Interest identification method, device, equipment and storage medium

Abstract

The present application provides an interest identification method, apparatus, device and storage medium, which relates to the field of big data. The method includes: obtaining voice call data, converting the voice call data into text data; constructing a heterogeneous graph based on the text data, and determining the high-frequency words in the heterogeneous graph by statistical methods; classifying the heterogeneous graph by means of a graph neural network to obtain at least one category corresponding to the text data; performing topic analysis on the documents in each category by means of latent Dirichlet distribution to obtain the topic of each category and the keywords in each topic; determining user interests based on the keywords and high-frequency words in each topic. The method of the present application realizes the rapid and accurate identification of user interests, avoids the inaccuracy of manual identification of user interests and the corresponding waste of personnel, reduces the waste of manpower, reduces the time of user interest analysis, and improves the efficiency of user interest identification.

Description

Technical Field

[0001] The present application relates to the field of big data, and in particular to an interest identification method, apparatus, device, and storage medium. Background Art

[0002] With the development of Internet technology and big data, the usage habits of Internet users have changed from initially searching for content by themselves to relying on content push from service providers. Therefore, accurately understanding the interests of each user can effectively help service providers provide personalized services and improve the user experience.

[0003] In related technologies, service providers can manually search for content that users may be interested in among a large amount of telephone call data between customer service and users. Service providers can also analyze user interest preferences by manually recording keywords.

[0004] However, existing interest identification methods consume a lot of manpower, and manual analysis takes a long time and is inefficient. Summary of the Invention

[0005] The present application provides an interest recognition method, apparatus, device and storage medium to solve the problem that the interest recognition method in the prior art consumes a lot of manpower, has a long manual analysis time and is inefficient.

[0006] In a first aspect, the present application provides an interest identification method, comprising:

[0007] Acquire voice call data, and convert the voice call data into text data;

[0008] Constructing a heterogeneous graph based on the text data, and determining high-frequency words in the heterogeneous graph by statistical methods;

[0009] Classify the heterogeneous graph using a graph neural network to obtain at least one classification corresponding to the text data;

[0010] Through the implicit Dirichlet allocation method, the documents in each category are subjected to topic analysis to obtain the topics of each category and the keywords in each topic.

[0011] User interests are determined based on the keywords and high-frequency words in each topic.

[0012] Here, the present application provides a method for automatically identifying user interests through voice call data. The method constructs the text data corresponding to the voice call data into a heterogeneous graph, and can quickly and accurately classify the text through a graph neural network. In combination with the latent Dirichlet distribution, the topics of interest to the user and the key information corresponding to each topic are identified, thereby achieving rapid and accurate identification of user interests, avoiding the inaccuracy of manual identification of user interests and the corresponding waste of personnel, reducing manpower waste, reducing user interest analysis time, and improving user interest identification efficiency.

[0013] Optionally, constructing a heterogeneous graph based on the text data includes:

[0014] Determine document nodes and word nodes based on the text data;

[0015] A heterogeneous graph is constructed according to the document nodes and word nodes.

[0016] Optionally, the heterogeneous graph includes a feature matrix and an adjacency matrix;

[0017] Accordingly, constructing a heterogeneous graph according to the document nodes and word nodes includes:

[0018] Constructing a feature matrix through the document nodes and the word nodes;

[0019] An adjacency matrix is ​​constructed through the relationships between all nodes in the feature matrix.

[0020] Among them, this application extracts documents and words from text data to form nodes of a heterogeneous graph. Specifically, document nodes and word nodes are used as nodes of a feature matrix, and an adjacency matrix is ​​formed through the relationship between each node. The above matrix combined with a graph neural network can accurately determine the similarity of the content in the document, thereby accurately classifying the document and further improving the accuracy of user interest identification.

[0021] Optionally, the classifying process of the heterogeneous graph by means of a graph neural network to obtain at least one classification corresponding to the text data includes:

[0022] The feature matrix and the adjacency matrix are input into a preset graph variational autoencoder model, and at least one classification corresponding to the text data is determined according to an output result of the preset graph variational autoencoder model.

[0023] Here, this application realizes text classification through graph variational autoencoders. The graph variational autoencoder can better classify documents, thereby identifying the topics that the current user is interested in, as well as some key information in each topic, further improving the accuracy of user interest identification.

[0024] Optionally, before inputting the feature matrix and the adjacency matrix into a preset graph variational autoencoder model and determining at least one classification corresponding to the text data through an output result of the preset graph variational autoencoder model, the method further includes:

[0025] Obtaining a graph variational autoencoder model training sample, wherein the graph variational autoencoder model training sample includes a feature matrix sample, an adjacency matrix sample, and a reconstructed adjacency matrix sample;

[0026] The graph variational autoencoder model training samples are input into the graph variational autoencoder model for training to obtain a preset graph variational autoencoder model.

[0027] Among them, the graph variational autoencoder model is an unsupervised model. Users do not need to label samples during training, which saves training time and steps and further improves the efficiency of user interest recognition.

[0028] Optionally, the topic analysis process is performed on the documents in each category by using implicit Dirichlet distribution to obtain the topic of each category and the keywords in each topic, including:

[0029] For each document in a category, the document is input into a preset latent Dirichlet distribution model;

[0030] The topic of each category and the keywords in each topic are determined through the output results of the latent Dirichlet distribution model.

[0031] Optionally, before inputting the document in each category into the preset latent Dirichlet distribution model, the method further includes:

[0032] Obtaining a latent Dirichlet distribution model training sample, wherein the latent Dirichlet distribution model training sample includes a term sample, a topic sample, and a document sample;

[0033] The implicit Dirichlet distribution model training samples are input into the implicit Dirichlet distribution model for training to obtain a preset implicit Dirichlet distribution model.

[0034] The Latent Dirichlet Allocation model is an unsupervised learning algorithm that doesn't require a manually labeled training set. It only requires a document set and a specified number of topics. This reduces training time and steps, further improving the efficiency of identifying user interests. Users can also specify the number of topics, providing greater flexibility. Another advantage of this model is that it can identify a number of words to describe each topic, resulting in high accuracy in identifying user interests.

[0035] Optionally, before converting the voice call data into text data, the method further includes:

[0036] Preprocessing the voice call data to obtain filtered voice call data;

[0037] Accordingly, converting the voice call data into text data includes:

[0038] The filtered voice call data is converted into text data.

[0039] Here, the present application can pre-process the voice call data in advance, filter out invalid data, reduce memory and processing time, and further improve the efficiency of user interest identification.

[0040] In a second aspect, the present application provides an interest identification device, comprising:

[0041] An acquisition module, configured to acquire voice call data and convert the voice call data into text data;

[0042] A graph construction module, configured to construct a heterogeneous graph based on the text data and determine high-frequency words in the heterogeneous graph by statistical methods;

[0043] A classification module, configured to classify the heterogeneous graph using a graph neural network to obtain at least one classification corresponding to the text data;

[0044] The topic extraction module is used to perform topic analysis on the documents in each category using the implicit Dirichlet allocation method to obtain the topics of each category and the keywords in each topic;

[0045] The interest determination module is used to determine the user's interest based on the keywords and high-frequency words in each topic.

[0046] Optionally, the graph construction module is specifically used to:

[0047] Determine document nodes and word nodes based on the text data;

[0048] A heterogeneous graph is constructed according to the document nodes and word nodes.

[0049] Optionally, the heterogeneous graph includes a feature matrix and an adjacency matrix;

[0050] Accordingly, the graph construction module is further specifically configured to:

[0051] Constructing a feature matrix through the document nodes and the word nodes;

[0052] An adjacency matrix is ​​constructed through the relationships between all nodes in the feature matrix.

[0053] Optionally, the classification module is specifically used to:

[0054] The feature matrix and the adjacency matrix are input into a preset graph variational autoencoder model, and at least one classification corresponding to the text data is determined according to an output result of the preset graph variational autoencoder model.

[0055] Optionally, before the classification module is used to input the feature matrix and the adjacency matrix into a preset graph variational autoencoder model and determine at least one classification corresponding to the text data based on the output result of the preset graph variational autoencoder model, the apparatus further includes a first training module for:

[0056] Obtaining a graph variational autoencoder model training sample, wherein the graph variational autoencoder model training sample includes a feature matrix sample, an adjacency matrix sample, and a reconstructed adjacency matrix sample;

[0057] The graph variational autoencoder model training samples are input into the graph variational autoencoder model for training to obtain a preset graph variational autoencoder model.

[0058] Optionally, the topic extraction module is specifically used to:

[0059] For each document in a category, the document is input into a preset latent Dirichlet distribution model;

[0060] The topic of each category and the keywords in each topic are determined through the output results of the latent Dirichlet distribution model.

[0061] Optionally, before the topic extraction module is used to input the documents in each category into a preset latent Dirichlet distribution model, the apparatus further includes a second training module for:

[0062] Obtaining a latent Dirichlet distribution model training sample, wherein the latent Dirichlet distribution model training sample includes a term sample, a topic sample, and a document sample;

[0063] The implicit Dirichlet distribution model training samples are input into the implicit Dirichlet distribution model for training to obtain a preset implicit Dirichlet distribution model.

[0064] Optionally, before the acquisition module converts the voice call data into text data, the acquisition module is further configured to:

[0065] Preprocessing the voice call data to obtain filtered voice call data;

[0066] Accordingly, the acquisition module is further specifically configured to:

[0067] The filtered voice call data is converted into text data.

[0068] In a third aspect, the present application provides an interest identification device, comprising: at least one processor and a memory;

[0069] The memory stores computer-executable instructions;

[0070] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the interest identification method described in the first aspect and various possible designs of the first aspect.

[0071] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the interest identification method described in the first aspect and various possible designs of the first aspect is implemented.

[0072] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the interest identification method as described in the first aspect and various possible designs of the first aspect.

[0073] The interest identification method, apparatus, device and storage medium provided in this application construct the text data corresponding to the voice call data into a heterogeneous graph, which can quickly and accurately classify the text through a graph neural network, and combine with the latent Dirichlet distribution to identify the topics of interest to the user and the key information corresponding to each topic, thereby achieving rapid and accurate identification of user interests, avoiding the inaccuracy of manual identification of user interests and the corresponding waste of manpower, reducing manpower waste, shortening user interest analysis time, and improving the efficiency of user interest identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0075] Figure 1 A schematic diagram of an interest recognition system architecture provided in an embodiment of the present application;

[0076] Figure 2 A flowchart of an interest identification method provided in an embodiment of the present application;

[0077] Figure 3 A flowchart of another interest identification method provided in an embodiment of the present application;

[0078] Figure 4A flowchart of another interest identification method provided in an embodiment of the present application;

[0079] Figure 5 A schematic diagram of the structure of an interest identification device provided in an embodiment of the present application;

[0080] Figure 6 A schematic diagram of the structure of an interest identification device provided in an embodiment of the present application.

[0081] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0082] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0083] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0084] It should be noted that the interest identification method, device, equipment and storage medium of the present application can be used in the field of big data, and can also be used in any field other than the field of big data. The application field of the interest identification method, device, equipment and storage medium of the present application is not limited.

[0085] The following explains the terms that appear in the examples of this application:

[0086] Graph neural network: refers to the general term for algorithms that use neural networks to learn graph-structured data, extract and discover features and patterns in graph-structured data, and meet the needs of graph learning tasks such as clustering, classification, prediction, segmentation, and generation.

[0087] Topic model: The topic model is used for document modeling, converting documents into numerical vectors, where each dimension of the numerical vector corresponds to a topic.

[0088] Variational Graph Auto-Encoders (VGAEs): Using a known graph, they encode (graph convolution) to learn the distribution of node vector representations. Node vector representations are sampled from this distribution, and then decoded (for link prediction) to reconstruct the graph. Graph variational autoencoders consist of an encoder and a decoder. The inputs are an adjacency matrix and a node feature matrix. The encoder generates a low-dimensional Gaussian distribution of node vectors, and the decoder generates the graph structure (for link prediction).

[0089] Latent Dirichlet Allocation (LDA) is a topic model that can give the topic of each document in a document set in the form of a probability distribution.

[0090] Currently, some of the calls made by service providers are handled by artificial intelligence (AI), so they need to prepare for possible scenarios and scripts in advance. This type of call content cannot record the topics that users are interested in, and a lot of call content will be wasted. If these contents can be fully mined, they can effectively meet the needs of users. In related technologies, service providers can manually search for content that users may be interested in in a large amount of telephone call data between AI or manual customer service and users, and service providers can analyze users' interest preferences by manually recording keywords. However, the interest identification method of the existing technology consumes a lot of manpower, and manual analysis takes a long time and is inefficient.

[0091] In order to solve the above problems, the embodiments of the present application provide an interest identification method, apparatus, device and storage medium, wherein the method constructs the text data corresponding to the voice call data into a heterogeneous graph, and can quickly and accurately classify the text through a graph neural network, and combines the latent Dirichlet distribution to identify the topics of interest to the user and the key information corresponding to each topic, and then determine the user's interest based on the key information.

[0092] Optional, Figure 1 This is a schematic diagram of an interest recognition system architecture provided in an embodiment of the present application. Figure 1 In the above architecture, the above architecture includes at least one of a receiving device 101, a processor 102 and a display device 103.

[0093] It is understood that the structure illustrated in the embodiments of this application does not constitute a specific limitation on the architecture of the interest identification system. In other feasible implementations of this application, the above architecture may include more or fewer components than shown, or combine or split certain components, or arrange the components differently. The specific configuration can be determined based on the actual application scenario and is not limited here. Figure 1 The components shown can be implemented in hardware, software, or a combination of software and hardware.

[0094] In a specific implementation process, the receiving device 101 may be an input / output interface or a communication interface.

[0095] Processor 102 can construct the text data corresponding to the voice call data into a heterogeneous graph, and can quickly and accurately classify the text through a graph neural network. In combination with latent Dirichlet distribution, it can identify the topics of interest to the user and the key information corresponding to each topic, and then determine the user's interest based on the key information.

[0096] The display device 103 can be used to display the above results and the like.

[0097] The display device may also be a touch screen display, which is used to receive user instructions while displaying the above content to achieve interaction with the user.

[0098] It should be understood that the above-mentioned processor can be implemented by the processor reading instructions in the memory and executing the instructions, or it can be implemented by a chip circuit.

[0099] In addition, the network architecture and business scenarios described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Ordinary technicians in this field can know that with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0100] The technical solution of this application is described in detail below with reference to specific embodiments:

[0101] Optionally, Figure 2 A flowchart of an interest identification method provided in an embodiment of the present application. The execution subject of the embodiment of the present application may be Figure 1 The specific execution subject of the processor 102 can be determined according to the actual application scenario. Figure 2 As shown, the method includes the following steps:

[0102] S201: Acquire voice call data and convert the voice call data into text data.

[0103] Optionally, pre-stored historical voice call data can be obtained, which can be stored in a cloud server in the form of cloud storage, or stored in a memory, or voice call data input by a user can be obtained, or voice call data sent by other terminal devices can be received.

[0104] Optionally, if there is directly stored text data, the text data is obtained.

[0105] Optionally, before converting the voice call data into text data, it also includes: preprocessing the voice call data to obtain filtered voice call data; accordingly, converting the voice call data into text data includes: converting the filtered voice call data into text data.

[0106] Here, the embodiment of the present application can pre-process the voice call data in advance, filter out invalid data, reduce memory and processing time, and further improve the efficiency of user interest identification.

[0107] Optionally, data preprocessing can be performed based on the length of the voice call data to filter out voice call data with a call time less than a preset call time threshold. The preset call time threshold can be determined based on actual conditions and is not specifically limited in this embodiment of the application.

[0108] Optionally, the call data may be filtered according to a preset user list, or according to a preset user classification, thereby effectively and accurately identifying the interests of users of different types and interests.

[0109] S202: Construct a heterogeneous graph based on the text data, and determine high-frequency words in the heterogeneous graph through statistical methods.

[0110] Optionally, constructing a heterogeneous graph based on the text data includes: determining document nodes and word nodes based on the text data; and constructing a heterogeneous graph based on the document nodes and word nodes.

[0111] Optionally, the heterogeneous graph includes a feature matrix and an adjacency matrix; accordingly, constructing a heterogeneous graph based on document nodes and word nodes includes: constructing a feature matrix through document nodes and word nodes; and constructing an adjacency matrix through the relationships between all nodes in the feature matrix.

[0112] Among them, the embodiment of the present application extracts documents and words from text data to form nodes of a heterogeneous graph. Specifically, document nodes and word nodes are used as nodes of a feature matrix, and an adjacency matrix is ​​formed through the relationship between each node. The above matrix combined with a graph neural network can accurately determine the similarity of the content in the document, thereby accurately classifying the document and further improving the accuracy of user interest identification.

[0113] Optionally, high-frequency words are determined by term frequency-inverse document frequency (TF-IDF).

[0114] TF-IDF is a statistical method used to assess the importance of a word within a document collection or corpus. The importance of a word increases with its occurrence in a document, but decreases inversely with its frequency in the corpus. Various forms of TF-IDF weighting are often used by search engines as a measure or rating of the relevance of documents to user queries.

[0115] Optionally, the graph structure is represented as follows: The graph structure has two node types: document nodes and word nodes. A word is a unique word in each document. The graph structure has two edge types: an edge between document nodes and an edge between document nodes and word nodes.

[0116] Optionally, the specific heterogeneous graph is represented as a feature matrix X and a corresponding adjacency matrix A. The dimension V of the matrix is ​​the number of document nodes + the number of word nodes. Because the inherent characteristics of each document node and word node are not considered in this paper, a V-dimensional identity matrix can be used for corresponding representation, that is, the feature matrix is ​​the identity matrix. Where i represents the row and j represents the column. The definition of each edge weight Aij in the adjacency matrix is:

[0117] When i and j are document nodes, i=j, then Aij=1, otherwise Aij=0.

[0118] When i and j are word nodes, Aij = PMI(i, j). This part collects co-occurrence statistics by using a fixed-size sliding window for all documents in the corpus. A positive Pointwise Mutual Information (PMI) value indicates that the semantic relevance of the words in the corpus is high, and a negative PMI value indicates that the semantic relevance of the words in the corpus is low or non-existent. This embodiment of the application only establishes edge nodes between word nodes with positive PMI.

[0119] When i is a document node and j is a word node, Aij obtains its value through TF-IDF. Use the TF-IDF algorithm to perform Jieba word segmentation: for the obtained text contents, the shorter ones can be combined into a document, and these documents are represented by d1, d2, d3...dn. TF = the number of times a word appears in di documents / the total number of words in di articles, which means the frequency of a single word appearing in a single document. IDF = the total number of documents / the number of documents in which word W appears. IDF reflects the distinctiveness of word W between documents. If W only appears in one document, it means that W can be used to distinguish the document from other documents, that is, IDF can reflect the uniqueness of W. TF*IDF can obtain the importance of the word.

[0120] S203: Classify the heterogeneous graph using a graph neural network to obtain at least one classification corresponding to the text data.

[0121] Optionally, the graph neural network is any model that can implement classification functions, such as a graph variational autoencoder or a graph attention network.

[0122] S204: Perform topic analysis on the documents in each category by using implicit Dirichlet allocation to obtain the topic of each category and the keywords in each topic.

[0123] S205: Determine user interests based on the keywords and high-frequency words in each topic.

[0124] Optionally, keywords and high-frequency words in each topic are compared to determine user interests.

[0125] Optionally, if the high-frequency words include a keyword, the keyword is determined to be the user's interest.

[0126] Optionally, after determining the user's interests, the embodiment of the present application may directly store the user's interests in the AI ​​database, or generate a question-and-answer scenario based on the user's interests, so that the AI ​​can communicate with the user based on the user's interests, thereby improving the user experience.

[0127] Optionally, after determining the user's interests, relevant information is pushed to the user based on the user's interests.

[0128] An embodiment of the present application provides a method for automatically identifying user interests through voice call data. The method constructs the text data corresponding to the voice call data into a heterogeneous graph, and can quickly and accurately classify the text through a graph neural network. In combination with the latent Dirichlet distribution, the method identifies the topics of interest to the user and the key information corresponding to each topic, thereby achieving rapid and accurate identification of user interests, avoiding the inaccuracy of manual identification of user interests and the corresponding waste of manpower, reducing manpower waste, shortening user interest analysis time, and improving the efficiency of user interest identification.

[0129] Optionally, the embodiment of the present application classifies text data by means of a graph variational autoencoder. Accordingly, Figure 3 A flowchart of another interest identification method provided in an embodiment of the present application is shown as follows: Figure 3 As shown, the method includes:

[0130] S301: Acquire voice call data and convert the voice call data into text data.

[0131] S302: Construct a heterogeneous graph based on the text data, and determine high-frequency words in the heterogeneous graph through statistical methods.

[0132] The implementation of step S301 and step S302 is similar to that of step S201 and step S202, and will not be described in detail here.

[0133] S303: Input the feature matrix and the adjacency matrix into a preset graph variational autoencoder model, and determine at least one category corresponding to the text data through an output result of the preset graph variational autoencoder model.

[0134] Optionally, before inputting the feature matrix and the adjacency matrix into a preset graph variational autoencoder model and determining at least one category corresponding to the text data through the output result of the preset graph variational autoencoder model, it also includes: obtaining graph variational autoencoder model training samples, wherein the graph variational autoencoder model training samples include feature matrix samples, adjacency matrix samples and reconstructed adjacency matrix samples; inputting the graph variational autoencoder model training samples into the graph variational autoencoder model for training to obtain the preset graph variational autoencoder model.

[0135] Among them, the graph variational autoencoder model is an unsupervised model. Users do not need to label samples during training, which saves training time and steps and further improves the efficiency of user interest recognition.

[0136] In one possible implementation, a graph neural network is used to further process the above-mentioned feature matrix. The embodiment of the present application uses a graph variational autoencoder to further process the above-mentioned matrix. The graph variational autoencoder uses a two-layer graph convolutional neural network as the encoding layer. The input content is a feature matrix X of the graph. Because each node does not have its own special features to be expressed, a unit matrix is ​​used for relevant representation. The other input content is the adjacency matrix A obtained in the previous step. By using the graph variational autoencoder, the adjacency matrix obtained above can be reconstructed. At this time, the similarity of the closer content in the matrix is ​​also higher. In this way, text classification can be performed.

[0137] S304: Perform topic analysis on the documents in each category by using implicit Dirichlet allocation to obtain the topic of each category and the keywords in each topic.

[0138] S305: Determine user interests based on the keywords and high-frequency words in each topic.

[0139] The implementation of step S304 and step S305 is similar to that of step S204 and step S205, and will not be described in detail here.

[0140] Here, the embodiment of the present application realizes text classification through a graph variational autoencoder. The graph variational autoencoder can better classify documents, thereby identifying the topics that the current user is interested in, as well as some key information in each topic, further improving the accuracy of user interest identification.

[0141] Optionally, the embodiment of the present application classifies text data by presetting a latent Dirichlet distribution model. Accordingly, Figure 4 A flowchart of another interest identification method provided in an embodiment of the present application is shown as follows: Figure 4 As shown, the method includes:

[0142] S401: Acquire voice call data and convert the voice call data into text data.

[0143] S402: Construct a heterogeneous graph based on the text data, and determine high-frequency words in the heterogeneous graph through statistical methods.

[0144] S403: Classify the heterogeneous graph using a graph neural network to obtain at least one classification corresponding to the text data.

[0145] The implementation of steps S401-S403 is similar to that of steps S201-S203, and will not be described in detail here.

[0146] S404: For each document in each category, input the document into a preset latent Dirichlet distribution model; determine the topic of each category and the keywords in each topic through the output result of the latent Dirichlet distribution model.

[0147] S405: Determine user interests based on the keywords and high-frequency words in each topic.

[0148] Optionally, before inputting the documents in each category into a preset latent Dirichlet distribution model, the method further includes: obtaining latent Dirichlet distribution model training samples, wherein the latent Dirichlet distribution model training samples include term samples, topic samples, and document samples; and inputting the latent Dirichlet distribution model training samples into the latent Dirichlet distribution model for training to obtain a preset latent Dirichlet distribution model.

[0149] In one possible implementation, topic analysis is performed using the implicit Dirichlet distribution. This method is an unsupervised machine learning method that pre-sets a three-layer structure of implicit Dirichlet distribution models: terms, topics, and documents, to obtain potential topic distribution information in large-scale documents. For each document, LDA uses a bag-of-words model to represent the text as a word frequency vector. The position and order of the terms in the model are irrelevant to the final topic distribution. LDA is obtained by adding a Bayesian framework to the Probabilistic Latent Semantic Analysis (PLSA) model and incorporating the influence of the Dirichlet prior distribution. The execution process is as follows:

[0150] For each document, extract a topic from the topic distribution;

[0151] Extract a topic from the word distribution corresponding to the above extracted topics;

[0152] Repeat the above process until every word in the document is traversed. In the specific implementation process, the number of topics to be generated can be set.

[0153] The Latent Dirichlet Allocation model is an unsupervised learning algorithm that doesn't require a manually labeled training set. It only requires a document set and a specified number of topics. This reduces training time and steps, further improving the efficiency of identifying user interests. Users can also specify the number of topics, providing greater flexibility. Another advantage of this model is that it can identify a number of words to describe each topic, resulting in high accuracy in identifying user interests.

[0154] The above method for processing text is relatively efficient and rapid. It allows for quicker understanding of recent user demands, enabling timely preparation and better resolution of user issues. While voice storage is currently expensive, it also holds significant value. By exploring users' potential interests through the above method, this data can be fully utilized, reducing storage costs. Using a graph variational autoencoder, it is possible to effectively identify topics of interest to the user and key information within each topic. This avoids the inaccuracies and associated waste of human resources associated with manual identification of user interests.

[0155] Figure 5 This is a structural diagram of an interest identification device provided in an embodiment of the present application, such as Figure 5 As shown, the apparatus of the embodiment of the present application includes: an acquisition module 501, a graph construction module 502, a classification module 503, a topic extraction module 504 and an interest determination module 505. The interest identification device here can be the above-mentioned processor itself, or a chip or integrated circuit that implements the function of the processor. It should be noted here that the division of the acquisition module 501, the graph construction module 502, the classification module 503, the topic extraction module 504 and the interest determination module 505 is only a division of logical functions. Physically, the two can be integrated or independent.

[0156] The acquisition module is used to acquire voice call data and convert the voice call data into text data;

[0157] The graph construction module is used to construct a heterogeneous graph based on text data and determine the high-frequency words in the heterogeneous graph through statistical methods;

[0158] A classification module is used to classify the heterogeneous graph using a graph neural network to obtain at least one category corresponding to the text data;

[0159] The topic extraction module is used to perform topic analysis on the documents in each category using the implicit Dirichlet allocation method to obtain the topics of each category and the keywords in each topic;

[0160] The interest determination module is used to determine user interests based on the keywords and high-frequency words in each topic.

[0161] Optionally, the graph construction module is specifically used to:

[0162] According to the text data, determine the document node and word node;

[0163] Construct a heterogeneous graph based on document nodes and word nodes.

[0164] Optionally, the heterogeneous graph includes a feature matrix and an adjacency matrix;

[0165] Accordingly, the graph construction module is further specifically used to:

[0166] Construct a feature matrix through document nodes and word nodes;

[0167] The adjacency matrix is ​​constructed through the relationship between all nodes in the feature matrix.

[0168] Optionally, the classification module is specifically used to:

[0169] The feature matrix and the adjacency matrix are input into a preset graph variational autoencoder model, and at least one classification corresponding to the text data is determined based on the output result of the preset graph variational autoencoder model.

[0170] Optionally, before the classification module is used to input the feature matrix and the adjacency matrix into a preset graph variational autoencoder model and determine at least one classification corresponding to the text data based on the output result of the preset graph variational autoencoder model, the apparatus further includes a first training module for:

[0171] Obtaining graph variational autoencoder model training samples, where the graph variational autoencoder model training samples include feature matrix samples, adjacency matrix samples, and reconstructed adjacency matrix samples;

[0172] The graph variational autoencoder model training samples are input into the graph variational autoencoder model for training to obtain a preset graph variational autoencoder model.

[0173] Optionally, the topic extraction module is specifically used to:

[0174] For each document in each category, the document is input into the preset latent Dirichlet distribution model;

[0175] The output results of the latent Dirichlet allocation model are used to determine the topics of each category and the keywords in each topic.

[0176] Optionally, before the topic extraction module is used to input the documents in each category into the preset latent Dirichlet distribution model, the apparatus further includes a second training module for:

[0177] Obtaining latent Dirichlet distribution model training samples, wherein the latent Dirichlet distribution model training samples include term samples, topic samples, and document samples;

[0178] The latent Dirichlet distribution model training samples are input into the latent Dirichlet distribution model for training to obtain a preset latent Dirichlet distribution model.

[0179] Optionally, before the acquisition module converts the voice call data into text data, the acquisition module is further configured to:

[0180] Preprocessing the voice call data to obtain filtered voice call data;

[0181] Accordingly, the acquisition module is further specifically used for:

[0182] Convert the filtered voice call data into text data.

[0183] refer to Figure 6 , which shows a schematic structural diagram of an interest identification device 600 suitable for implementing an embodiment of the present disclosure. The interest identification device 600 can be a terminal device or a server. The terminal device can include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, personal digital assistants (PDAs), tablet computers (Portable Android Devices, PADs), portable multimedia players (PMPs), in-vehicle terminals (such as in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The illustrated interest identification device is merely an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0184] like Figure 6 As shown, the interest identification device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. Various programs and data required for the operation of the interest identification device 600 are also stored in the RAM 603. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0185] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the interest identification device 600 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6The interest identification apparatus 600 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0186] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network via the communication device 609, or installed from the storage device 608, or installed from the ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of the embodiment of the present disclosure are performed.

[0187] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0188] The computer-readable medium may be included in the interest identification device, or may exist independently without being incorporated into the interest identification device.

[0189] The computer-readable medium carries one or more programs. When the one or more programs are executed by the interest identification device, the interest identification device executes the method shown in the above embodiment.

[0190] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0191] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0192] The units involved in the embodiments described in this disclosure may be implemented in software or hardware. In some cases, the name of a unit does not limit the unit itself. For example, the first acquisition unit may also be described as a "unit for acquiring at least two Internet Protocol addresses."

[0193] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0194] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0195] The interest identification device of the embodiment of the present application can be used to execute the technical solutions in the above-mentioned method embodiments of the present application. Its implementation principles and technical effects are similar and will not be repeated here.

[0196] An embodiment of the present application further provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by a processor, they are used to implement any of the above-mentioned interest identification methods.

[0197] An embodiment of the present application further provides a computer program product, including a computer program, which, when executed by a processor, is used to implement any of the above-mentioned interest identification methods.

[0198] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0199] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0200] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the following claims.

[0201] It should be understood that the present application is not limited to the exact structure described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for identifying interests, characterized in that: include: Acquire voice call data, and convert the voice call data into text data; Constructing a heterogeneous graph based on the text data, and determining high-frequency words in the heterogeneous graph by statistical methods; Classify the heterogeneous graph using a graph neural network to obtain at least one classification corresponding to the text data; Through the implicit Dirichlet allocation method, the documents in each category are subjected to topic analysis to obtain the topics of each category and the keywords in each topic. User interests are determined based on the keywords and high-frequency words in each topic.

2. The method according to claim 1, characterized in that The step of constructing a heterogeneous graph based on the text data includes: Determine document nodes and word nodes based on the text data; A heterogeneous graph is constructed according to the document nodes and word nodes.

3. The method according to claim 2, characterized in that The heterogeneous graph includes a feature matrix and an adjacency matrix; Accordingly, constructing a heterogeneous graph according to the document nodes and word nodes includes: Constructing a feature matrix through the document nodes and the word nodes; An adjacency matrix is ​​constructed through the relationships between all nodes in the feature matrix.

4. The method according to claim 3, characterized in that The classifying process of the heterogeneous graph by means of a graph neural network to obtain at least one classification corresponding to the text data includes: The feature matrix and the adjacency matrix are input into a preset graph variational autoencoder model, and at least one classification corresponding to the text data is determined according to an output result of the preset graph variational autoencoder model.

5. The method according to claim 4, characterized in that Before inputting the feature matrix and the adjacency matrix into a preset graph variational autoencoder model and determining at least one classification corresponding to the text data according to an output result of the preset graph variational autoencoder model, the method further includes: Obtaining a graph variational autoencoder model training sample, wherein the graph variational autoencoder model training sample includes a feature matrix sample, an adjacency matrix sample, and a reconstructed adjacency matrix sample; The graph variational autoencoder model training samples are input into the graph variational autoencoder model for training to obtain a preset graph variational autoencoder model.

6. The method according to any one of claims 1 to 4, characterized in that The document in each category is subjected to topic analysis by implicit Dirichlet distribution to obtain the topic of each category and the keywords in each topic, including: For each document in a category, the document is input into a preset latent Dirichlet distribution model; The topic of each category and the keywords in each topic are determined through the output results of the latent Dirichlet distribution model.

7. The method according to claim 6, characterized in that Before inputting the document into the preset latent Dirichlet distribution model for each document in each category, the method further includes: Obtaining a latent Dirichlet distribution model training sample, wherein the latent Dirichlet distribution model training sample includes a term sample, a topic sample, and a document sample; The implicit Dirichlet distribution model training samples are input into the implicit Dirichlet distribution model for training to obtain a preset implicit Dirichlet distribution model.

8. The method according to any one of claims 1 to 4, characterized in that Before converting the voice call data into text data, the method further includes: Preprocessing the voice call data to obtain filtered voice call data; Accordingly, converting the voice call data into text data includes: The filtered voice call data is converted into text data.

9. An interest recognition device, characterized in that: include: An acquisition module, configured to acquire voice call data and convert the voice call data into text data; A graph construction module, configured to construct a heterogeneous graph based on the text data and determine high-frequency words in the heterogeneous graph by statistical methods; A classification module, configured to classify the heterogeneous graph using a graph neural network to obtain at least one classification corresponding to the text data; The topic extraction module is used to perform topic analysis on the documents in each category using the implicit Dirichlet allocation method to obtain the topics of each category and the keywords in each topic; The interest determination module is used to determine the user's interest based on the keywords and high-frequency words in each topic.

10. An interest recognition device, characterized in that: include: a processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 8 when executed by a processor.

12. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images