CN112231569A - News recommendation method and device, computer equipment and storage medium - Google Patents

News recommendation method and device, computer equipment and storage medium Download PDF

Info

Publication number
CN112231569A
CN112231569A CN202011148692.6A CN202011148692A CN112231569A CN 112231569 A CN112231569 A CN 112231569A CN 202011148692 A CN202011148692 A CN 202011148692A CN 112231569 A CN112231569 A CN 112231569A
Authority
CN
China
Prior art keywords
news
vector
model
entity
expression vector
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011148692.6A
Other languages
Chinese (zh)
Inventor
尹曦
赵亮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Life Insurance Company of China Ltd
Original Assignee
Ping An Life Insurance Company of China Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Life Insurance Company of China Ltd filed Critical Ping An Life Insurance Company of China Ltd
Priority to CN202011148692.6A priority Critical patent/CN112231569A/en
Publication of CN112231569A publication Critical patent/CN112231569A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/95Retrieval from the web
    • G06F16/953Querying, e.g. by the use of web search engines
    • G06F16/9535Search customisation based on user profiles and personalisation
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24Querying
    • G06F16/245Query processing
    • G06F16/2455Query execution
    • G06F16/24553Query execution of query operations
    • G06F16/24554Unary operations; Data partitioning operations
    • G06F16/24556Aggregation; Duplicate elimination
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/28Databases characterised by their database models, e.g. relational or object models
    • G06F16/284Relational databases
    • G06F16/288Entity relationship models
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/31Indexing; Data structures therefor; Storage structures
    • G06F16/313Selection or weighting of terms for indexing
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • G06F16/335Filtering based on additional data, e.g. user or group profiles
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35Clustering; Classification
    • G06F16/355Class or cluster creation or modification
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/95Retrieval from the web
    • G06F16/958Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/205Parsing
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/279Recognition of textual entities
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/30Semantic analysis

Abstract

The embodiment of the application belongs to the technical field of artificial intelligence, can be applied to the fields of intelligent government affairs and/or medical science and technology, and relates to a news recommendation method, a news recommendation device, computer equipment and a storage medium, wherein the news recommendation method comprises the steps of inputting a received news document into a semantic extraction model of a preset initial recommendation model to obtain a semantic feature expression vector; calculating entity feature representation vectors of single entities in news documents; aggregating all entity feature representation vectors to obtain an aggregate representation vector; processing the aggregation expression vector, the semantic feature expression vector and a preset user vector to obtain a prediction result; and calculating a loss function value of the initial recommendation model, iteratively training the initial recommendation model until the initial recommendation model is converged, obtaining a trained recommendation model, and recommending news to a user based on the trained recommendation model. The trained recommendation model may be stored in a blockchain. The method and the device enhance the dimensionality of news recommendation.

Description

News recommendation method and device, computer equipment and storage medium
Technical Field
The present application relates to the field of artificial intelligence technologies, and in particular, to a news recommendation method and apparatus, a computer device, and a storage medium.
Background
At present, the personalized recommendation system is widely applied to various industries such as the Internet, electronic commerce, life, entertainment and the like. Compared with a traditional search system, a user does not need to actively input target information or query (query), but the system automatically matches appropriate news information according to user portrait information and scene context information and then pushes the news information to the user, and the mode greatly enhances the efficiency of user information acquisition and user experience, so that the user scale of products is promoted.
The news recommendation methods are various and include a news recommendation method based on a collaborative filtering algorithm of an article, a news recommendation method based on a latent semantic model algorithm, a news recommendation method based on a recommendation algorithm of an association rule, and the like. However, the existing news recommendation methods have the problem that the recommended news content is single, so that the user experience is poor.
Disclosure of Invention
The embodiment of the application aims to provide a news recommendation method, a news recommendation device, computer equipment and a storage medium, so that the dimensionality of news recommendation is enhanced, and the diversity and the accuracy of model recommendation are effectively improved.
In order to solve the above technical problem, an embodiment of the present application provides a news recommendation method, which adopts the following technical solutions:
receiving a news document, inputting the news document into a semantic extraction model of a preset initial recommendation model, and obtaining a semantic feature expression vector output by the semantic extraction model, wherein the news document carries a tag;
identifying single entities from the news documents, and calculating entity feature expression vectors of the single entities in the news documents;
based on the semantic feature representation vectors, aggregating the entity feature representation vectors of all the single entities in the news document to obtain an aggregate representation vector;
processing the aggregation expression vector, the semantic feature expression vector and a preset user vector to obtain a prediction result;
calculating a loss function value of the initial recommendation model based on the prediction result and the label corresponding to the news document, iteratively training the initial recommendation model until the initial recommendation model is converged, obtaining a trained recommendation model, and recommending news to a user based on the trained recommendation model.
Further, the step of calculating the entity feature representation vector of the single entity in the news document comprises:
respectively calculating a knowledge map embedded expression vector of a single entity and a context expression vector of the single entity in a news document;
and adding the knowledge graph embedding expression vector of the single entity and the context expression vector of the single entity in the news document in corresponding dimensions to obtain an entity feature expression vector of the single entity.
Further, the step of calculating the knowledge-graph embedded representation vector of the single entity comprises:
and acquiring a knowledge graph, and performing knowledge representation learning on a single entity based on the knowledge graph and a graph model in the initial recommendation model to obtain a knowledge graph embedded representation vector of the single entity.
Further, the step of calculating the context representation vector of the single entity in the news document comprises:
identifying the number of times that a single entity appears in a news document, the position where the single entity appears and the category of the single entity;
and respectively converting the times of the single entity appearing in the news document, the position of the single entity appearing and the category of the single entity into corresponding numbers based on a preset corresponding relation table to obtain the context expression vector of the single entity in the news document.
Further, the step of aggregating the entity feature representation vectors of all the single entities in the news document based on the semantic feature representation vectors to obtain an aggregate representation vector includes:
taking the semantic feature expression vector as a query vector, and paying attention to each single entity contained in the document based on an attention mechanism set in the initial recommendation model;
respectively converting the noted entity feature representation vectors corresponding to the single entity into a key vector and a value vector;
calculating the inner product of the query vector and each key vector, and normalizing all the inner products to obtain the attention weight of the entity feature expression vector;
and carrying out weighted summation on the entity feature representation vectors based on the attention weight to obtain the aggregation representation vector.
Further, the step of processing the aggregate expression vector, the semantic feature expression vector, and a preset user vector to obtain a prediction result includes:
splicing the aggregation expression vector and the semantic feature expression vector;
inputting the spliced vector into a full connection layer in the initial recommendation model to obtain a news document feature vector;
and inputting the news document feature vector and the user vector into a binary classification model of the initial recommendation model to obtain a prediction result.
Further, the semantic extraction model includes a long and short term memory network layer, and the step of inputting the news document into a preset semantic extraction model to obtain the semantic feature expression vector output by the semantic extraction model includes:
converting each word in the news document into a word vector based on a semantic extraction model;
extracting all word vectors in each sentence based on a long-short term memory network layer in a semantic extraction model, and performing weighted summation on the word vectors in each sentence through an attention mechanism in the semantic extraction model to generate sentence characteristic vectors;
and carrying out weighted summation on the sentence vectors through an attention mechanism in a semantic extraction model to generate document feature vectors serving as the semantic feature expression vectors.
In order to solve the above technical problem, an embodiment of the present application further provides a news recommendation device, which adopts the following technical solutions:
a news recommender, comprising:
the receiving module is used for receiving a news document, inputting the news document into a semantic extraction model of a preset initial recommendation model, and obtaining a semantic feature expression vector output by the semantic extraction model, wherein the news document carries a tag;
the calculation module is used for identifying single entities from the news documents and calculating entity feature expression vectors of the single entities in the news documents;
the aggregation module is used for aggregating the entity feature representation vectors of all the single entities in the news document based on the semantic feature representation vectors to obtain an aggregate representation vector;
the processing module is used for processing the aggregation expression vector, the semantic feature expression vector and a preset user vector to obtain a prediction result; and
and the iteration module is used for calculating a loss function value of the initial recommendation model based on the prediction result and the label corresponding to the news document, iteratively training the initial recommendation model until the initial recommendation model is converged to obtain a trained recommendation model, and recommending news to a user based on the trained recommendation model.
In order to solve the above technical problem, an embodiment of the present application further provides a computer device, which adopts the following technical solutions:
a computer device comprising a memory having computer readable instructions stored therein and a processor that when executed implements the steps of the news recommendation method described above.
In order to solve the above technical problem, an embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solutions:
a computer readable storage medium having computer readable instructions stored thereon which, when executed by a processor, implement the steps of the news recommendation method described above.
Compared with the prior art, the embodiment of the application mainly has the following beneficial effects:
according to the method and the device, the entity feature representation vectors of the single entity obtained from the news document are aggregated through the semantic feature representation vectors, the aggregated representation vectors are generated, and the richness of the aggregated representation vectors is improved. The aggregation expression vector, the semantic feature expression vector and the preset user vector are processed to generate a prediction result, and the prediction result is generated based on the aggregation expression vector and the semantic feature expression vector, so that the recommendation dimensionality is enhanced, the model recommendation diversity and accuracy are effectively improved, and the common problem that the user experience is reduced due to the fact that the model recommendation is more narrow is effectively avoided.
Drawings
In order to more clearly illustrate the solution of the present application, the drawings needed for describing the embodiments of the present application will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present application, and that other drawings can be obtained by those skilled in the art without inventive effort.
FIG. 1 is an exemplary system architecture diagram in which the present application may be applied;
FIG. 2 is a flow diagram of one embodiment of a news recommendation method according to the present application;
FIG. 3 is a schematic block diagram of one embodiment of a news recommender in accordance with the present application;
FIG. 4 is a schematic block diagram of one embodiment of a computer device according to the present application.
Reference numerals: 200. a computer device; 201. a memory; 202. a processor; 203. a network interface; 300. a news recommendation device; 301. a receiving module; 302. a calculation module; 303. a polymerization module; 304. a processing module; 305. and (5) an iteration module.
Detailed Description
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "including" and "having," and any variations thereof, in the description and claims of this application and the description of the above figures are intended to cover non-exclusive inclusions. The terms "first," "second," and the like in the description and claims of this application or in the above-described drawings are used for distinguishing between different objects and not for describing a particular order.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.
As shown in fig. 1, the system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the terminal devices 101, 102, 103 and the server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, to name a few.
The user may use the terminal devices 101, 102, 103 to interact with the server 105 via the network 104 to receive or send messages or the like. The terminal devices 101, 102, 103 may have various communication client applications installed thereon, such as a web browser application, a shopping application, a search application, an instant messaging tool, a mailbox client, social platform software, and the like.
The terminal devices 101, 102, 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III, mpeg compression standard Audio Layer 3), MP4 players (Moving Picture Experts Group Audio Layer IV, mpeg compression standard Audio Layer 4), laptop portable computers, desktop computers, and the like.
The server 105 may be a server providing various services, such as a background server providing support for pages displayed on the terminal devices 101, 102, 103.
It should be noted that, the news recommendation method provided in the embodiment of the present application is generally executed by a server/terminal device, and accordingly, the news recommendation apparatus is generally disposed in the server/terminal device.
It should be understood that the number of terminal devices, networks, and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.
With continued reference to FIG. 2, a flow diagram of one embodiment of a news recommendation method in accordance with the present application is shown. The news recommending method comprises the following steps:
s1: receiving a news document, inputting the news document into a semantic extraction model of a preset initial recommendation model, and obtaining a semantic feature expression vector output by the semantic extraction model, wherein the news document carries a label.
In this embodiment, a news document received in the application is a training sample, and is obtained based on historical click data of a user, and a tag corresponding to the news document is used to identify whether the user has clicked the news document. The current news recommendation model generally employs a deep knowledge aware network DKN. The semantic extraction model used in the present application is HAN (Hierarchical Attention network) and the whole network structure includes four parts, a word sequence encoder, an Attention layer based on word level, a sentence encoder and an Attention layer based on sentence level. And extracting the news document through the HAN to obtain a semantic feature expression vector of the news document. And a semantic feature expression vector is obtained through a semantic extraction model of the initial recommendation model, so that the subsequent model training process is facilitated.
In this embodiment, the electronic device (e.g., the server/terminal device shown in fig. 1) on which the news recommendation method operates may receive the news document through a wired connection or a wireless connection. It should be noted that the wireless connection means may include, but is not limited to, a 3G/4G connection, a WiFi connection, a bluetooth connection, a WiMAX connection, a Zigbee connection, a uwb (ultra wideband) connection, and other wireless connection means now known or developed in the future.
Specifically, the semantic extraction model includes a long and short term memory network layer, and in step S1, that is, the step of inputting the news document into a preset semantic extraction model to obtain a semantic feature expression vector output by the semantic extraction model includes:
converting each word in the news document into a word vector based on a semantic extraction model;
extracting all word vectors in each sentence based on a long-short term memory network layer in a semantic extraction model, and performing weighted summation on the word vectors in each sentence through an attention mechanism in the semantic extraction model to generate sentence characteristic vectors;
and carrying out weighted summation on the sentence vectors through an attention mechanism in a semantic extraction model to generate document feature vectors serving as the semantic feature expression vectors.
In this embodiment, the existing extraction of the document feature vectors is mainly divided into three types: the LDA unsupervised topic model is a large-scale pre-training language model based on a DSSM frame hidden space semantic representation model, BERT and the like. Relative to the models, a semantic extraction model HAN (semantic attribute network) model adopted in the application is used for weighting and summing hidden vectors of each level according to sequential dimensions from words to sentences to paragraphs and sections step by using an LSTM (long-short term memory) network in combination with an attribute mechanism, so that semantic features of different levels can be extracted. The logic of the method is clearer and the interpretability is stronger.
S2: and identifying single entities from the news documents, and calculating entity feature expression vectors of the single entities in the news documents.
In this embodiment, the entity feature representation vectors for individual entities in the news document are calculated based on an initial recommendation model. The single entities are identified from the news documents, either by using a common vocabulary or by single entity identification tools. Single entities are identified and extracted from news documents. The general vocabulary comprises preset single entities, and the single entities are identified from the news documents based on the general vocabulary. The single Entity Recognition tool may be an NER (Named Entity Recognition), and may be selected according to actual needs in a specific operation process.
It should be noted that: the single entity identified in the present application is a proper noun that exists in reality, including a name of a person, a name of a place, a name of an organization, and the like.
Specifically, in step S2, the step of calculating the entity feature representation vector of the single entity in the news document includes:
respectively calculating a knowledge map embedded expression vector of a single entity and a context expression vector of the single entity in a news document;
and adding the knowledge graph embedding expression vector of the single entity and the context expression vector of the single entity in the news document in corresponding dimensions to obtain an entity feature expression vector of the single entity.
In this embodiment, the calculation of the entity feature expression vector of a single entity includes two parts: calculating a knowledge graph representation vector of a single entity and calculating a context representation vector of the single entity in a news document. Adding the knowledge graph embedded representation of the entity and the context representation of the entity in the news document in corresponding dimensions, enriching the dimensions of the entity feature representation vector, and expanding the feature range represented by the entity feature representation vector. For example: the knowledge-graph embedded expression vector of the entity is (0,1,1, 0); and if the context expression vector of the entity in the news document is (1,0,0,1), adding the knowledge graph embedded expression of the entity and the context expression of the entity in the news document in corresponding dimensions, and obtaining an entity feature expression vector of a single entity, wherein the entity feature expression vector is (1,1,1, 1).
Further, the step of calculating the knowledge-graph embedded representation vector of the single entity comprises:
and acquiring a knowledge graph, and performing knowledge representation learning on a single entity based on the knowledge graph and a graph model in the initial recommendation model to obtain a knowledge graph embedded representation vector of the single entity.
In this embodiment, the present application computes a knowledge-embedded representation of an entity using a graph neural network-based approach. The knowledge representation learning is carried out on the entity by using a graph model of GraphSAGE (inductive graph representation learning), namely the graph model in the initial recommendation model of the application refers to the graph model of GraphSAGE. Compared with the classical GCN (graph convolution network), the GraphSAGE graph model has better generalization capability on unknown data, samples neighborhood nodes and supports various neighborhood node information aggregation modes, so that the method for carrying out knowledge representation on an entity is more efficient and flexible. In addition, the knowledge graph is obtained from an open source network database, and the encyclopedic knowledge graph is selected, wherein the encyclopedic knowledge graph comprises other entities and relations related to a single entity of the news document. And performing knowledge representation learning (knowledge representation learning refers to the representation of learning entity vectors and relationship vectors) on single entities of the news documents in the structured knowledge graph through a GraphSAGE graph model, and obtaining knowledge graph embedded representation vectors of the single entities. According to the method and the device, the knowledge map embedded expression vector with enhanced external knowledge is obtained by introducing the external knowledge map and the GraphSAGE map model, and the knowledge range of the entity characteristic expression vector of a single entity generated by the knowledge map embedded expression vector is further effectively enhanced.
It should be noted that the knowledge graph of the present application can be obtained from an open-source network database, or can be obtained from a knowledge graph library established by an individual according to actual needs, where the knowledge graph library includes different kinds of knowledge graphs, such as a news knowledge graph, a science and technology knowledge graph, and a medicine knowledge graph.
Correspondingly, the step of calculating the context representation vector of the single entity in the news document comprises the following steps:
identifying the number of times of the single entity appearing in the news document, the position of the single entity and the category of the single entity as the contextual characteristic representation of the single entity;
and respectively converting the times of the single entity appearing in the news document, the position of the single entity appearing and the category of the single entity into corresponding numbers based on a preset corresponding relation table to obtain the context expression vector of the single entity in the news document.
In this embodiment, the context may be different for the same single entity as it appears in the news, so its contextual characterizations should be different. Therefore, the present application selects three representative features as the contextual characterization of a single entity: total number of Times (TF) a single entity appears in a news document; the position of the single entity, specifically, the position of the single entity in the news document is the number row in the number section; categories of single entities, categories include star, sports, music, and so on. According to a preset corresponding relation table, the times of the single entity appearing in the news document, the position of the single entity appearing and the category of the single entity are respectively converted into corresponding numbers, the numbers can be input into a line function, and corresponding vectors are generated, wherein the line function refers to a vector generation instruction. The corresponding relation table stores the preset corresponding relation between the number of times of the single entity appearing in the news document and the number, the corresponding relation between the position of the single entity appearing in the news document and the number and the corresponding relation between the category of the single entity and the number, and the corresponding number can be obtained by searching the number of times of the single entity appearing in the news document, the position of the single entity appearing in the news document or the category of the single entity in the corresponding relation table. And calculating a context representation vector of the single entity in the news document through the occurrence number, the position and the category of the single entity, and representing the context representation vector. For example: the total occurrence frequency of the current single entity is 10 times, and the corresponding number is 1; and if the number corresponding to the 3 rd row in the 3 rd segment is 1 and the number corresponding to the star category in the category is 0, the context expression vector of the current single entity is (1,1, 0).
It should be noted that: the application does not limit the number, position and dimensional position of the number corresponding to the category in the converted vector, and the example is only used as a reference. In the practical application process, the corresponding dimension position can be determined according to the practical requirement.
S3: and aggregating the entity feature representation vectors of all the single entities in the news document based on the semantic feature representation vectors to obtain an aggregate representation vector.
In this embodiment, based on the initial recommendation model and the semantic feature representation vector, the entity feature representation vectors of all the single entities in the news document are aggregated to obtain an aggregated representation vector. The method and the device have the advantages that the entity feature representation vectors of all the single entities in the news documents are aggregated based on the semantic feature representation vectors, and the richness of the aggregation representation vectors is effectively improved.
Specifically, in step S3, that is, the step of aggregating the entity feature representation vectors of all the single entities in the news document based on the semantic feature representation vectors to obtain an aggregate representation vector includes:
taking the semantic feature expression vector as a query vector, and paying attention to each single entity contained in the document based on an attention mechanism set in the initial recommendation model;
respectively converting the noted entity feature representation vectors corresponding to the single entity into a key vector and a value vector;
calculating the inner product of the query vector and each key vector, and normalizing all the inner products to obtain the attention weight of the entity feature expression vector;
and carrying out weighted summation on the entity feature representation vectors based on the attention weight to obtain the aggregation representation vector.
In this embodiment, the attention mechanism of the present application is to pay attention to each single entity contained in the document by using the semantic feature representation vector obtained in step S1 as a query (query vector). The entity feature representation vector of a single entity is converted into a key vector (key vector) and a value vector (value vector). Calculating the inner product of the query vector and the key vector to obtain the attention weight; and normalizing the attention weights of all the entities to realize that the sum of all the attention weights is 1, and carrying out weighted summation on the feature expression vectors corresponding to the single entity based on the attention weights to obtain an aggregation expression vector. Compared with the prior art that the attention weight is calculated through MLP after the query vector and the key vector are spliced, the method has the advantages that the attention weight is obtained by directly calculating the inner product of the query vector and the key vector, so that the network architecture is simpler, and the network training is more efficient. Furthermore, the semantic feature representation vector is used as a query, and the entity feature representation vector is subjected to weighted summation, so that the richness of the aggregation representation vector is improved.
S4: and processing the aggregation expression vector, the semantic feature expression vector and a preset user vector to obtain a prediction result.
In this embodiment, the news document feature representation vector and a preset user vector are processed based on the initial recommendation model, where the prediction result indicates whether the news is recommended, specifically, whether the news is recommended or not recommended. The user vector mainly relates to personal information (including name, age, and the like) of the user, user interest and hobby pictures, and the like. Because the prediction result is generated based on the aggregation expression vector and the semantic feature expression vector, the recommendation dimensionality is enhanced, and the diversity and accuracy of model recommendation are effectively improved.
Specifically, the step of processing the aggregate expression vector, the semantic feature expression vector, and a preset user vector to obtain a prediction result includes:
splicing the aggregation expression vector and the semantic feature expression vector;
inputting the spliced vector into a full connection layer in the initial recommendation model to obtain a news document feature vector;
and inputting the news document feature vector and the user vector into a binary classification model of the initial recommendation model to obtain a prediction result.
In this embodiment, after obtaining the aggregate representation vector of all entities in each news document from step S3, dimension concatenation is directly performed on the vector and the original semantic feature representation vector of the corresponding news obtained in step S1, where the dimension concatenation may refer to data merging according to a specified dimension, and when the dimension concatenation is specified, it is required to ensure that values of other dimensions are the same, for example: the vectors (3,45,8) and the vectors (5,45,8) are spliced in the first dimension, the vectors after dimension splicing are (8,45,8), or the method can be used for directly adding the dimensions of the corresponding vectors. And then processing the spliced vector through a full connection layer containing the tanh activation function to obtain a news document feature vector. Through dimension splicing and neural network fusion, the feature expression capability is enhanced, and the knowledge enhancement of the feature vector of the news document is realized. The knowledge-enhanced news document feature representation vector can also be used for other downstream tasks, such as news topic classification, popularity prediction and the like. After the knowledge-enhanced news document expression vector is obtained, supervised interactive learning is carried out on the news document feature expression vector and the user vector through a two-classification model based on the click log, and a prediction result is obtained. Specifically, the click log includes the click condition of the user on the news document, and different news clicked and not clicked by the user are obtained through the click log. For the current news to be judged whether to be recommended or not, based on the news document characteristic vector, the news with the highest similarity to the news to be judged is found in the news clicked by the user, whether the user likes the news or not is known through the user vector, if the user likes the news, the fact that the user also likes the news to be judged is judged, and then the news is recommended to serve as a prediction result. And the supervised finger automatically carries out loss function calculation on the two classification models, and iteratively trains the preset model until convergence to obtain the trained two classification models. The training process of the model can be recorded, so that problems occurring in the training process of the model can be traced conveniently.
For example: the above two-classification model can be implemented by using a conventional two-classifier (e.g., LR (Logistic Regression), Xgboost (Gradient Boosting) model) or DNN (Deep Neural Network) model, determining the news clicked and not clicked by the user by clicking a diary, taking the clicked news as a positive sample, and then randomly sampling some of the rest samples from the news large disk as negative samples to train the two-classification model, or using a Factorization Machine (FM) frequently used in a recommendation system algorithm, wherein the Factorization Machine (FM) compensates two major disadvantages of the conventional two-classifier, i.e., the influence of many features on the result is implemented by joint action, the FM introduces cross features by feature combination to enhance the learning and generalization capabilities of the model, and the high-dimensional sparse matrix is decomposed to obtain a low-dimensional hidden vector, the calculation efficiency and the training convergence speed are improved.
It should be noted that, in the present application, the execution processes of S1 to S4 are executed based on the initial recommendation model, and S1 to S4 are training processes of the initial recommendation model.
S5: calculating a loss function value of the initial recommendation model based on the prediction result and the label corresponding to the news document, iteratively training the initial recommendation model until the initial recommendation model is converged, obtaining a trained recommendation model, and recommending news to a user based on the trained recommendation model.
In this embodiment, the click log is used to record a click condition of the user on news, compare whether the user clicks on the news with a corresponding tag of the news document, and compare the result with a prediction result predicted by a preset initial recommendation model, and iteratively train the initial recommendation model until the initial recommendation model converges to obtain a final recommendation model. And the model convergence refers to setting the maximum iteration times, determining the model convergence when the maximum iteration times are reached, and stopping training. The calculation process of the loss function value can adopt a 0-1 loss function, the prediction result and the label are not equal to 1, otherwise, the prediction result and the label are equal to 0, and the specific formula is as follows:
where Y represents the prediction result, and f (x) represents the tag corresponding to the news document.
Of course, according to actual needs, the loss function value of the present application may also be calculated by using other existing loss functions, and the present application is applicable.
It should be noted that: the semantic extraction model and the GraphSAGE graph model in the initial recommendation model can be pre-trained models, and parameters of the semantic extraction model and the GraphSAGE graph model do not need to be changed in the process of iteratively training the initial recommendation model.
Certainly, the semantic extraction model and the graphics SAGE graph model included in the initial recommendation model of the application are randomly initialized models, and parameters of the semantic extraction model and the graphics SAGE graph model need to be iteratively updated at the same time in the iterative training of the initial recommendation model.
The two modes are flexibly selected according to actual requirements in the actual application process.
It is emphasized that, in order to further ensure the privacy and security of the trained recommendation model, the trained recommendation model may also be stored in a node of a blockchain.
The block chain referred by the application is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism, an encryption algorithm and the like. A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like.
This application can be applied to in the wisdom government affairs field to promote the construction in wisdom city. The application can be applied to the field of medical science and technology, and the news document can be a medical science and technology news document and a medical health news document, so that the popularization degree of medical knowledge is improved.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware associated with computer readable instructions, which can be stored in a computer readable storage medium, and when executed, can include processes of the embodiments of the methods described above. The storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a Random Access Memory (RAM).
It should be understood that, although the steps in the flowcharts of the figures are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and may be performed in other orders unless explicitly stated herein. Moreover, at least a portion of the steps in the flow chart of the figure may include multiple sub-steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed alternately or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
With further reference to fig. 3, as an implementation of the method shown in fig. 2, the present application provides an embodiment of a news recommendation apparatus, which corresponds to the embodiment of the method shown in fig. 2, and which can be applied to various electronic devices.
As shown in fig. 3, the news recommender 300 according to this embodiment includes: the system comprises a receiving module 301, a calculating module 302, an aggregation module 303, a processing module 304 and an iteration module 305, wherein the receiving module 301 is configured to receive a news document, input the news document into a semantic extraction model of a preset initial recommendation model, and obtain a semantic feature expression vector output by the semantic extraction model, wherein the news document carries a tag; a calculating module 302, configured to identify a single entity from the news document, and calculate an entity feature representation vector of the single entity in the news document; the aggregation module 303 is configured to aggregate the entity feature representation vectors of all the single entities in the news document based on the semantic feature representation vector to obtain an aggregate representation vector; the processing module 304 is configured to process the aggregation representation vector, the semantic feature representation vector, and a preset user vector to obtain a prediction result; and an iteration module 305, configured to calculate a loss function value of the initial recommendation model based on the prediction result and the label corresponding to the news document, and iteratively train the initial recommendation model until the initial recommendation model converges, obtain a trained recommendation model, and recommend news to a user based on the trained recommendation model.
In the embodiment, the entity feature representation vectors of the single entity obtained from the news document are aggregated through the semantic feature representation vectors to generate the aggregation representation vectors, so that the richness of the aggregation representation vectors is improved. The aggregation expression vector, the semantic feature expression vector and the preset user vector are processed to generate a prediction result, and the prediction result is generated based on the aggregation expression vector and the semantic feature expression vector, so that the recommendation dimensionality is enhanced, the model recommendation diversity and accuracy are effectively improved, and the common problem that the user experience is reduced due to the fact that the model recommendation is more narrow is effectively avoided.
The semantic extraction model comprises a long-term and short-term memory network layer, and the receiving module 301 comprises a conversion sub-module, an extraction sub-module and a weighting sub-module. The conversion submodule is used for converting each word in the news document into a word vector based on the semantic extraction model; the extraction submodule is used for extracting all word vectors in each sentence based on a long-short term memory network layer in the semantic extraction model, and carrying out weighted summation on the word vectors in each sentence through an attention mechanism in the semantic extraction model to generate sentence characteristic vectors; and the weighting submodule is used for carrying out weighted summation on the sentence vectors through an attention mechanism in the semantic extraction model to generate document feature vectors serving as the semantic feature expression vectors.
The calculation module 302 includes a calculation sub-module and an obtaining sub-module. The calculation submodule is used for respectively calculating a knowledge map embedded expression vector of a single entity and a context expression vector of the single entity in a news document; and the obtaining submodule is used for adding the knowledge map embedded expression vector of the single entity and the context expression vector of the single entity in the news document in corresponding dimensions to obtain the entity feature expression vector of the single entity.
In some optional implementation manners of this embodiment, the calculation sub-module is further configured to obtain a knowledge graph, and perform knowledge representation learning on a single entity based on the knowledge graph and a graph model in the initial recommendation model to obtain a knowledge graph embedded representation vector of the single entity.
The calculation submodule comprises a recognition unit and a conversion unit. The identification unit is used for identifying the number of times of the single entity appearing in the news document, the position of the single entity and the category of the single entity as the context feature representation of the single entity; the conversion unit is used for respectively converting the times of the single entity appearing in the news document, the position of the single entity appearing and the category of the single entity into corresponding numbers based on a preset corresponding relation table to obtain the context expression vector of the single entity in the news document.
The aggregation module 303 includes an attention sub-module, a conversion sub-module, a normalization sub-module, and a summation sub-module. The attention submodule is used for taking the semantic feature expression vector as a query vector and paying attention to each single entity contained in the document based on an attention machine set in the initial recommendation model; the conversion module is used for converting the noted entity feature representation vectors corresponding to the single entities into key vectors and value vectors respectively; the normalization submodule is used for calculating the inner product of the query vector and each key vector, normalizing all the inner products and obtaining the attention weight of the entity feature expression vector; and the summation submodule is used for carrying out weighted summation on the entity feature representation vectors based on the attention weight to obtain the aggregation representation vector.
The processing module 304 includes a splicing sub-module, a fully connected layer sub-module, and a processing sub-module. The splicing submodule is used for splicing the aggregation expression vector and the semantic feature expression vector; the full-connection layer submodule is used for inputting the spliced vector into a full-connection layer in the initial recommendation model to obtain a news document feature vector; and the processing submodule is used for inputting the news document feature vector and the user vector into the two classification models of the initial recommendation model to obtain a prediction result.
According to the method and the device, the entity feature representation vectors of the single entity obtained from the news document are aggregated through the semantic feature representation vectors, the aggregated representation vectors are generated, and the richness of the aggregated representation vectors is improved. The aggregation expression vector, the semantic feature expression vector and the preset user vector are processed to generate a prediction result, and the prediction result is generated based on the aggregation expression vector and the semantic feature expression vector, so that the recommendation dimensionality is enhanced, the model recommendation diversity and accuracy are effectively improved, and the common problem that the user experience is reduced due to the fact that the model recommendation is more narrow is effectively avoided.
In order to solve the technical problem, an embodiment of the present application further provides a computer device. Referring to fig. 4, fig. 4 is a block diagram of a basic structure of a computer device according to the present embodiment.
The computer device 200 comprises a memory 201, a processor 202, a network interface 203 communicatively connected to each other via a system bus. It is noted that only computer device 200 having components 201 and 203 is shown, but it is understood that not all of the illustrated components are required and that more or fewer components may alternatively be implemented. As will be understood by those skilled in the art, the computer device is a device capable of automatically performing numerical calculation and/or information processing according to a preset or stored instruction, and the hardware includes, but is not limited to, a microprocessor, an Application Specific Integrated Circuit (ASIC), a Programmable Gate Array (FPGA), a Digital Signal Processor (DSP), an embedded device, and the like.
The computer device can be a desktop computer, a notebook, a palm computer, a cloud server and other computing devices. The computer equipment can carry out man-machine interaction with a user through a keyboard, a mouse, a remote controller, a touch panel or voice control equipment and the like.
The memory 201 includes at least one type of readable storage medium including a flash memory, a hard disk, a multimedia card, a card type memory (e.g., SD or DX memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a Read Only Memory (ROM), an Electrically Erasable Programmable Read Only Memory (EEPROM), a Programmable Read Only Memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the storage 201 may be an internal storage unit of the computer device 200, such as a hard disk or a memory of the computer device 200. In other embodiments, the memory 201 may also be an external storage device of the computer device 200, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), or the like, provided on the computer device 200. Of course, the memory 201 may also include both internal and external storage devices of the computer device 200. In this embodiment, the memory 201 is generally used for storing an operating system installed in the computer device 200 and various application software, such as computer readable instructions of a news recommendation method. Further, the memory 201 may also be used to temporarily store various types of data that have been output or are to be output.
The processor 202 may be a Central Processing Unit (CPU), controller, microcontroller, microprocessor, or other data Processing chip in some embodiments. The processor 202 is generally operative to control overall operation of the computer device 200. In this embodiment, the processor 202 is configured to execute computer-readable instructions stored in the memory 201 or process data, for example, execute computer-readable instructions of the news recommendation method.
The network interface 203 may comprise a wireless network interface or a wired network interface, and the network interface 203 is generally used for establishing communication connection between the computer device 200 and other electronic devices.
In the embodiment, the dimensionality of recommending the news by the model is enhanced, the diversity and the accuracy of model recommendation are effectively improved, and the common problem that the user experience is reduced due to the fact that the model recommendation is more narrow and more narrow is effectively avoided.
The present application further provides another embodiment, which is to provide a computer-readable storage medium storing computer-readable instructions executable by at least one processor to cause the at least one processor to perform the steps of the news recommendation method as described above.
In the embodiment, the dimensionality of recommending the news by the model is enhanced, the diversity and the accuracy of model recommendation are effectively improved, and the common problem that the user experience is reduced due to the fact that the model recommendation is more narrow and more narrow is effectively avoided.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present application may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal device (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present application.
It is to be understood that the above-described embodiments are merely illustrative of some, but not restrictive, of the broad invention, and that the appended drawings illustrate preferred embodiments of the invention and do not limit the scope of the invention. This application is capable of embodiments in many different forms and is provided for the purpose of enabling a thorough understanding of the disclosure of the application. Although the present application has been described in detail with reference to the foregoing embodiments, it will be apparent to one skilled in the art that the present application may be practiced without modification or with equivalents of some of the features described in the foregoing embodiments. All equivalent structures made by using the contents of the specification and the drawings of the present application are directly or indirectly applied to other related technical fields and are within the protection scope of the present application.

Claims (10)

1. A news recommendation method, comprising the steps of:
receiving a news document, inputting the news document into a semantic extraction model of a preset initial recommendation model, and obtaining a semantic feature expression vector output by the semantic extraction model, wherein the news document carries a tag;
identifying single entities from the news documents, and calculating entity feature expression vectors of the single entities in the news documents;
based on the semantic feature representation vectors, aggregating the entity feature representation vectors of all the single entities in the news document to obtain an aggregate representation vector;
processing the aggregation expression vector, the semantic feature expression vector and a preset user vector to obtain a prediction result;
calculating a loss function value of the initial recommendation model based on the prediction result and the label corresponding to the news document, iteratively training the initial recommendation model until the initial recommendation model is converged, obtaining a trained recommendation model, and recommending news to a user based on the trained recommendation model.
2. A news recommendation method according to claim 1, wherein the step of calculating an entity feature representation vector for a single entity in the news document comprises:
respectively calculating a knowledge map embedded expression vector of a single entity and a context expression vector of the single entity in a news document;
and adding the knowledge graph embedding expression vector of the single entity and the context expression vector of the single entity in the news document in corresponding dimensions to obtain an entity feature expression vector of the single entity.
3. A news recommendation method according to claim 2, wherein the step of computing a knowledge-graph embedded representation vector for a single entity comprises:
and acquiring a knowledge graph, and performing knowledge representation learning on a single entity based on the knowledge graph and a graph model in the initial recommendation model to obtain a knowledge graph embedded representation vector of the single entity.
4. A news recommendation method according to claim 2 or 3, wherein the step of calculating a context representation vector of a single entity in a news document comprises:
identifying the number of times that a single entity appears in a news document, the position where the single entity appears and the category of the single entity;
and respectively converting the times of the single entity appearing in the news document, the position of the single entity appearing and the category of the single entity into corresponding numbers based on a preset corresponding relation table to obtain the context expression vector of the single entity in the news document.
5. A news recommendation method according to claim 1, wherein said step of aggregating entity feature representation vectors of all single entities in a news document based on said semantic feature representation vectors to obtain an aggregate representation vector comprises:
taking the semantic feature expression vector as a query vector, and paying attention to each single entity contained in the document based on an attention mechanism set in the initial recommendation model;
respectively converting the noted entity feature representation vectors corresponding to the single entity into a key vector and a value vector;
calculating the inner product of the query vector and each key vector, and normalizing all the inner products to obtain the attention weight of the entity feature expression vector;
and carrying out weighted summation on the entity feature representation vectors based on the attention weight to obtain the aggregation representation vector.
6. A news recommendation method according to claim 1, wherein said step of processing said aggregate representation vector, semantic feature representation vector and preset user vector to obtain prediction result comprises:
splicing the aggregation expression vector and the semantic feature expression vector;
inputting the spliced vector into a full connection layer in the initial recommendation model to obtain a news document feature vector;
and inputting the news document feature vector and the user vector into a binary classification model of the initial recommendation model to obtain a prediction result.
7. The news recommendation method according to claim 1, wherein the semantic extraction model includes a long-short term memory network layer, and the step of inputting the news document into a preset semantic extraction model to obtain the semantic feature expression vector output by the semantic extraction model includes:
converting each word in the news document into a word vector based on a semantic extraction model;
extracting all word vectors in each sentence based on a long-short term memory network layer in a semantic extraction model, and performing weighted summation on the word vectors in each sentence through an attention mechanism in the semantic extraction model to generate sentence characteristic vectors;
and carrying out weighted summation on the sentence vectors through an attention mechanism in a semantic extraction model to generate document feature vectors serving as the semantic feature expression vectors.
8. A news recommender, comprising:
the receiving module is used for receiving a news document, inputting the news document into a semantic extraction model of a preset initial recommendation model, and obtaining a semantic feature expression vector output by the semantic extraction model, wherein the news document carries a tag;
the calculation module is used for identifying single entities from the news documents and calculating entity feature expression vectors of the single entities in the news documents;
the aggregation module is used for aggregating the entity feature representation vectors of all the single entities in the news document based on the semantic feature representation vectors to obtain an aggregate representation vector;
the processing module is used for processing the aggregation expression vector, the semantic feature expression vector and a preset user vector to obtain a prediction result; and
and the iteration module is used for calculating a loss function value of the initial recommendation model based on the prediction result and the label corresponding to the news document, iteratively training the initial recommendation model until the initial recommendation model is converged to obtain a trained recommendation model, and recommending news to a user based on the trained recommendation model.
9. A computer device comprising a memory having computer readable instructions stored therein and a processor that when executed performs the steps of the news recommendation method of any one of claims 1-7.
10. A computer-readable storage medium having computer-readable instructions stored thereon which, when executed by a processor, implement the steps of the news recommendation method of any one of claims 1-7.
CN202011148692.6A 2020-10-23 2020-10-23 News recommendation method and device, computer equipment and storage medium Pending CN112231569A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011148692.6A CN112231569A (en) 2020-10-23 2020-10-23 News recommendation method and device, computer equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011148692.6A CN112231569A (en) 2020-10-23 2020-10-23 News recommendation method and device, computer equipment and storage medium

Publications (1)

Publication Number Publication Date
CN112231569A true CN112231569A (en) 2021-01-15

Family

ID=74110474

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011148692.6A Pending CN112231569A (en) 2020-10-23 2020-10-23 News recommendation method and device, computer equipment and storage medium

Country Status (1)

Country Link
CN (1) CN112231569A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113222328A (en) * 2021-03-25 2021-08-06 中国科学技术大学先进技术研究院 Air quality monitoring equipment point arrangement and site selection method based on road section pollution similarity
CN113327691A (en) * 2021-06-01 2021-08-31 平安科技(深圳)有限公司 Query method and device based on language model, computer equipment and storage medium

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113222328A (en) * 2021-03-25 2021-08-06 中国科学技术大学先进技术研究院 Air quality monitoring equipment point arrangement and site selection method based on road section pollution similarity
CN113222328B (en) * 2021-03-25 2022-02-25 中国科学技术大学先进技术研究院 Air quality monitoring equipment point arrangement and site selection method based on road section pollution similarity
CN113327691A (en) * 2021-06-01 2021-08-31 平安科技(深圳)有限公司 Query method and device based on language model, computer equipment and storage medium
CN113327691B (en) * 2021-06-01 2022-08-12 平安科技(深圳)有限公司 Query method and device based on language model, computer equipment and storage medium

Similar Documents

Publication Publication Date Title
US11288593B2 (en) Method, apparatus and device for extracting information
CN112231569A (en) News recommendation method and device, computer equipment and storage medium
Lavanya et al. Twitter sentiment analysis using multi-class SVM
US20210042476A1 (en) Intelligent Routing Services and Systems
CN111090987A (en) Method and apparatus for outputting information
CN111625715B (en) Information extraction method and device, electronic equipment and storage medium
Suhaili et al. Service chatbots: A systematic review
CN111400584A (en) Association word recommendation method and device, computer equipment and storage medium
CN112214601A (en) Social short text sentiment classification method and device and storage medium
Kim et al. SAO2Vec: Development of an algorithm for embedding the subject–action–object (SAO) structure using Doc2Vec
CN110827115A (en) Attention mechanism-based product recommendation method and system
Salami et al. Recurrent convolutional neural networks for poet identification
US11281860B2 (en) Method, apparatus and device for recognizing text type
CN113987169A (en) Text abstract generation method, device and equipment based on semantic block and storage medium
WO2021135455A1 (en) Semantic recall method, apparatus, computer device, and storage medium
CN112199954A (en) Disease entity matching method and device based on voice semantics and computer equipment
Li et al. RETRACTED: Multi-neural network-based sentiment analysis of food reviews based on character and word embeddings
Hajek et al. Opinion mining of consumer reviews using deep neural networks with word-sentiment associations
US10915756B2 (en) Method and apparatus for determining (raw) video materials for news
US10579655B2 (en) Method and apparatus for compressing topic model
Oliveira et al. Sentiment analysis of stock market behavior from Twitter using the R Tool
CN113505601A (en) Positive and negative sample pair construction method and device, computer equipment and storage medium
CN114357117A (en) Transaction information query method and device, computer equipment and storage medium
CN114385814A (en) Information retrieval method and device, computer equipment and storage medium
CN112307738A (en) Method and device for processing text

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination