Theme recommendation method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a theme recommendation method and device, electronic equipment and a storage medium, and belongs to the technical field of computers. The method comprises the steps that the popularity degree of candidate topics is calculated according to the project number corresponding to the candidate topics; calculating node weights of the candidate topics in the knowledge graph according to the popularity; calculating correlation scores of the candidate topics and the items according to the correlation probabilities of the candidate topics and the items and the node weights; and performing descending sorting on the candidate theme list according to the relevance score, and returning a plurality of topics as the content of the target recommendation list. According to the embodiment of the invention, the accuracy of topic recommendation during retrieval can be improved, and the noise of topic retrieval is reduced.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a topic recommendation method and device, an electronic device, and a storage medium. Background Art

[0002] Keywords used in user searches are a form of metadata, generally referred to as topics or tags, which are annotated based on the core concepts of the entity to match the key features in the search results. Topics can help improve the visibility of related entities to user queries.

[0003] Due to phenomena such as label explosion and topic entanglement, the performance of the prediction model in the recommendation system is negatively affected, resulting in reduced accuracy of topic recommendations during retrieval and the presence of noise in the retrieved topics.

[0004] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the invention

[0005] The main purpose of the embodiments of the present application is to propose a topic recommendation method and device, an electronic device and a storage medium, aiming to improve the accuracy of topic recommendation during retrieval and reduce the noise of retrieval topics.

[0006] To achieve the above object, an embodiment of the present application provides a topic recommendation method, which includes the following steps:

[0007] Calculating the popularity of a candidate topic according to the number of items corresponding to the candidate topic, the candidate topic belonging to the candidate topic list;

[0008] Calculate the node weight of the candidate topic in the knowledge graph according to the popularity;

[0009] Calculate the relevance score between the candidate topic and the project according to the relevance probability between the candidate topic and the project and the node weight;

[0010] The candidate topic list is sorted in descending order according to the relevance score, and the first several topics are returned as the contents of the target recommendation list.

[0011] In some embodiments, the method further includes the step of obtaining a candidate topic list, wherein obtaining the candidate topic list includes:

[0012] If there is a search topic record of the user, generating the candidate topic list according to the search topic record;

[0013] If there is an initial topic in the repository, the candidate topic list is generated according to the initial topic.

[0014] In some embodiments, the method further includes the step of obtaining a candidate topic list, wherein obtaining the candidate topic list includes:

[0015] If the initial topic does not exist in the repository and the search topic record of the user does not exist, acquiring text data of the repository;

[0016] Converting the text data into a term frequency-inverse document frequency vector;

[0017] Training a topic set recommendation model by using the term frequency-inverse document frequency vector;

[0018] The target project is input into the topic set recommendation model to obtain a candidate topic list.

[0019] In some embodiments, the step of calculating the node weight of the candidate topic in the knowledge graph according to the popularity includes:

[0020] Calculating the node degree of the candidate topic in the knowledge graph;

[0021] The node weight of the candidate topic in the knowledge graph is calculated according to the popularity and the node degree.

[0022] In some embodiments, the method further includes a step of constructing a knowledge graph, wherein constructing the knowledge graph includes:

[0023] Initialize the knowledge graph using topics, relationship types, and relationships between topics;

[0024] Updating the knowledge graph by triangulation;

[0025] Acquire knowledge about topics, relationship types, and relationships between topics, and expand the knowledge graph through the knowledge;

[0026] The knowledge graph is evaluated through a labeling task.

[0027] In some embodiments, the calculating the relevance score between the candidate topic and the project according to the relevance probability between the candidate topic and the project and the node weight includes:

[0028] Obtain the final fuzzy representation of the user and the final fuzzy representation of the item by using a hierarchical fusion fuzzy deep network;

[0029] The relevance score between the candidate topic and the project is calculated according to the final fuzzy representation of the user, the final fuzzy representation of the project, the relevance probability between the candidate topic and the project, and the node weight.

[0030] In some embodiments, the calculating the node weight of the candidate topic in the knowledge graph according to the popularity and the node degree includes:

[0031] Get the proportional coefficient of popularity and the proportional coefficient of node degree;

[0032] The node weight of the candidate topic in the knowledge graph is calculated according to the popularity, the node degree, the proportional coefficient of the popularity and the proportional coefficient of the node degree. The expression of the node weight is as follows:

[0033] W t =α×P t +β×D t ,

[0034] α+β=1

[0035] Among them, W t is the node weight of the candidate topic in the knowledge graph, α is the proportional coefficient of the popularity, P t is the popularity, β is the proportional coefficient of the node degree, D t is the node degree.

[0036] To achieve the above objective, another aspect of the present application provides a topic recommendation device, the device comprising:

[0037] A first calculation module, configured to calculate the popularity of a candidate topic according to the number of items corresponding to the candidate topic, the candidate topic belonging to the candidate topic list;

[0038] A second calculation module is used to calculate the node weight of the candidate topic in the knowledge graph according to the popularity;

[0039] A third calculation module, used to calculate the relevance score between the candidate topic and the project according to the relevance probability between the candidate topic and the project and the node weight;

[0040] The recommendation module is used to sort the candidate topic list in descending order according to the relevance score and return the first several topics as the content of the target recommendation list.

[0041] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application provides an electronic device, the electronic device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the above-mentioned method when executing the computer program.

[0042] To achieve the above objective, another aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0043] The embodiments of the present application include at least the following beneficial effects: the present application provides a topic recommendation method and device, an electronic device and a storage medium, which calculates the popularity of the candidate topic according to the number of projects corresponding to the candidate topic, reflects the popularity of the candidate topic, and helps determine the priority of the recommendation; calculates the node weight of the candidate topic in the knowledge graph according to the popularity, which helps to highlight important topics and thus reduce the influence of irrelevant or low-value topics; calculates the correlation score between the candidate topic and the project according to the correlation probability of the candidate topic and the project and the node weight, which is conducive to evaluating the correlation between each candidate topic and the project, so that the recommendation is more in line with the actual needs of the user; sorts the candidate topic list in descending order according to the correlation score, and returns the first several topics as the content of the target recommendation list, focusing on the actual needs of the user, reducing unnecessary information interference, improving the accuracy of topic recommendations during retrieval, and reducing the noise of retrieval topics. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of a topic recommendation method provided by an embodiment of the present application;

[0045] Figure 2 It is a flowchart of the step of obtaining a candidate topic list of the topic recommendation method provided in an embodiment of the present application;

[0046] Figure 3 yes Figure 2 Flow chart of step S203 in FIG.

[0047] Figure 4 yes Figure 1 Flow chart of step S102 in FIG.

[0048] Figure 5 It is a flowchart of the steps of constructing a knowledge graph of the topic recommendation method provided in the embodiment of the present application;

[0049] Figure 6 yes Figure 1 Flow chart of step S103 in FIG.

[0050] Figure 7 This is an optional flow chart of the topic recommendation method provided in the embodiment of the present application applied to a topic recommendation system;

[0051] Figure 8 It is a schematic diagram of the knowledge graph construction process provided in the embodiment of the present application;

[0052] Fig. 9 is a structural schematic diagram of a topic recommendation device provided in an embodiment of the present application;

[0053] Fig.10 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. They are only examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the attached claims.

[0055] It is understood that the terms "first", "second", etc. used in this application can be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another concept. For example, without departing from the scope of the embodiment of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determination".

[0056] The terms "at least one", "multiple", "each", "any", etc. used in this application, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0058] Before describing the embodiments of the present application in detail, some nouns and terms involved in the embodiments of the present application are first described. The nouns and terms involved in the embodiments of the present application are subject to the following explanations.

[0059] 1) Knowledge Graph (KG) is a structured knowledge representation method. Knowledge graph stores the relationship between entities in the form of a graph, usually including nodes (representing entities) and edges (representing the relationship between entities).

[0060] 2) Knowledge Graph Recommendation (KGRec) is a recommendation system model that combines knowledge graphs. The core idea of ​​the KGRec model is to use the rich information in the knowledge graph to enhance the traditional recommendation algorithm, making the recommendation more accurate, personalized and explanatory.

[0061] Keywords used in user searches are a form of metadata, generally referred to as topics or tags, which are annotated based on the core concepts of the entity to match the key features in the search results. Topics can help improve the visibility of related entities to user queries.

[0062] Due to phenomena such as label explosion and topic entanglement, the performance of the prediction model in the recommendation system is negatively affected, resulting in reduced accuracy of topic recommendations during retrieval and the presence of noise in the retrieved topics.

[0063] In summary, the technical problems existing in the relevant technologies need to be improved.

[0064] In view of this, a topic recommendation method, apparatus, device and medium are provided in an embodiment of the present application. The scheme calculates the popularity of candidate topics according to the number of projects corresponding to the candidate topics, reflects the popularity of the candidate topics, and helps determine the priority of recommendation; calculates the node weight of the candidate topic in the knowledge graph according to the popularity, which helps to highlight important topics, thereby reducing the influence of irrelevant or low-value topics; calculates the correlation score between the candidate topic and the project according to the correlation probability of the candidate topic and the project and the node weight, which is conducive to evaluating the correlation between each candidate topic and the project, so that the recommendation is more in line with the actual needs of the user; sorts the candidate topic list in descending order according to the correlation score, and returns the first several topics as the content of the target recommendation list, focusing on the actual needs of the user, reducing unnecessary information interference, improving the accuracy of topic recommendation during retrieval, and reducing the noise of retrieval topics.

[0065] The topic recommendation method provided in the embodiment of the present application relates to the field of computer technology. The topic recommendation method provided in the embodiment of the present application can be applied to a terminal, can be applied to a server, or can be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and a car terminal, etc., but is not limited to this; the server side can be configured as an independent physical server, or a server cluster or distributed system composed of multiple physical servers, and can also be configured to provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and cloud servers for basic cloud computing services such as big data and artificial intelligence platforms. The server can also be a node server in a blockchain network; the software can be an application that implements the topic recommendation method, etc., but is not limited to the above forms.

[0066] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0067] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0068] Figure 1 is an optional flowchart of the topic recommendation method provided in the embodiment of the present application. Figure 1The method may include but is not limited to steps S101 to S104.

[0069] Step S101, calculating the popularity of the candidate topic according to the number of items corresponding to the candidate topic.

[0070] Specifically, the candidate topics belong to a candidate topic list, and the repository of the platform includes several projects, each of which includes several topics.

[0071] In some embodiments, the candidate topic list can be obtained from the search topic records of the three channels of search box, general area and industry area, or from the initial topics in the repository. Optionally, the general area can use a general search channel, and the industry area can use a search channel for a specific industry.

[0072] In some embodiments, each project includes a set of topics T = {t1, t2, ..., t m}, where T is the subject group, t1, t2, …, t m is the topic and m is the number of topics assigned to the repository.

[0073] It can be understood that a project includes a subject, that is, a subject corresponds to a project.

[0074] Optionally, count each candidate topic and count the number of projects corresponding to it (i.e., count the number of projects containing the candidate topic). This can be achieved using database queries, data frame operations (such as using the Pandas library), or SQL aggregate functions (such as COUNT).

[0075] In some embodiments, the expression of the node weight of the popularity of the candidate topic is as follows:

[0076]

[0077] Among them, P t is the popularity of the candidate topic, t is the candidate topic, n t is the number of items in the platform tagged with topic t.

[0078] In this embodiment, the popularity of the candidate topics is calculated according to the number of items corresponding to the candidate topics, which can intuitively reflect the popularity of each candidate topic, help determine the priority of recommendation, and prepare for the subsequent calculation of node weights.

[0079] Step S102, calculating the node weight of the candidate topic in the knowledge graph according to its popularity.

[0080] Specifically, the knowledge graph includes topics, relationship types, and relationships between topics.

[0081] In some embodiments, the node degree of the candidate topic in the knowledge graph is calculated; and the node weight of the candidate topic in the knowledge graph is calculated based on the popularity and the node degree.

[0082] Optionally, the knowledge graph is annotated based on the popularity of the topic in the platform and the node weight of the topic node in the knowledge graph.

[0083] It is understandable that the node weight of a candidate topic in the knowledge graph is related to its popularity.

[0084] Furthermore, the node weight of the candidate topic in the knowledge graph is also related to the node degree, where the node degree refers to the number of all edges directly connected to the candidate topic node (i.e., the total degree of the candidate topic).

[0085] In some embodiments, the node weight may be affected by a proportionality coefficient of popularity to a proportionality coefficient of node degree.

[0086] For example, if the proportional coefficient of popularity is greater than the proportional coefficient of node degree, the recommendation results will mainly rely on a set of the most popular topics, that is, on topics that have not been frequently searched by the platform.

[0087] In this embodiment, the node weights of candidate topics in the knowledge graph are calculated based on popularity, so that in the subsequent relevance score calculation, the impact of popularity on the candidate topics can be more reasonably reflected in the node weights, which helps to highlight important topics and thus reduce the impact of irrelevant or low-value topics.

[0088] Step S103, calculating the relevance score between the candidate topic and the project according to the relevance probability between the candidate topic and the project and the node weight.

[0089] In some embodiments, a final fuzzy representation of the user and a final fuzzy representation of the project of a hierarchical fusion fuzzy deep network are obtained; and a relevance score between the candidate topic and the project is calculated based on the final fuzzy representation of the user, the final fuzzy representation of the project, the relevance probability between the candidate topic and the project, and the node weights.

[0090] Optionally, a text analysis technique may be used to calculate the probability of correlation between candidate topics and items, or a classifier may be used to calculate the probability of correlation between topics and items. The text analysis technique may use TF-IDF, cosine similarity, etc., and the classifier may use a naive Bayes text classifier, a multi-class multi-label Logit model classifier, etc.

[0091] In some embodiments, the calculation formula of the relevance score is as follows:

[0092]

[0093]

[0094] in, is the relevance score, is the probability of the candidate topic being associated with the project, W t N(I) represents the set of topics that are directly adjacent to at least one topic in the initial set of topics of the project.

[0095] In this embodiment, the correlation score between the candidate topic and the project is calculated based on the correlation probability between the candidate topic and the project and the node weight, which is conducive to accurately evaluating the relevance of each candidate topic to the project, making the recommendation more in line with the actual needs of the user and enhancing the personalization and accuracy of the system recommendation.

[0096] Step S104, sorting the candidate topic list in descending order according to the relevance score, and returning the first several topics as the contents of the target recommendation list.

[0097] In some embodiments, the candidate topic list is sorted in descending order according to the relevance scores using a sorting algorithm such as quick sort or merge sort.

[0098] The present application does not limit the length of the returned target recommendation list. Optionally, the target recommendation list length can be determined based on a preset recommendation list length, or candidate topics whose relevance scores exceed a preset threshold can be added to the target recommendation list to obtain the target recommendation list length.

[0099] In this embodiment, the candidate topic list is sorted in descending order according to the relevance score, and the first several topics are returned as the content of the target recommendation list, focusing on the actual needs of the user, reducing unnecessary information interference, improving the accuracy of topic recommendations during retrieval, and reducing the noise of retrieval topics.

[0100] Steps S101 to S104 shown in the embodiment of the present application calculate the popularity of the candidate topic according to the number of projects corresponding to the candidate topic, thereby reflecting the popularity of the candidate topic and helping to determine the priority of the recommendation; calculating the node weight of the candidate topic in the knowledge graph according to the popularity helps to highlight important topics, thereby reducing the influence of irrelevant or low-value topics; calculating the correlation score between the candidate topic and the project according to the correlation probability of the candidate topic and the project and the node weight is conducive to evaluating the correlation between each candidate topic and the project, so that the recommendation is more in line with the actual needs of the user; sorting the candidate topic list in descending order according to the correlation score, and returning the first several topics as the content of the target recommendation list, focusing on the actual needs of the user, reducing unnecessary information interference, improving the accuracy of topic recommendations during retrieval, and reducing the noise of retrieval topics.

[0101] See also Figure 2In some embodiments, the topic recommendation method provided in the embodiment of the present application further includes the step of obtaining a candidate topic list. The step of obtaining a candidate topic list may include but is not limited to steps S201 to S203:

[0102] Step S201: if there is a search topic record of the user, a candidate topic list is generated according to the search topic record.

[0103] In step S201 of some embodiments, if there are search subject records of the user, the user search subject records are obtained, and invalid or irrelevant search subject records are removed, such as empty searches, repeated searches, spelling errors, etc.

[0104] Among them, search topic records can be obtained through three channels: search box, general area and industry area.

[0105] Optionally, keywords are extracted from the user's search topic records, the frequency of occurrence of each keyword in the user's search topic records is calculated, and a candidate topic list is generated according to the keyword frequency.

[0106] In this embodiment, if there is a search topic record of the user, a candidate topic list is generated according to the search topic record, so that the recommendation is more personalized and the relevance of the recommendation result is improved.

[0107] Step S202: if there is an initial topic in the repository, a candidate topic list is generated according to the initial topic.

[0108] In step S202 of some embodiments, if there is an initial topic in the repository, the initial topic is obtained, the relevance score of each topic is calculated, and a candidate topic list is generated according to the initial topic and its relevance score.

[0109] For example, the initial topic is abc123+- / , and by calculating the relevance scores of each topic, the candidate topic list is obtained as 123+- / abc. .

[0110] It is to be understood that the initial topic may include a single topic or a collection of topics.

[0111] Step S203: if the initial topic does not exist in the repository and there is no search topic record of the user, the text data of the repository is obtained, and a candidate topic list is generated according to the text data.

[0112] In step S203 of some embodiments, if the initial topic does not exist in the repository and there is no user's search topic record, the text data of the project in the repository is obtained; the text data is converted into a term frequency-inverse document frequency vector; the topic set recommendation model is trained through the term frequency-inverse document frequency vector; the target project is input into the topic set recommendation model to obtain a candidate topic list.

[0113] See also Figure 3 In some embodiments, step S203 may include but is not limited to steps S301 to S304:

[0114] Step S301: if there is no initial topic in the repository and there is no search topic record of the user, the text data of the repository is obtained.

[0115] In step S301 of some embodiments, if the initial topic does not exist in the repository and there is no search topic record of the user, the text data is connected.

[0116] Step S302: convert the text data into a term frequency-inverse document frequency vector.

[0117] Specifically, the term frequency-inverse document frequency vector is used to train the classifier of the topic set recommendation model.

[0118] In step S302 of some embodiments, the text data is converted into respective TF-IDF vectors.

[0119] Step S303: training a topic set recommendation model through the word frequency-inverse document frequency vector.

[0120] Optionally, the topic set recommendation model includes a multinomial naive Bayes (MNB) text classifier and a multi-class multi-label Logit model (LR) classifier.

[0121] In step S303 of some embodiments, a classifier of a topic set recommendation model is trained using text data and designated topics.

[0122] Step S304: input the target item into the topic set recommendation model to obtain a candidate topic list.

[0123] In step S304 of some embodiments, text data of the target project is obtained, and the text data is used as input to a multinomial naive Bayes text classifier to predict the correlation probability between the topic and the project, and the first several topics with the highest correlation probability are used as a candidate topic list.

[0124] See also Figure 4 In some embodiments, step S102 may include but is not limited to steps S401 to S402:

[0125] Step S401, calculate the node degree of the candidate topic in the knowledge graph.

[0126] Specifically, node weight is related to popularity and node degree, which refers to the number of edges directly connected to the node in the knowledge graph.

[0127] In step S401 of some embodiments, the node degree of the candidate topic in the knowledge graph is calculated, and the expression of the node weight of the node degree of the candidate topic is as follows:

[0128]

[0129] Among them, D t is the node degree of the candidate topic, t is the candidate topic, d t is the total degree of candidate topic t in the knowledge graph.

[0130] Step S402, calculating the node weight of the candidate topic in the knowledge graph according to popularity and node degree.

[0131] In step S402 of some embodiments, the proportional coefficient of popularity and the proportional coefficient of node degree are obtained; the node weight of the candidate topic in the knowledge graph is calculated according to the popularity, the node degree, the proportional coefficient of popularity and the proportional coefficient of node degree, and the expression of the node weight is as follows:

[0132] W t =α×P t +β×D t ,

[0133] α+β=1

[0134] Among them, W t is the node weight of the candidate topic in the knowledge graph, α is the proportional coefficient of popularity, P t is the popularity, β is the proportional coefficient of node degree, D t is the node degree.

[0135] For example, if the proportional coefficient of popularity is greater than the proportional coefficient of node degree, the recommendation results will mainly rely on a set of most popular topics; if the proportional coefficient of popularity is less than the proportional coefficient of node degree, it will rely on topics that have not been frequently searched by the platform.

[0136] See also Figure 5 In some embodiments, the topic recommendation method provided in the embodiment of the present application further includes a step of constructing a knowledge graph, and the step of constructing a knowledge graph may also include but is not limited to steps S501 to S504:

[0137] Step S501, initialize the knowledge graph using topics, relationship types and relationships between topics.

[0138] In step S501 of some embodiments, topics, relationship types, and relationships between topics are encoded, and a knowledge graph is initialized using a set of topics, relationship types, and relationships.

[0139] Step S502, update the knowledge graph through the triangulation verification method.

[0140] In step S502 of some embodiments, the subject set is iterated multiple times, each time defining and correcting relationships, and reviewing the relationships defined with each other through triangulation to verify their correctness and validity.

[0141] It should be noted that if there is a disagreement during the triangulation process, a third-party expert will intervene in the discussion and make the final decision on the relationship.

[0142] Among them, if no consensus is reached, conflicting relationships will be added to the original KG (i.e. knowledge graph), allowing participants to make the final decision on the correctness of these relationships as the final party.

[0143] Step S503, obtain knowledge about topics, relationship types and relationships between topics, and expand the knowledge graph through knowledge.

[0144] This application does not specifically limit the method of acquiring knowledge, and it can be flexibly selected in combination with actual detection needs. For example, it can be adopted to acquire new community knowledge about topics, relationship types and relationships between topics, or it can be adopted to acquire encyclopedic knowledge about topics, relationship types and relationships between topics.

[0145] In step S503 of some embodiments, knowledge about topics, relationship types, and relationships between topics is obtained, and the knowledge graph is expanded and updated through the knowledge.

[0146] Understandably, because the field is dynamic, new topics emerge and old topics evolve, so that previously defined relationships may be affected.

[0147] In this embodiment, knowledge about topics, relationship types, and relationships between topics is obtained, and the knowledge graph is expanded by knowledge, so that the validity period of the knowledge graph can be extended.

[0148] Step S504, evaluating the knowledge graph through the labeling task.

[0149] In step S504 of some embodiments, all relations in the knowledge graph are evaluated through an objective labeling task, and the evaluators agree or disagree with each relation by marking it as "True" or "False"; when each relation has at least three votes from credible evaluators and only when the majority of evaluators who review these relations mark it as "True", it will be added to the knowledge graph, otherwise, it will be deleted from the knowledge graph.

[0150] See also Figure 6 In some embodiments, step S103 includes but is not limited to steps S601 to S602:

[0151] Step S601, obtaining the final fuzzy representation of the user and the final fuzzy representation of the item of the hierarchical fusion fuzzy depth network.

[0152] In step S601 of some embodiments, the joint context of the user and the item is aligned into an embedding space by a hierarchical fusion fuzzy deep network, so as to obtain a final fuzzy representation of the user and a final fuzzy representation of the item. The hierarchical fusion fuzzy deep network includes fuzzy neural network learning and a multi-layer deep neural network.

[0153] Step S602, calculating the relevance score between the candidate topic and the project according to the final fuzzy representation of the user, the final fuzzy representation of the project, the relevance probability between the candidate topic and the project, and the node weight.

[0154] In step S602 of some embodiments, the final fuzzy representation of the user and the final fuzzy representation of the project are added to the knowledge graph recommendation model as auxiliary information.

[0155] In some embodiments, the knowledge graph recommendation model calculates the relevance score between the candidate topic and the project according to the final fuzzy representation of the user, the final fuzzy representation of the project, the relevance probability between the candidate topic and the project, and the node weight. The expression of the node weight is as follows:

[0156]

[0157]

[0158] in, is the relevance score between the candidate topic and the project, W t is the node weight, is the probability of the candidate topic being associated with the project, α and β are hyperparameters, and X u,FDNN is the final fuzzy representation of the user, γ i is the potential factor matrix of project households, γ μ is the user's latent factor matrix, x i,FDNN is the final fuzzy representation of the project, and N(I) is the set of topics that are directly adjacent to at least one topic in the initial set of topics of the project.

[0159] Figure 7 is an optional flow chart of the topic recommendation method provided in the embodiment of the present application applied to a topic recommendation system. Figure 7 The method may include but is not limited to the following steps.

[0160] Step 1: knowledge graph construction.

[0161] It should be noted that the repository of the atomic capability platform of the recommendation system is regarded as S = {r1, r2, …, r n}, where ri Represents a single project. Each repository may contain various types of information, such as capability name, billing type, price, description, available volume, service provider, etc. Each project includes a set of topics T = {t1, t2, …, t m}, where m is the number of topics assigned to the repository.

[0162] The system consists of two parts: recommendation model and business system. Building the recommendation model consists of two stages: first, adding i The first step is to create an initial topic set for a given topic-free repository r j Recommend a set of topics, leverage semantic sources of information to build more powerful recommendation systems, and use semantic relationships between high-quality topics to enhance recommendation models.

[0163] The business system provides three channels for searching on the atomic capability platform: search box, general area, and industry area, and provides capability recommendations on the homepage. The corresponding model prediction results are returned in a structured manner through the backend and model service. At the same time, the backend will report the retrieval to facilitate subsequent model iterations.

[0164] It is understandable that an excellent method for building a recommendation system requires high-quality topics and their semantic relationships. KG (i.e., knowledge graph) is a solution for storing such information in a structured format. For example, the schematic diagram of the knowledge graph construction process is as follows: Figure 8 shown.

[0165] In the initialization phase, coders manually code, using a set of topics, relationship types, and relationship initialization KGs, and cross-evaluating the coding results in combination with triangulation. Furthermore, each coder iterates the theme set multiple times, defining and correcting relationships each time, and reviewing each other's defined relationships through triangulation to verify their correctness and validity. If there is a disagreement, a third-party expert will intervene in the discussion and make the final decision on the relationship. If no consensus is reached, conflicting relationships will be added to the initial KG, allowing participants to make the final decision on the correctness of these relationships as the final party.

[0166] Furthermore, in the expansion phase, the relationships defined in the previous step need to be evaluated and expanded into a more comprehensive KG. In this process, while evaluating the pre-defined relationships, new community knowledge about topics, relationship types, and relationships between topics is obtained to extend the validity period of the KG. Because the domain is dynamic, new topics will emerge and old topics will develop, which may affect the previously defined relationships.

[0167] In the evaluation phase, all relations defined in the first two steps are evaluated through an objective labeling task. Evaluators approve or disapprove each relation by marking it as "True" or "False", with at least three credible evaluators voting, and only when the majority of evaluators who reviewed the relation mark it as "True" will it be added to the KG.

[0168] This construction method will cause label explosion when the evaluator submits the suggestion text in free format, resulting in redundancy of relationship types and topic semantics. Additional auxiliary strategies and functions, such as evaluator reliability check, redundant duplication check, and regular random check, can be used to ensure the semantic uniqueness of relationship types and reduce the probability of other abnormal situations. Through the above steps, a high-quality atomic capability topic KG can be obtained.

[0169] Furthermore, after obtaining the high-quality knowledge graph of the atomic capability platform, corresponding topic prediction models were proposed for two different scenarios: topic enhancement during user search and recommendation of topic sets when no search was performed.

[0170] Step 2: topic set recommendation.

[0171] It should be noted that, assuming that the repository does not mark any initial topic or a new user does not have any search topic record, how to achieve cold start? This embodiment proposes KGRec+ as a topic set recommendation model.

[0172] In some embodiments, the text data is first concatenated and converted into their own TF-IDF vectors (i.e., term frequency-inverse document frequency vectors) for use by the classifier. After training on the text data of the repository and the specified topics, the model can predict a list of related topics for a given project. This set is predicted using a machine learning-based component and supplemented with a KGRec model.

[0173] Among them, two classifiers are used in the model. One is a multinomial naive Bayes (MNB) text classifier, which takes text data in the software project as input and predicts the Pr_t^p of the feature topic. The other is a multi-class multi-label Logit model (LR) classifier, which changes the loss function to cross entropy loss and predicts the probability distribution as a multinomial probability distribution to natively support multi-class classification problems.

[0174] Further, the model is trained to predict the Take the highest The KGRec model takes this set of topics as input and expands the list of m topics with g or more topics to return a list of k=m+g topics as recommendations (i.e., the target recommendation list).

[0175] Step 3, search for topic enhancements.

[0176] In some embodiments, if there is a marked initial topic in the repository or the user has a topic record, search topic enhancement is performed. Search topic enhancement is performed through the KGRec model (i.e., knowledge graph recommendation model) and the hierarchical fusion fuzzy network architecture FDNN to enhance the topic, and through the diffusion activation theory and fuzzy learning, it effectively reduces feature noise and recommendation uncertainty, and improves recommendation performance.

[0177] Among them, the KGRec model is an extended initial set. It applies the diffusion activation theory based on semantic networks on KG. It labels a set of source nodes with weights or activations, and then iteratively propagates or diffuses the activations to other nodes connected to the source nodes.

[0178] First, the KG is annotated with node weights based on the popularity of the topic in the Atomic Capability Platform (hot words) and the topic nodes in the KG. Because if a topic is frequently searched in the platform, it must be a more useful and valuable topic, and topics with higher value in the KG will be associated with more topics, so they are more likely to be associated with more projects that may be assigned related topics. The weight calculation formula is shown below.

[0179] W t =α×P t +β×D t ,

[0180] α+β=1

[0181] Among them, W t represents the weight of the node corresponding to topic t, P t and D t They are measures of the popularity (i.e., prevalence) and degree (i.e., node degree) of topic t, respectively. α and β are proportional coefficients of the scores.

[0182] It is understandable that for different values ​​of α and β, the results will be different. For example, for α>β, the results will mainly rely on a set of the most popular atomic capability topics, and for α<β, they will rely on topics that have not been frequently searched by the platform.

[0183] The calculation formulas for popularity and node degree are as follows:

[0184]

[0185] Among them, P t is the popularity of topic t, D t is the node degree of topic t, n t is the number of items in the platform tagged with topic t, d t is the total degree of topic t in KG, and T is the set of topics in KG.

[0186] Exemplarily, in order to consider both popularity and node degree, α=β=0.5 is set.

[0187] Furthermore, considering that topic t (i.e., candidate topic) is related to item p, the probability is I is the initial set of topics assigned to item p, and N(I) represents the set of topics that are directly adjacent to at least one topic in I. For all t in N(I), the probability is expanded using the following formula to calculate the relevance score of topic t with the projection p along the KG edge:

[0188]

[0189]

[0190] in, is the relevance score between the candidate topic and the project, W t is the node weight, is the probability of the candidate topic being associated with the project, and N(I) is the topic set that is directly adjacent to at least one topic in the initial topic set of the project.

[0191] Furthermore, the model is based on Sort the candidate topic list and return the top k topics as the recommendation list (i.e., the target recommendation list).

[0192] Optionally, in order to reduce the noise of model learning, a hierarchical fusion fuzzy network is proposed to align the joint context of users and items into an embedding space, and form a complete potential embedding output through a fusion mechanism to assist the KGRec model in achieving better recommendation effects.

[0193] Among them, fuzzy learning avoids learning data feature noise by integrating a given network architecture with different fuzzy rule learning layers, thereby better fine-tuning multiple task-driven learning objectives. This application proposes a hierarchical fusion fuzzy deep network (FDNN) to align the joint context of users and items into an embedding space. The entire FDNN consists of two main parts, each input vector is processed by a component based on fuzzy learning or mapping embedding, and is fused through a fusion mechanism before being fed to the fully connected layer to form a complete potential embedding output.

[0194] The first part is fuzzy neural network learning, which mainly includes two input layers. First, when constructing KG, the topic set is obtained, and the corresponding embedding is obtained through the metapath2vec algorithm as an input embedding vector. Then, the user-related description text corresponding to each topic is represented by the LSTM structure, and the output h of the last layer is obtained as an input. Then, each input layer in this part is linked to different member layers, which can distinguish the corresponding potential features of each node type. In each member layer, the input variables are associated with the specific dimensions and potential features of each input vector.

[0195] Each fuzzy neuron is a mapping function, represented by f fuzz (.):R→[0,1]. The i-th fuzzy neuron in the l-th layer is represented as Used to calculate the fuzziness of a specific k-th input variable, represents the output of the i-th fuzzy neuron in the l-th fuzzy neural layer, where μ and σ 2 are the mean and variance respectively. For the fuzzy rule layer, the “AND” fuzzy logic operation connecting the input node sets between the two layers is uniformly used. The expression of the output node weight of the i-th fuzzy neuron in the l-th fuzzy neural layer is as follows:

[0196]

[0197] in, is the output of the fuzzy neuron in the lth fuzzy neural layer, indicates that the i-th fuzzy neuron is in the l-th layer, Used to calculate the fuzziness of a specific k-th input variable, is the exponential part of an exponential function, Represents the effect of the distance between the input and the mean under normalization of the variance.

[0198] Optionally, a multi-layer deep neural network is used in the second part to refine the previously obtained topic embedding and associated text output into a high-level representation before feeding it to the fusion layer. The neural layer is defined as a linear fully connected layer with a sigmoid activation function that connects each latent feature and input variable of the vector between two consecutive neural layers. The formula representation of each layer is as follows:

[0199]

[0200] in, represents the linear combination output of the i-th neuron in the l-th layer, Linear represents linear transformation, is the deep feature representation output of the previous layer l-1, is the deep feature representation output of the lth layer, and Sigmod represents the Sigmoid activation function.

[0201] Furthermore, the fusion mechanism is combined and defined as a nonlinear function expressed as The specific formula is as follows: W i and W j are all learnable matrices, and b is a constant. Finally, the output is sent to the fully connected layer to complete the prediction task and obtain the final fuzzy representation X of the user and item. u,FDNN , X i,FDNN .

[0202]

[0203] in, is the fuzzy representation of the output of the lth layer after the fusion mechanism, is the deep feature representation of the l-1th layer, is the output of the fuzzy neuron in the l-1th fuzzy neural layer, W i and W j are all learnable matrices, and b is a constant.

[0204] Finally, the fuzzy representation is added to the KGRec model as auxiliary information, and the formula is as follows:

[0205]

[0206] in, is the relevance score between the candidate topic and the project, W t is the node weight, is the probability of the candidate topic being associated with the project, α and β are hyperparameters, and X u,FDNN is the final fuzzy representation of the user, γ i is the potential factor matrix of project households, γ μ is the user's latent factor matrix, X i,FDNN is the final fuzzy representation of the project, and N(I) is the set of topics that are directly adjacent to at least one topic in the initial set of topics of the project.

[0207] Understandably, W t represents the weight of the node corresponding to topic t. The first half is the same as before, calculating the correlation score between topic t and the projection p along the KG edge. In the second half, α and β are hyperparameters, γ i and γ μ is the latent factor matrix of items and users. Finally, the top k topics are returned as the recommendation list.

[0208] In this embodiment, an initial data set of an atomic capability platform is constructed through expert guidance and crowdsourcing, and the semantic relations within it are extracted. The relational information is optimized through evaluation, suggestion, etc., and encapsulated in a knowledge graph of a specific field. Then, for the topic enhancement scenario, a KGRec model is proposed to predict missing topics. In addition to the topics directly assigned according to the user's search keywords, the missing topics related to the search entity are supplemented to improve the completeness of the assigned topic set. At the same time, in order to reduce the noise of model learning, a hierarchical fusion fuzzy network is proposed to align the joint context of users and projects into an embedding space, and a complete potential embedding output is formed through a fusion mechanism to assist the KGRec model in obtaining better recommendation effects. Afterwards, through the extended activation technology, a KGRec+ model is established for topic prediction for the topic set recommendation scenario, based on the text data of the project and the semantics obtained in the knowledge graph, and the semantic relations between high-quality topics are used to enhance the recommendation model. Finally, three channels are provided on the atomic capability platform for search: search box, general area, and industry area, and capability recommendations are provided on the homepage. The deployed offline model service will return the recommended results in a structured manner when the user searches the page.

[0209] This embodiment uses the expert mode and crowdsourcing method to sort out a data set specifically for the atomic capability platform theme, and encapsulates it in the corresponding knowledge graph. By integrating the semantic relationship between the topics and the hierarchical fusion fuzzy network, two models are constructed to solve the problem of topic enhancement for missing topic completion and the problem of predicting related topics for users. On the one hand, considering the diversity of topics, different levels of abstraction, and the sparsity and dispersion of data on specific topics, a main body manual method is adopted, combined with automation technology to promote the acquisition and evaluation of knowledge. On the other hand, by integrating the semantic relationship between topics, the accuracy of atomic capability recommendations is improved, and the problem of label explosion is solved by carefully evaluating the input set of topics to reduce interference in the specified topic and recommend related topics for the project. At the same time, fuzzy learning is used in the recommendation task to assist the original recommendation task and reduce the data noise in the results obtained based on the diffusion activation theory to obtain better accuracy performance in the recommendation problem.

[0210] See also Fig. 9 The present application also provides a topic recommendation device, which can implement the above topic recommendation method. The device includes:

[0211] A first calculation module 901 is used to calculate the popularity of a candidate topic according to the number of items corresponding to the candidate topic, and the candidate topic belongs to a candidate topic list;

[0212] The second calculation module 902 is used to calculate the node weight of the candidate topic in the knowledge graph according to the popularity;

[0213] The third calculation module 903 is used to calculate the correlation score between the candidate topic and the project according to the correlation probability between the candidate topic and the project and the node weight;

[0214] The recommendation module 904 is used to sort the candidate topic list in descending order according to the relevance score, and return the first several topics as the content of the target recommendation list.

[0215] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0216] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above topic recommendation method when executing the computer program. The electronic device can be any smart terminal including a tablet computer, a car computer, etc.

[0217] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0218] See also Fig.10 , Fig.10 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0219] The processor 1001 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0220] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other applications. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1002, and the processor 1001 calls and executes the subject recommendation method of the embodiment of this application;

[0221] Input / output interface 1003, used to implement information input and output;

[0222] The communication interface 1004 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0223] A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 );

[0224] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0225] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned topic recommendation method is implemented.

[0226] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiments, the functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0227] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0228] The topic recommendation method, topic recommendation device, electronic device and storage medium provided in the embodiments of the present application calculate the popularity of candidate topics according to the number of projects corresponding to the candidate topics, thereby reflecting the popularity of the candidate topics and helping to determine the priority of recommendation; calculate the node weight of the candidate topic in the knowledge graph according to the popularity, which helps to highlight important topics and thus reduce the influence of irrelevant or low-value topics; calculate the correlation score between the candidate topic and the project according to the correlation probability of the candidate topic and the project and the node weight, which is conducive to evaluating the correlation between each candidate topic and the project, so that the recommendation is more in line with the actual needs of the user; sort the candidate topic list in descending order according to the correlation score, and return the first several topics as the content of the target recommendation list, focusing on the actual needs of the user, reducing unnecessary information interference, improving the accuracy of topic recommendation during retrieval, and reducing the noise of retrieval topics.

[0229] The embodiments 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. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0230] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0231] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0232] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0233] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0234] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0235] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, 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 through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0236] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0237] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0238] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0239] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A topic recommendation method, characterized in that: Applied to the knowledge graph recommendation model, the method comprises the following steps: Calculating the popularity of a candidate topic according to the number of items corresponding to the candidate topic, the candidate topic belonging to the candidate topic list; Calculate the node weight of the candidate topic in the knowledge graph according to the popularity, wherein the knowledge graph includes topics, relationship types, and relationships between topics; Calculate the relevance score between the candidate topic and the project according to the relevance probability between the candidate topic and the project and the node weight; The candidate topic list is sorted in descending order according to the relevance score, and the first several topics are returned as the contents of the target recommendation list.

2. The method according to claim 1, characterized in that The method further comprises the step of obtaining a candidate topic list, wherein obtaining the candidate topic list comprises: If there is a search topic record of the user, generating the candidate topic list according to the search topic record; If there is an initial topic in the repository, the candidate topic list is generated according to the initial topic.

3. The method according to claim 1, characterized in that The method further comprises the step of obtaining a candidate topic list, wherein obtaining the candidate topic list comprises: If the initial topic does not exist in the repository and the search topic record of the user does not exist, acquiring text data of the repository; Converting the text data into a term frequency-inverse document frequency vector; Training a topic set recommendation model by using the term frequency-inverse document frequency vector; The target project is input into the topic set recommendation model to obtain a candidate topic list.

4. The method according to claim 1, characterized in that: The step of calculating the node weight of the candidate topic in the knowledge graph according to the popularity includes: Calculating the node degree of the candidate topic in the knowledge graph; The node weight of the candidate topic in the knowledge graph is calculated according to the popularity and the node degree.

5. The method according to claim 1, characterized in that The method further includes a step of constructing a knowledge graph, wherein the step of constructing a knowledge graph includes: Initialize the knowledge graph using topics, relationship types, and relationships between topics; Updating the knowledge graph by triangulation; Acquire knowledge about topics, relationship types, and relationships between topics, and expand the knowledge graph through the knowledge; The knowledge graph is evaluated through a labeling task.

6. The method according to claim 1, characterized in that The calculating the relevance score between the candidate topic and the project according to the relevance probability between the candidate topic and the project and the node weight includes: Obtain the final fuzzy representation of the user and the final fuzzy representation of the item by using a hierarchical fusion fuzzy deep network; The relevance score between the candidate topic and the project is calculated according to the final fuzzy representation of the user, the final fuzzy representation of the project, the relevance probability between the candidate topic and the project, and the node weight.

7. The method according to claim 4, characterized in that The step of calculating the node weight of the candidate topic in the knowledge graph according to the popularity and the node degree includes: Get the proportional coefficient of popularity and the proportional coefficient of node degree; The node weight of the candidate topic in the knowledge graph is calculated according to the popularity, the node degree, the proportional coefficient of the popularity and the proportional coefficient of the node degree. The expression of the node weight is as follows: IN t =α×P t +β×D t , α+β＝1 Among them, W t is the node weight of the candidate topic in the knowledge graph, α is the proportional coefficient of the popularity, P t is the popularity, β is the proportional coefficient of the node degree, D t is the node degree.

8. A topic recommendation device, characterized in that: The device comprises: A first calculation module, configured to calculate the popularity of a candidate topic according to the number of items corresponding to the candidate topic, the candidate topic belonging to the candidate topic list; A second calculation module is used to calculate the node weight of the candidate topic in the knowledge graph according to the popularity; A third calculation module, used to calculate the relevance score between the candidate topic and the project according to the relevance probability between the candidate topic and the project and the node weight; The recommendation module is used to sort the candidate topic list in descending order according to the relevance score and return the first several topics as the content of the target recommendation list.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing 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 7 is implemented.

Images