to the candidate service scene. When the target service scenario is the candidate service scenario 1, the node model 1 corresponding to the candidate service scenario 1 may be called to process the first dialogue information and the second dialogue information. When the target service scenario is the candidate service scenario 2, the node model 2 corresponding to the candidate service scenario 2 may be called to process the first dialogue information and the second dialogue information. When the target service scenario is the candidate service scenario 3, the node model 3 corresponding to the candidate service scenario 3 may be called to process the first dialogue information and the second dialogue information.

In one embodiment, the node model mentioned in the present application is similar to the traffic classification model, and the node model may also be a classification model trained based on a machine learning algorithm. In one embodiment, the customer service device may invoke a node model in a target service scenario to process the feature vector, determine a prediction probability of each candidate node of the feature vector in the target service scenario, and determine the target node according to the prediction probability of each candidate node.

In one embodiment, before the target node model in the target service scenario is called to process the feature vector and determine the target node, the target node model needs to be trained. Specifically, the training process of the target node model may include s21-s22:

s 21: and acquiring a target conversation sample set in a target service scene, and determining the feature vector of each target conversation sample in the target conversation sample set.

It should be noted that the embodiment of determining the feature vector of the target dialog sample is similar to the manner of determining the feature vector of the target dialog information, and reference may be specifically made to the foregoing description, which is not described herein again.

s22: classifying the target dialogue sample set according to the feature vector corresponding to each target dialogue sample to obtain target dialogue sample subsets under different categories, wherein the feature vector corresponding to the target dialogue sample in the target dialogue sample subset under one category corresponds to one candidate node under a target service scene.

Specifically, a target node model can be constructed based on a machine learning algorithm, and parameter tuning is continuously performed in the construction process to construct an optimal target node model. Specifically, the customer service device may classify, by using the initial classification model, the feature vector corresponding to each target conversation sample in the target conversation sample set to obtain different types of target conversation sample subsets, and update the initial parameters of the initial classification model according to the different types of target conversation sample subsets. And after multiple updates, the target node model can be obtained through training. The feature vector corresponding to the target dialogue sample contained in the target dialogue sample subset of one category corresponds to one candidate node in the target business scene.

S204: and outputting the target node association language, wherein the target node association language is response information corresponding to the first dialogue information.

In one embodiment, association needs to be configured for each candidate node in each candidate service scenario in advance. Alternatively, candidate nodes may be associated with a utterance through a node management interface. Specifically, a node management interface can be displayed, and the node management interface comprises a node identification entry column and a language entry column; and entering a candidate node in the node identification entry column and entering a dialect in the dialect entry column so that the candidate node indicated by the node identification entry column is associated with the dialect entered in the dialect entry column.

For example, a candidate node whose node identification content is "node 1" may be entered in the node identification entry column, and a new customer who may be registered on the jargon entry column for "days 6.1 to 6.20 may additionally obtain a 200-dollar coupon in addition to enjoying all the campaign offers of 618. When the customer service device determines that the target node is a candidate node with the node identifier of "node 1" through S201 to S203, the customer service device may output the utterance content entered in the above-mentioned utterance entry field.

Optionally, the customer service device may directly output the terminology associated with the target node; optionally, the customer service device may further send the target node-associated terminology to the user-associated terminal device, so that the user-associated terminal device outputs the target node-associated terminology.

As can be seen from the above description of the embodiment of the method shown in fig. 2, the tactical recommendation method shown in fig. 2 may determine the target node according to the first dialogue information and the second dialogue information. In some embodiments, product recommendations may also be made to the user based on the target nodes for marketing convenience. Based on this, the embodiment of the application also provides another dialectical recommendation method. As shown in fig. 3, the conversational recommendation method includes S301 to S303:

s301: user portrait information of a target user and a product recommendation model corresponding to the user portrait information are obtained.

The user portrait information may refer to information related to a user and describing features of the user. Optionally, the user representation information may include user attribute information and/or user tag information. The user attribute information may refer to attribute information inherent to the user, and the user attribute information may include, but is not limited to: one or more of name, nickname, age, gender, residence, nationality, occupation, constellation, blood type, and identification. The user tag information is obtained by abstracting and classifying a certain feature of a user, and specifically, the user tag information may be obtained based on user behavior data analysis. The user behavior data refers to behavior data generated when the user uses a business product, for example, when the user uses a financial business, the behavior data corresponding to the user in the financial business may include, but is not limited to, a deposit of the user, a loan product of the user, a loan number of the user, a loan amount of the user, a financial product purchased by the user, a financial activity in which the user participates, and the like. For another example, when the user uses the social service, the behavior data in the social service corresponds to the social service, which may include but is not limited to the social account number of the user, the rank of the user, the dynamics published by the user on the social platform, and so on.

The product recommendation model can also be a classification model obtained based on machine learning algorithm training.

S302: and calling a product recommendation model to process the user portrait information and the target node to obtain target index values of the candidate products under the target node under the evaluation index, and determining the target product according to the target index values of the candidate products.

Wherein, the evaluation index can be used for indicating the matching result of the candidate product and the user. The evaluation index may include: one or more of a conversation turn, customer satisfaction, customer complaint rate, problem resolution rate, and marketing conversion rate. The conversation turns can refer to the conversation turns of the user and the candidate products, and the larger the conversation turns are, the higher the matching degree of the candidate products and the user is. The customer satisfaction may be used to indicate a satisfaction score of the user for the candidate product, the higher the customer satisfaction, the higher the matching of the candidate product with the user. The customer complaint rate can also be used for indicating the satisfaction degree score of the user on the candidate product, and the lower the customer complaint rate is, the higher the matching degree of the candidate product and the user is. Wherein, the problem solving rate can also be used for indicating the satisfaction degree score of the user to the candidate product, and the higher the problem solving rate is, the higher the matching degree of the candidate product and the user is. Wherein the marketing conversion rate is used for indicating the commercial value of the candidate product, and the higher the marketing conversion rate is, the higher the matching degree of the candidate product and the user is.

In one embodiment, the number of evaluation indexes may be one. Then a maximum target index value may be determined from the target index values of each candidate product under the evaluation index, and the candidate product indicated by the maximum target index value may be determined as the target product. For example, the evaluation index may be a marketing conversion rate, and assuming that there are 4 candidate products under the target node, the target index values of the 4 candidate products under the evaluation index of the marketing conversion rate are respectively: if the target index value of the candidate product 1 is 4, the target index value of the candidate product 2 is 3, the target index value of the candidate product 3 is 5, and the target index value of the candidate product 4 is 2, it may be determined that the target index value of the candidate product 3 is the maximum, and the candidate product 3 may be determined as the target product.

In another embodiment, the number of the evaluation indexes may be multiple, and the reference value of each candidate product may be determined according to the attention mechanism, and then the target product may be determined according to the reference value of each candidate product. Specifically, the weight of each candidate product under the plurality of evaluation indexes may be obtained, the target index value of each candidate product under the plurality of evaluation indexes is weighted and summed according to the attention weight of each candidate product under the plurality of evaluation indexes, the reference value of each candidate product is determined, and the candidate product with the largest reference value is determined as the target product. The attention weight may be determined according to a business requirement, and if a certain evaluation index is important, the corresponding attention weight may be set to be larger, and if a certain evaluation index is not important enough, the corresponding attention weight may be set to be smaller. For example, if a candidate product is illustrated, if the candidate product has 2 evaluation indexes, the target index values under the 2 evaluation indexes are a and b, respectively. When the attention weights of the candidate product under the 2 evaluation indexes are q1 and q2, respectively, the target index values of each candidate product under the multiple evaluation indexes can be weighted and summed according to the attention weights of each candidate product under the multiple evaluation indexes, so that the reference value of the candidate product is a q1+ b q 2.

S303: and checking the target product, if the target product passes the checking, carrying out consensus verification on the target product through a consensus node in the block chain network, if the consensus verification passes, packaging the target product into a block, and writing the block into the block chain.

The block chain is a chain data structure formed by combining data blocks in a sequential connection mode according to a time sequence, and a distributed account book which can not be tampered and forged of data is guaranteed in a cryptographic mode. Multiple independent distributed nodes maintain the same record. The blockchain technology realizes decentralization and becomes a foundation for credible digital asset storage, transfer and transaction.

In the embodiment of the application, the customer service equipment can call the product recommendation model to process the user portrait information and the target node to obtain the target index values of the candidate products under the target node under the evaluation index, and the target product is determined according to the target index values of the candidate products. Due to the fact that the target indexes of the candidate products under the target nodes under the evaluation indexes are considered, the matching degree of the target products and the user is determined to be higher based on the target index values of the candidate products, and accurate recommendation of products interesting to the user is facilitated.

The embodiment of the application also discloses a conversational recommendation device, which can be a computer program (comprising program codes) running in the customer service equipment. The conversational recommendation device may perform the method shown in fig. 2 or fig. 3. Referring to fig. 4, the speech recommendation apparatus may operate as follows:

an obtaining unit 401, configured to obtain first session information, and obtain second session information associated with the first session information based on a knowledge graph;

a processing unit 402, configured to invoke a service classification model to process the first dialog information and the second dialog information, and determine a target service scenario;

the processing unit 402 is further configured to obtain a target node model corresponding to the target service scene, and call the target node model to process the first session information and the second session information, so as to determine a target node in the target service scene;

an output unit 403, configured to output a target node-associated terminology, where the target node-associated terminology is response information corresponding to the first session information.

In a possible implementation manner, before the obtaining unit 401 is configured to obtain the second session information associated with the first session information based on the knowledge-graph, the obtaining unit 401 is further configured to:

acquiring dialogue data, wherein the dialogue data comprises dialogue information, the attribute of the dialogue information and the incidence relation between the dialogue information;

and constructing the knowledge graph based on the dialog information, the attributes of the dialog information and the incidence relation among the dialog information.

In a possible implementation manner, the processing unit 402 is configured to invoke the service classification model to process the first dialog information and the second dialog information, and determine the target service scenario, including:

extracting the features of the first dialogue information and the second dialogue information to obtain a feature vector;

calling a service classification model to process the feature vector, and determining the prediction probability of the feature vector in each candidate service scene;

and determining a target service scene according to the prediction probability under each candidate service scene.

In a possible implementation manner, the processing unit 402 is configured to perform feature extraction on the first dialog information and the second dialog information to obtain a feature vector, and includes:

determining target dialogue information based on the first dialogue information and the second dialogue information;

performing word segmentation processing on the target dialogue information to obtain segmented target dialogue information, and mapping the segmented target dialogue information to obtain word vectors;

and calling a feature extraction model to extract features of the word vectors to obtain feature vectors.

In one possible implementation, the processing unit 402 is configured to determine the target session information based on the first session information and the second session information, and includes:

respectively acquiring the attention weight of the first dialogue information and the attention weight of the second dialogue information;

the first dialog information and the second dialog information are processed based on the attention weight of the first dialog information and the attention weight of the second dialog information, and the target dialog information is determined.

In a possible implementation, after the processing unit 402 is configured to determine the target node in the target service scenario, the processing unit 402 is further configured to:

acquiring user portrait information of a target user and a product recommendation model corresponding to the user portrait information;

and calling a product recommendation model to process the user portrait information and the target node to obtain target index values of the candidate products under the target node under the evaluation index, and determining the target product according to the target index values of the candidate products.

In a possible implementation, after the processing unit 402 is configured to determine the target product according to the index value of each candidate product, the processing unit 402 is further configured to:

checking the target product, and if the target product passes the checking, carrying out consensus verification on the target product through a consensus node in the block chain network;

and if the consensus verification is passed, packaging the target product into blocks, and writing the blocks into the block chain.

According to another embodiment of the present application, the units in the recommendation device shown in fig. 4 may be respectively or entirely combined into one or several other units to form one or several other units, or some unit(s) may be further split into multiple units with smaller functions to form one or several other units, which may achieve the same operation without affecting the achievement of the technical effect of the embodiment of the present application. The units are divided based on logic functions, and in practical applications, the functions of one unit may be implemented by a plurality of units, or the functions of a plurality of units may be implemented by one unit. In other embodiments of the present application, the speech-based recommendation apparatus may also include other units, and in practical applications, these functions may also be implemented by assistance of other units, and may be implemented by cooperation of multiple units.

According to another embodiment of the present application, the Processing element and the memory element may include a Central Processing Unit (CPU), a random access memory medium (RAM), a read only memory medium (ROM), and the like. A general purpose computing device, such as a computer, runs a computer program (including program code) capable of executing the steps involved in the corresponding method as shown in fig. 2 or fig. 3 to construct a conversational recommendation apparatus as shown in fig. 4 and to implement the conversational recommendation method of the embodiments of the present application. The computer program may be recorded on a computer-readable recording medium, for example, and loaded and executed in the customer service apparatus via the computer-readable recording medium.

In an embodiment of the application, the conversation recommendation device may acquire first conversation information and acquire second conversation information associated with the first conversation information based on a knowledge graph; calling a service classification model to process the first dialogue information and the second dialogue information, determining a target service scene, calling a target node model corresponding to the target service scene to process the first dialogue information and the second dialogue information, determining a target node under the target service scene, and outputting a dialogue associated with the target node, wherein the dialogue associated with the target node is response information corresponding to the first dialogue information. Due to the fact that the business classification model and the node model are combined, the target business scene is determined according to the business classification model, and then the target node is determined according to the target node model under the target business scene, accuracy of the target node can be effectively improved, and accuracy of word operation related to the target node is further improved. In addition, the first dialogue information and the second dialogue information are combined, and the target service scene determined based on the first dialogue information and the second dialogue information is more accurate, so that the target node determined based on the target node model in the target service scene is more accurate, and the accuracy of the dialogue associated with the target node is effectively improved.

Based on the description of the embodiment of the speech technology recommendation method, the embodiment of the application further discloses customer service equipment. Referring to fig. 5, the customer service device includes at least a processor 501, an input interface 502, an output interface 503, and a computer storage medium 504, which may be connected by a bus or other means.

The computer storage medium 504 is a memory device in the customer service device for storing programs and data. It is understood that the computer storage medium 504 herein may include a built-in storage medium of the customer service device, and may also include an extended storage medium supported by the customer service device. The computer storage media 504 provides storage space that stores the operating system of the customer service device. Also stored in this memory space are one or more instructions, which may be one or more computer programs (including program code), suitable for loading and execution by processor 501. Note that the computer storage media herein can be high-speed RAM memory; optionally, the system may further include at least one computer storage medium remote from the processor, where the processor may be referred to as a Central Processing Unit (CPU), and is a core and a control center of the customer service device, and is adapted to implement one or more instructions, and specifically load and execute the one or more instructions to implement the corresponding method flow or function.

In one embodiment, one or more instructions stored in the computer storage medium 504 may be loaded and executed by the processor 501 to implement the steps involved in performing the corresponding method as shown in fig. 2 or fig. 3, and in particular, one or more instructions stored in the computer storage medium 504 may be loaded and executed by the processor 501 to implement the steps of:

In a possible implementation, before the obtaining unit 401 of the processor 501 is configured to obtain the second session information associated with the first session information based on the knowledge-graph, the processor 501 is further configured to:

In a possible implementation, the processor 501 is configured to invoke the service classification model to process the first dialogue information and the second dialogue information, and determine the target service scenario, including:

In a possible implementation manner, the processor 501 is configured to perform feature extraction on the first dialog information and the second dialog information to obtain a feature vector, and includes:

In one possible implementation, the processor 501 is configured to determine the target dialog information based on the first dialog information and the second dialog information, and includes:

In one possible embodiment, after the processor 501 is configured to determine a target node in a target service scenario, the processor 501 is further configured to:

In a possible implementation, after the processor 501 is configured to determine the target product according to the index value of each candidate product, the processor 501 is further configured to:

It should be noted that the present application also provides a computer program product or a computer program, where the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the customer service device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the customer service device performs the steps performed in fig. 2 or fig. 3 of the above-mentioned customer service recommendation method embodiment.

While the foregoing is directed to embodiments of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.

Claims

1. A method for conversational recommendation, the method comprising:

acquiring a target node model corresponding to the target service scene, calling the target node model to process the first dialogue information and the second dialogue information, and determining a target node in the target service scene;

2. The method of claim 1, wherein prior to the obtaining second session information associated with the first session information based on the knowledge-graph, the method further comprises:

and constructing the knowledge graph based on the dialogue information, the attribute of the dialogue information and the incidence relation among the dialogue information.

3. The method of claim 1, wherein said invoking the traffic classification model to process the first dialog information and the second dialog information to determine a target traffic scenario comprises:

extracting features of the first dialogue information and the second dialogue information to obtain a feature vector;

calling the service classification model to process the feature vector, and determining the prediction probability of the feature vector under each candidate service scene;

and determining the target service scene according to the prediction probability under each candidate service scene.

4. The method of claim 3, wherein the feature extracting the first dialog information and the second dialog information to obtain a feature vector comprises:

determining target dialog information based on the first dialog information and the second dialog information;

and calling a feature extraction model to extract features of the word vectors to obtain the feature vectors.

5. The method of claim 4, wherein the determining target dialog information based on the first dialog information and the second dialog information comprises:

respectively acquiring an attention weight of the first dialogue information and an attention weight of the second dialogue information;

and processing the first dialogue information and the second dialogue information based on the attention weight of the first dialogue information and the attention weight of the second dialogue information to determine the target dialogue information.

6. The method of any of claims 1-5, wherein after determining the target node in the target traffic scenario, the method further comprises:

and calling the product recommendation model to process the user portrait information and the target node to obtain target index values of the candidate products under the target node under the evaluation index, and determining the target product according to the target index values of the candidate products.

7. The method of claim 6, wherein after determining the target product based on the indicator values for each candidate product, the method further comprises:

and if the consensus verification is passed, packaging the target product into a block, and writing the block into a block chain.

8. A tactical recommendation apparatus, comprising:

the processing unit is used for calling a service classification model to process the first dialogue information and the second dialogue information and determine a target service scene;

the processing unit is further configured to obtain a target node model corresponding to the target service scene, and call the target node model to process the first dialogue information and the second dialogue information, so as to determine a target node in the target service scene;

and an output unit, configured to output the target node-associated terminology, where the target node-associated terminology is response information corresponding to the first session information.

9. Customer service device, comprising a processor and a memory, wherein the memory is adapted to store a computer program comprising a program, wherein the processor is configured to invoke the program to perform the conversational recommendation method of any of claims 1-7.

10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program comprising program instructions that, when executed by a processor, cause the processor to perform the tactical recommendation method of any of claims 1-7.