Article classification method and device, equipment and medium

Abstract

The invention provides an article classification method and device, equipment and a medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining a plurality of to-be-classified field articles, determining a first vector of each field article of a first type through a network model according to the content identifier and content information of each field article, and according to the feature vector of each domain article, matching with a known domain article vector of a second type in a pre-constructed vector space model, determining a second vector of the domain article of the second type, and according to the first vector and the second vector, classifying the plurality of domain articles to be classified. According to the technical scheme of the invention, the method achieves the targeted recognition of the domain articles of the corresponding types through the model and vector matching, achieves the automatic marking, finally carries out the classification, solves the problems of low efficiency and inconsistency of manual one-by-one marking, and improves the marking expansibility and accuracy.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to an article classification method, apparatus, device, and medium. Background Art

[0002] With the development of Internet technology, a large number of articles and information in different fields have been published on various platforms. Articles in each field cover various aspects of the content, and the number of articles is huge. However, when users search and browse these articles, they often face the problems of information overload and inaccuracy. Therefore, it is necessary to add tags to these articles for classification.

[0003] Related technologies rely primarily on manual labeling and classification, which is subjective and inconsistent. Different people may assign different labels to the same article, requiring a high level of expertise. Furthermore, as the number of articles increases, the manual labeling workload will grow exponentially, making it difficult to meet the rapidly growing demand for knowledge in various fields. Consequently, manual labeling suffers from low efficiency, poor consistency, and poor scalability.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the Invention

[0005] The purpose of the present disclosure is to provide a method that, at least to a certain extent, solves the problems of low efficiency, poor consistency, and poor scalability in tagging field articles in the prior art.

[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by practice of the present disclosure.

[0007] According to a first aspect of the present disclosure, an embodiment of the present disclosure provides an article classification method, the method comprising:

[0008] Obtain multiple field articles to be classified; the field articles belong to the same preset field;

[0009] Determining a first vector for each domain article of the first type through a network model based on the content identifier and content information of each domain article; the first vector is used to represent an attribute or state of a domain entity in the domain article;

[0010] According to the feature vector of each domain article, the feature vector is matched with the known second type of domain article vectors in the pre-built vector space model to determine the second vector of the second type of domain article; the second vector is used to represent the event or behavior in the domain article;

[0011] A plurality of articles in a field to be classified are classified according to the first vector and the second vector.

[0012] In a possible embodiment, the field articles include: medical articles in the medical field; the first types include: department type, disease type, and index type; the second types include: drug type, symptom type, category type, and tag type;

[0013] The method further comprises:

[0014] Obtain multiple medical articles to be classified;

[0015] According to the content identifier and content information of each medical article, the department vector of the medical article of department type, the disease vector of the medical article of disease type, and the index vector of the medical article of index type are determined through the network model;

[0016] Based on the feature vector of each medical article, the feature vector is matched with the medical article vectors of known drug type, symptom type, category type, and label type in the pre-built vector space model to determine the drug vector, symptom vector, category vector, and label vector of the medical article of drug type, symptom type, category type, and label type respectively;

[0017] The plurality of medical articles to be classified are classified according to the department vector, disease vector, index vector, drug vector, symptom vector, category vector and label vector.

[0018] In a possible embodiment, determining the first vector of each field article of the first type through a network model based on the content identifier and content information of each field article includes:

[0019] Inputting the plurality of field articles into a generative pre-training model;

[0020] Inputting instruction information for identifying a first type of domain article into the generative pre-training model, so that the generative pre-training model identifies the first type of domain article based on the content identifier and content information of each domain article;

[0021] Adding a label vector to be verified for the first type of field articles;

[0022] The tag vector to be checked is checked to obtain a first vector.

[0023] In a possible embodiment, verifying the tag vector to be verified to obtain the first vector includes:

[0024] Determining a first similarity based on the to-be-verified label vector and the first type of domain article vectors stored in the label vector library;

[0025] Verifying the first vector according to the first similarity;

[0026] If the first similarity is greater than the first threshold, the verification is passed, and the passed label vector to be verified is used as the first vector of the first type of field article.

[0027] In a possible embodiment, matching the feature vector of each domain article with known vectors of second-type domain articles in a pre-built vector space model to determine the second vector of the second-type domain article includes:

[0028] Determining a second similarity between the feature vector of each field article and the known field article vector of the second type;

[0029] For any feature vector of a field article to be classified, if the corresponding second similarity is greater than a second threshold, the type corresponding to the field article to be classified is determined to be the second type, and the second vector of the field article to be classified is determined.

[0030] In a possible embodiment, constructing a vector space model includes:

[0031] Obtain domain articles with known labels and determine the known label vectors corresponding to domain articles with known label vectors;

[0032] According to a vector matching algorithm, the field articles with the known label vectors are matched, and classifications are constructed to obtain a vector space model; wherein each classification corresponds to a type.

[0033] In a possible embodiment, obtaining multiple field articles includes:

[0034] In response to the user's indication of the classification of the domain articles, under the batch processing architecture, the existing domain articles are read in batches from the existing database in a declarative input and output manner as multiple domain articles to be classified.

[0035] In a possible embodiment, the method further includes:

[0036] Pre-process the field articles to obtain content identification and content information;

[0037] Using the content identifier as a unique identifier, natural language processing technology is used to generate vectors for the content information, and the feature vectors of the obtained field articles are stored in the existing vector database;

[0038] Among them, after obtaining multiple field articles, it also includes:

[0039] The content identifier and content information of each field article are obtained, and the feature vector of each field article is obtained from the stock vector database.

[0040] In a possible embodiment, pre-processing the field articles to obtain content identifiers and content information includes:

[0041] Screening the articles in the field, excluding the test data existing in the articles in the field, and obtaining the screened articles in the field;

[0042] Deleting the markup language tags and blank lines in the data paragraphs corresponding to the articles in the screening field to obtain the deleted field articles;

[0043] Aggregate the deleted domain articles, extract the content identifier and content information in each deleted domain article into an object, and obtain the content identifier and content information of the domain article.

[0044] In a possible embodiment, before obtaining a plurality of field articles, the method further includes:

[0045] receiving an incremental domain article through a message queue; the incremental domain article and the domain article are in the same preset domain;

[0046] Preprocessing and natural language processing technology are performed on the incremental field articles to obtain content identification, content information and corresponding feature vectors corresponding to the incremental field articles;

[0047] Storing the feature vectors corresponding to the incremental field articles in the stock vector database;

[0048] The content identifiers and content information corresponding to the incremental field articles are stored in the stock database and classified according to the stock field articles.

[0049] In a possible embodiment, after determining the first vector and the second vector, the method further includes:

[0050] Determining a marking result according to the first vector and the second vector;

[0051] The formal data and draft data of the marking results are stored in the stock database and the cache database respectively;

[0052] Different management permissions are set for the official data and the draft data.

[0053] In a possible embodiment, after storing the official data and the draft data of the marking results in the stock database and the cache database respectively, the method further includes:

[0054] In response to a user's query request for field articles, the marking results are preferentially extracted from the cache database and displayed to the user.

[0055] In a possible embodiment, the method further includes:

[0056] Construct a marking record sheet;

[0057] In the process of classifying the field articles, the marking status of the field articles is recorded in the marking status record table.

[0058] According to another aspect of the present disclosure, an embodiment of the present disclosure provides an article classification device, comprising:

[0059] An acquisition unit, configured to acquire a plurality of field articles to be classified; the field articles belong to the same preset field;

[0060] A first label adding unit is configured to determine a first vector of each domain article of the first type through a network model based on the content identifier and content information of each domain article; the first vector is used to represent an attribute or state of a domain entity in the domain article;

[0061] A second label adding unit is configured to match the feature vector of each domain article with a known second type of domain article vector in a pre-built vector space model to determine a second vector of the second type of domain article; the second vector is used to represent an event or behavior in the domain article;

[0062] A classification unit is used to classify multiple articles in the field to be classified according to the first vector and the second vector.

[0063] In a possible embodiment, the article classification device further includes a preprocessing unit configured to:

[0064] Pre-process the field articles to obtain content identification and content information;

[0065] Using the content identifier as a unique identifier, natural language processing technology is used to generate vectors for the content information, and the feature vectors of the obtained field articles are stored in the existing vector database;

[0066] Among them, after obtaining multiple field articles, it also includes:

[0067] The content identifier and content information of each field article are obtained, and the feature vector of each field article is obtained from the stock vector database.

[0068] In a possible embodiment, the article classification device further includes an incremental article processing unit configured to:

[0069] receiving an incremental domain article through a message queue; the incremental domain article and the domain article are in the same preset domain;

[0070] Preprocessing and natural language processing technology are performed on the incremental field articles to obtain content identification, content information and corresponding feature vectors corresponding to the incremental field articles;

[0071] Storing the feature vectors corresponding to the incremental field articles in the stock vector database;

[0072] The content identifiers and content information corresponding to the incremental field articles are stored in the stock database and classified according to the stock field articles.

[0073] In a possible embodiment, the article classification device further includes: a marking result management unit, configured to:

[0074] Determine a marking result according to the first vector and the second vector;

[0075] The formal data and draft data of the marking results are stored in the stock database and the cache database respectively;

[0076] Different management permissions are set for the official data and the draft data.

[0077] In a possible embodiment, the marking result management unit is further used to:

[0078] In response to a user's query request for field articles, the marking results are preferentially extracted from the cache database and displayed to the user.

[0079] In a possible embodiment, the article classification device further includes a real-time monitoring unit configured to:

[0080] Construct a marking record sheet;

[0081] In the process of classifying the field articles, the marking status of the field articles is recorded in the marking status record table.

[0082] According to another aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform any one of the methods in the first aspect by executing the executable instructions.

[0083] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method according to any one of the first aspects is implemented.

[0084] The embodiments of the present disclosure provide an article classification method, apparatus, device, and medium, the method comprising: obtaining multiple domain articles to be classified; the domain articles belong to the same preset domain; determining a first vector for each domain article of a first type through a network model based on the content identifier and content information of each domain article; the first vector is used to represent the attributes or status of the domain entity in the domain article; matching the feature vector of each domain article with a known second type of domain article vector in a pre-constructed vector space model to determine a second vector for the second type of domain article; the second vector is used to represent the event or behavior in the domain article; and classifying the multiple domain articles to be classified based on the first vector and the second vector. The corresponding type of domain articles are specifically identified and automatically labeled using both model and vector matching methods. The model is used to identify and label vectors representing the attributes or status of domain entities, with good consistency, high efficiency in labeling and classification, and guaranteed accuracy. Labeling and classifying vectors representing events or behaviors through vector matching can also solve the problems of low efficiency and inconsistency in manual labeling, thereby improving the scalability and accuracy of labeling.

[0085] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] The accompanying drawings are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification, are used to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0087] Figure 1 A schematic diagram showing the structure of an article classification system according to an embodiment of the present disclosure is shown;

[0088] Figure 2 A schematic diagram showing the structure of an article classification system according to an embodiment of the present disclosure is shown;

[0089] Figure 3 A schematic diagram showing a flow chart of an article classification method in an embodiment of the present disclosure;

[0090] Figure 4 A schematic diagram of a generative pre-training model labeling process in an embodiment of the present disclosure is shown;

[0091] Figure 5 A schematic diagram showing a process of marking in a vector matching manner according to an embodiment of the present disclosure is shown;

[0092] Figure 6A schematic diagram showing a process of classifying medical articles according to an embodiment of the present disclosure is shown;

[0093] Figure 7 A schematic diagram showing a process of constructing a vector space model in an embodiment of the present disclosure;

[0094] Figure 8 A schematic structural diagram of an article classification device according to an embodiment of the present disclosure is shown;

[0095] Figure 9 A schematic structural diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0096] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0097] In addition, the accompanying drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0098] Because related technologies rely primarily on manual labor for labeling and classification, processing capabilities are severely insufficient and scalability is poor. Furthermore, manual processing is highly subjective, with inconsistent results and high requirements for personnel expertise.

[0099] Based on this, the present disclosure provides an article classification method, device, equipment and medium, which includes: obtaining multiple field articles to be classified; the field articles belong to the same preset field; determining the first vector of each field article of the first type through a network model based on the content identifier and content information of each field article; the first vector is used to characterize the attributes or status of the field entity in the field article; matching the feature vector of each field article with the known second type of field article vector in the pre-constructed vector space model to determine the second vector of the second type of field article; the second vector is used to characterize the event or behavior in the field article; and classifying the multiple field articles to be classified based on the first vector and the second vector. Through the two methods of model and vector matching, the corresponding type of field articles are targetedly identified and automatically marked. The vectors representing the attributes or status of the field entity are identified and marked by the model, with good consistency, high efficiency of marking and classification, and guaranteed accuracy. The vectors representing events or behaviors are marked and classified by vector matching, which can also solve the problems of low efficiency and inconsistency of manual marking one by one, and improve the scalability and accuracy of marking.

[0100] The solutions provided by the embodiments of the present disclosure involve technologies such as article classification methods, which are specifically described through the following embodiments:

[0101] Figure 1 The schematic diagram of an article classification system structure applicable to the article classification method of the embodiment of the present disclosure is shown, which includes: a plurality of terminals 120 and a server cluster 140 .

[0102] The terminal 120 may be a mobile terminal such as a mobile phone, a game console, a tablet computer, an e-book reader, smart glasses, an MP4 (Moving Picture Experts Group Audio Layer IV) player, a smart home device, an AR (Augmented Reality) device, a VR (Virtual Reality) device, or the terminal 120 may also be a personal computer (PC), such as a laptop computer and a desktop computer.

[0103] The terminal 120 is used to deploy the front-end system, and the user can process through the front-end interface of the front-end system.

[0104] The terminal 120 is connected to the server cluster 140 via a communication network. Optionally, the communication network is a wired network or a wireless network.

[0105] Server cluster 140 is a server, or a combination of multiple servers, or a virtualization platform, or a cloud computing service center. Server cluster 140 is used to provide background services for the application that provides the virtual chat method. Optionally, server cluster 140 performs primary computing tasks, and terminal 120 performs secondary computing tasks; alternatively, server cluster 140 performs secondary computing tasks, and terminal 120 performs primary computing tasks; alternatively, a distributed computing architecture is used between terminal 120 and server cluster 140 for collaborative computing. In this disclosure, the primary computing tasks are performed by the server.

[0106] Optionally, the application clients installed in different terminals 120 are the same, or the application clients installed on the two terminals 120 are clients of the same type of application on different control system platforms. Based on the different terminal platforms, the specific form of the application client may also be different. For example, the application client may be a mobile phone client, a PC client, or a World Wide Web (Web) client.

[0107] Those skilled in the art will appreciate that the number of the terminals 120 may be greater or less. For example, there may be only one terminal, or there may be dozens, hundreds, or even more terminals. The embodiments of this disclosure do not limit the number and device types of the terminals.

[0108] The backend management system may be deployed in the server 140 .

[0109] Optionally, the wireless network or wired network described above uses standard communication technologies and / or protocols. The network is typically the Internet, but may also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a private network, or any combination of a virtual private network. In some embodiments, technologies and / or formats including Hypertext Markup Language (HTML), Extensible Markup Language (XML), etc. are used to represent data exchanged over the network. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), and Internet Protocol Security (IPsec) may be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies may be used to replace or supplement the above-mentioned data communication technologies.

[0110] The development environment of the article classification method in the present disclosure can be a computer, and the operating environment can be Linux, which is an operating system.

[0111] Furthermore, Figure 2 FIG. 4 shows a structural diagram of another article classification system that can be applied to the article classification method in the embodiment of the present disclosure; FIG. Figure 2 As shown, it includes a front-end system 210 / back-end management system 220, a service capability module 230 and a service capability module 230.

[0112] The architecture disclosed herein is based on a batch processing architecture as its core, and various service capabilities and components are configured, specifically a Spring Batch architecture.

[0113] Front-end system 210 provides a front-end page that may include at least the following functions: Article search and display: Provides the function of searching for articles in a certain field and displays the search results to the user in the form of a list or card. Article classification and tag display: Provides the function of classifying articles and displays the results of automated classification and tagging to the user, allowing users to understand the category or tag to which each article belongs.

[0114] The backend management system 220 manages the system's configuration and content. This includes at least the following functions: User management: Managing user accounts, including adding, deleting, and modifying user information, and setting permissions for different users. Article management: Managing domain article data, including adding, editing, deleting, and reviewing articles. Tag management: Managing article tags, including adding, editing, and deleting tags.

[0115] Furthermore, the user inputs a request in the front-end system 210 and calls various service capabilities in the service capability module 230 through the interface gateway, such as: (1) Preprocessing, which may specifically include: content cleaning, screening and aggregation, etc. (2) Content tagging. Add tags to domain articles. (3) Message Queue (MQ) consumption. Get newly added domain articles from the message queue. (4) Content vector service. (5) Tag vector service. (6) Tagging statistics, statistics and records of tagging status. (7) Idempotent tagging. (8) Tag source service, which can mark the source of the tag. (9) Tag classification, classify the tags. (10) Tag completion. When processing the data set, predict and fill in the tags for some samples.

[0116] Idempotence means that a single operation has the same effect as multiple operations. Idempotent tagging means that no matter how many times the same tagging operation is performed, the result remains the same, without changing the original business state. During automated tagging and classification, idempotent tagging ensures data consistency and accuracy.

[0117] Furthermore, through the Spring Batch architecture, Figure 2 The batch processing engine in the Spring Batch architecture can be understood as the batch processing core, and the infrastructure module 240 can be understood as the batch processing infrastructure.

[0118] The batch processing engine reads and writes data to various clusters in the infrastructure module 240. These clusters may include MySQL (My Structured Query Language) clusters, Elasticsearch (ES) cache clusters, and message queue clusters. MySQL is an open-source relational database management system. The databases in the MySQL cluster can be considered stock databases, storing existing domain articles.

[0119] Batch processing engines process data in batches. This means that after reading data in batches, they process the data in the current batch immediately, and then process the next batch after that batch. After each batch is read, it can be processed by multiple processes, or multiple processes can be used, each of which processes the data in batches.

[0120] In a possible embodiment, in the case of content data changes, for example, when there are new domain articles, the data is read through the MQ cluster reader and saved in the database of the Mysql cluster, so that the new incremental domain articles are converted into existing domain articles, and the existing domain articles are obtained in batches as domain articles to be classified.

[0121] Data is read from the MySQL cluster and cache cluster and stored after processing. After the field article classification process is completed, the labeling results of the classification process can be stored separately. The official data and draft data of the labeling results can be stored separately in the inventory database and cache database, and different management permissions can be set for the official data and draft data. For users with only browsing or review permissions, only the draft data can be displayed, and for users with editing permissions, the official data can be provided.

[0122] Furthermore, the Generative Pre-trained Transformer (GPT) gateway can be used to call the Natural Language Processing (NLP) service module 250. Through the NLP service, services such as vector query, similarity query, vector generation, and GPT tagging can be called. These services can be called by configuring them as interfaces to process domain articles.

[0123] Vector query is used to query content vectors based on unique identifiers.

[0124] Similarity queries are used to query the similarity between articles in a certain field, and between articles in a certain field and tags. The similarity between articles in a certain field and tags is used for labeling during vector matching.

[0125] Vector generation is used to generate feature vectors of domain articles, as well as pre-process domain articles with known labels to obtain domain article vectors, etc.

[0126] Among them, the Spring Batch architecture can support core functions and features such as transaction management, block-partition-based processing, declarative input and output IO, flexible control of starting, stopping and restarting the data processing process, retry and skip.

[0127] The core functions and features supported by the above architecture enable domain articles to be labeled and classified in batches, blocks, and partitions during data processing. This can be flexibly started, stopped, and restarted based on user control. For processing failures, for example, domain articles that failed to be labeled can be retried, and can be skipped if the number of failures exceeds the failure threshold.

[0128] Process in blocks, for example, if there are 100 pieces of data, 10 pieces in each block, and process 10 times in a loop.

[0129] Through the spring batch architecture and declarative IO, various operation parameters can be specified at the beginning of the article classification process, such as where to start reading data, how much to read each time, where to write data, how much to write each time, etc.

[0130] Operational parameters can be dynamically configured. That is, when designing the system, some performance-related configuration parameters can be set to be dynamically adjustable. During the specific article classification process, system performance can be controlled by modifying configuration parameters. For example, by passing data to be processed via SQL, the amount and range of data to be processed each time can be dynamically controlled, as can the amount of data to be processed in each batch, and the thread pool size.

[0131] During dynamic adjustment, operating parameters can be automatically or manually reset based on the duration of each task, the amount of data processed, system memory, and CPU usage. The next data processing cycle is then repeated, and through continuous practice, CPU and memory usage are maintained at the highest available and stable levels.

[0132] This batch data processing approach ensures that even with large data volumes, performance bottlenecks are eliminated, and normal business queries are not impacted, resulting in minimal performance loss. Furthermore, operational parameters can be dynamically expanded and adjusted during processing to ensure stable performance.

[0133] It should be noted that the architecture for implementing the above functions in related technologies generally uses the MVC (Model-View-Controller) architecture to develop software through the business (Service) layer and the data access (Dao) layer. The embodiment of the present disclosure adopts the Spring Batch architecture, and configures the operation parameters based on the declarative IO method. The domain articles are processed in batches, partitions and blocks, and various operation parameters can be dynamically adjusted to ensure that the processing capacity is maintained at the highest and the performance stability is maintained. In the application scenario of domain article classification, the embodiment of the present disclosure will not cause performance bottlenecks due to the large amount of data for a large number of domain articles, which may involve tens of millions of articles. It can also automatically and accurately mark and classify them.

[0134] In the following content, the article classification method is described through the embodiments of the present disclosure, and is specifically described through the following embodiments.

[0135] Figure 3FIG. 1 is a flow chart showing a method for classifying articles in an embodiment of the present disclosure. Figure 3 As shown, the following steps are included:

[0136] S302: Obtain multiple articles in the field to be classified.

[0137] Among them, the domain articles belong to the same preset domain, and the domain articles to be classified are all domain articles without labels.

[0138] In a possible embodiment, in response to a user's indication of the classification of a domain article, under a batch processing architecture, in a declarative input and output manner, existing domain articles are read in batches from an existing database as a plurality of domain articles to be classified.

[0139] In a possible embodiment, the existing field articles in the present disclosure can be processed in advance, and the information needed can be directly obtained during the classification process, for example, the content identifier (content_id) and content information of each field article to be classified can be obtained, and the feature vector of each field article to be classified can be obtained from the existing vector database.

[0140] The content information of the domain article may include a content identifier (content_name) and a content paragraph (paragraph_content).

[0141] The pre-processing process of the above information includes: pre-processing the field articles to obtain content identification and content information; using the content identification as a unique identifier, generating vectors for the content information through natural language processing technology, and obtaining the feature vectors of the field articles and storing them in the stock vector database.

[0142] Exemplarily, the preprocessing process for domain articles in the present disclosure may include: screening the domain articles, excluding the test data existing in the domain articles, and obtaining the screened domain articles; deleting the markup language tags and blank lines in the data paragraphs corresponding to the screened domain articles, and obtaining the deleted domain articles; aggregating the deleted domain articles, extracting the content identifier and content information in each deleted domain article into an object, and obtaining the content identifier and content information of the domain articles.

[0143] Exemplarily, the process of generating a feature vector may include: using the content identifier as a unique identifier (key), embedding the content identifier and content paragraph through natural language processing technology to obtain the corresponding content identifier embedding vector and content paragraph embedding vector, and further obtaining the feature vector of the domain article.

[0144] S304: Determine a first vector of each field article of the first type through a network model according to the content identifier and content information of each field article.

[0145] The first vector is used to represent the attributes or states of domain entities in domain articles. The first vector is relatively easy to identify and, when processed by the network model, has good recognition effects, improving the accuracy of identifying the type of domain articles and achieving good consistency, thereby increasing processing efficiency.

[0146] In a possible embodiment, domain articles in different fields have different specific first types.

[0147] For example, if the specific field is the medical field, the field article can be a medical article, and the first type can include department type, disease type, and index type. The specific description information corresponding to the above types is given as an example, for example:

[0148] Diseases: such as "high blood pressure" and "cough".

[0149] Department: such as "Internal Medicine" and "Surgery".

[0150] Index: such as "keyword index" and "author index".

[0151] In the medical field, the specific description information of the first vector is used to describe the attributes or states of medical entities in medical articles.

[0152] For example, if the specific field is the e-commerce field, the first type may include: order number, user mobile phone number, product category, product price, etc.

[0153] For example, if the specific field is the e-commerce field, the entity may be a commodity, and the first type may include an order number, a user's mobile phone number, a commodity category, a commodity price, and the like.

[0154] For example, if the specific field is logistics, the entity may be goods, etc., and the first type may include: location information, time information, cost, inventory, etc.

[0155] In a possible embodiment, the network model can be a generative pre-trained model, which calls the GPT labeling interface in the natural language processing service through the GPT gateway. Since it is impossible to determine which are the first type of domain articles before labeling, the multiple acquired domain articles are input into GPT for identification. GPT can identify the first type of domain articles and add a first vector to the first type of domain articles.

[0156] It should be noted that by inputting all domain articles into the GPT tagging interface, the feature vectors in each domain article can also be extracted and stored in the stock vector database, which can be used to identify the second vector during the vector matching process.

[0157] In one possible embodiment, Figure 4 A schematic diagram of the process of labeling through a generative pre-training model is shown, as Figure 4 As shown, the following steps are included:

[0158] S402: Input multiple field articles into the generative pre-training model.

[0159] S404: Inputting instruction information for identifying the first type of domain articles into the generative pre-training model, so that the generative pre-training model identifies the first type of domain articles according to the content identifier and content information of each domain article.

[0160] S404: Add a tag vector to be verified for the first type of domain articles.

[0161] S406: Verify the tag vector to be verified to obtain a first vector.

[0162] It should be noted that since the GPT model needs to recognize the text in the article to determine what type of article it is, medical articles are directly input into the GPT model instead of feature vectors.

[0163] GPT adds text labels to articles in the first category. These labels can include, for example, "disease-cough." The labels are then verified to produce the first vector.

[0164] Furthermore, the process of verifying the label may include: determining a first similarity based on the label vector to be verified and the first type of domain article vector stored in the label vector library; verifying the first vector based on the first similarity; if the first similarity is greater than a first threshold, the verification is passed, and the label vector to be verified that passes the verification is used as the first vector of the first type of domain article.

[0165] Labeling the first type of domain articles through the above-mentioned GPT model is a targeted step performed based on the first type of domain articles. It is not only highly accurate and efficient, but also has low CPU processing power usage.

[0166] Exemplarily, if the field is the medical field, multiple medical articles are input into the generative pre-training model, and indication information for identifying medical articles of department type, disease type, and index type is input into the generative pre-training model, so that the generative pre-training model identifies medical articles of department type, disease type, and index type based on the content identification and content information of each medical article, adds labels to the medical articles of department type, disease type, and index type, and verifies the label vector to be verified corresponding to the label to obtain a first vector.

[0167] S306: Match the feature vector of each domain article with the known second type of domain article vectors in the pre-built vector space model to determine a second vector of the second type of domain article.

[0168] Among them, the known second type of domain article vectors can be understood as some existing second vectors in the vector space model. These existing second vectors are matched with the feature vectors corresponding to the domain articles to be classified to mark the feature vectors of the domain articles to be classified.

[0169] Among them, the second vector is used to characterize events or behaviors in the field articles. By determining the second type of field articles and determining the second vector by vector matching, compared with the manual labeling method, the subjective judgment of the field articles is reduced, the consistency is better, and the efficiency of meeting the standards is improved. For example, if the specific field is the medical field, the second type may include: drug type, symptom type, category type and label type, etc. Among them, the corresponding specific descriptive information can be illustrated by example, for example:

[0170] Medicines: such as "aspirin", "insulin" and "antibiotics".

[0171] Category: such as "drug classification" and "disease classification".

[0172] Labels: such as "contraindications", "side effects" and "dosage and administration".

[0173] Symptoms: such as "fever", "cough" and "headache", etc.

[0174] The above information mainly describes events or behaviors in medical articles. If it is also input into the GPT model for recognition and labeling, the recognition effect is not accurate enough. Labeling through vector matching is more accurate.

[0175] For example, if the specific field is the e-commerce field, the second type may include transaction events, historical purchase records, user preferences, etc.

[0176] For example, if the specific field is logistics, the second type may include: the actual distribution status of the current goods, etc.

[0177] Other specific areas will not be discussed here.

[0178] In a possible embodiment, similarities between vectors are matched through a similarity query interface. Figure 5 A schematic diagram of a vector matching marking process is shown in FIG. Figure 5 As shown, the following steps are included:

[0179] S502: Determine a second similarity between the feature vector of each domain article and a known second type of domain article vector.

[0180] For example, the similarity may be calculated by a vector matching algorithm.

[0181] S504: For any feature vector of the field article to be classified, if the corresponding second similarity is greater than the second threshold, the type corresponding to the field article to be classified is determined to be the second type, and the second vector of the field article to be classified is determined.

[0182] For example, the similarity between vectors is used to determine which known second-type domain article vectors the feature vector corresponding to the domain article better matches and has a higher similarity, and then the second vector is added to the domain article to be classified.

[0183] In a possible embodiment, if the field is the medical field, the vector matching method may include: determining the second similarity of each type between the feature vector of each field article and the medical article vectors of known drug types, symptom types, category types, and label types; for the feature vector of any medical article, if the second similarity of the corresponding type is greater than a second threshold, determining that the type corresponding to any medical article to be classified is the second type, and determining the second vector of any field article.

[0184] For example, if the second similarity between the feature vector of any medical article and the medical article vector of a known drug type is greater than the second threshold, it can be determined that any medical article belongs to the drug type, and any medical article vector to be classified is a drug vector.

[0185] It should be noted that through vector matching, the feature vector of each field article is directly input. The feature vector itself exists in the form of a vector, but it may not be in the form of a second vector. For example, the original feature vector may only be the vector "aspirin". After vector matching, it is determined that its second similarity with the known second type of field article vector is greater than the second threshold. In this case, the format of the feature vector needs to be unified with the second vector proposed in this disclosure. The feature vector is then added in the format of "Drug-Aspirin". This is only for example. The specific first vector and second vector may have more format requirements.

[0186] Determining the second vector of the domain article by vector matching can save the process of recognizing the text part of the domain article, and the accuracy of vector matching is high.

[0187] pass Figure 4 and Figure 5 Compared with using only GPT for labeling, the accuracy of labeling the second type of domain articles can be improved. GPT's accuracy in identifying the second type of domain articles is relatively low. Compared with using only vector matching, the method disclosed herein uses GPT labeling to improve the efficiency and speed of labeling the first type of domain articles. The method disclosed herein selects the best labeling method for different types of domain articles and ultimately classifies them, ensuring good consistency while improving accuracy and efficiency.

[0188] S308: Classify the plurality of articles in the field to be classified according to the first vector and the second vector.

[0189] In a possible embodiment, based on the first vector and the second vector, a labeling result of the domain article to be classified can be obtained, and the domain article is classified in the vector space model according to the labeling result and classified with the known domain articles in the vector space model.

[0190] If the field is medical, based on the first and second types described above, the first and second vectors of medical articles in this disclosure may include: disease tags, department tags, index tags, drug tags, symptom tags, category tags, and label tags. Each vector includes a specific content label vector. For example, the disease tag may be a vector corresponding to "disease-cough," and so on. Alternatively, the first and second vectors may be understood as a table containing multiple specific content label vectors.

[0191] In a possible embodiment, after classification, in response to a query request in which a user inputs a unique identifier (contentId) of any field article to conduct a search, the corresponding field article and the added vector may be displayed.

[0192] In one possible embodiment, a marking result is determined based on the first vector and the second vector; the official data and draft data of the marking result are stored in a stock database and a cache database, respectively; and different management permissions are set for the official data and the draft data. By using different permission management methods, data security can be improved.

[0193] In a possible embodiment, in response to a user's query request for domain article tags, the tagging results are preferentially extracted from the cache database to display the query results to the user.

[0194] Furthermore, based on the user's authority level, it is determined whether the query result to be displayed is formal data or draft data.

[0195] In one possible embodiment, the present disclosure can monitor tagging status. Specifically, this can include: constructing a tagging status record table; during the process of categorizing field articles, recording the tagging status of field articles in the tagging status record table. This allows for real-time recording of tagging status, tracking successful and failed tagging, and providing insights into tagging progress and results. This is then displayed in a visual interface, improving user convenience.

[0196] In a possible embodiment, the field articles to be classified in the present disclosure are mainly existing field articles obtained from the existing database. In the process of article classification, incremental field articles may exist at any time. The incremental field articles can be processed into existing field articles and then processed in the same way as the existing field articles.

[0197] Exemplarily, the processing method of incremental medical articles in the present disclosure may include: receiving incremental domain articles through a message queue; the incremental domain articles and the domain articles are in the same preset domain; preprocessing and natural language processing technology are performed on the incremental domain articles to obtain the content identification, content information and corresponding feature vector corresponding to the incremental domain article; storing the feature vector corresponding to the incremental domain article in an existing vector database; storing the content identification and content information corresponding to the incremental domain article in an existing database, and classifying them according to the method of existing domain articles.

[0198] In a possible embodiment, the present disclosure can also utilize graph database technology to construct knowledge graphs for different preset domains based on domain articles, vectors, content identifiers, content information, and domain entities, relationships, and attributes in domain articles to represent domain entities, relationships, and attributes in different domains and support complex query and analysis operations.

[0199] Through the above-mentioned method, the method disclosed in the present invention can process a large number of field articles, which may include existing field articles and incremental field articles, and by introducing the Spring Batch architecture, NLP technology, vector space model and GPT model, etc., automatic field article classification and labeling are realized, which gets rid of the inefficiency and inconsistency of manual labeling one by one. Semantic analysis is realized by using NLP technology. By applying NLP technology, semantic analysis of field articles is realized, including part-of-speech tagging, named entity recognition, keyword extraction, etc., to improve the accuracy of article classification and labeling. And combined with field knowledge, big data analysis is carried out through GPT model, vector matching and other methods to improve the accuracy and scalability of field article classification and labeling.

[0200] Through the spatial vector model, which includes the relationship between various features and vectors of field articles, and the relationship and attributes between various features and different types, the efficiency and accuracy of users in the query and analysis process can be improved.

[0201] Furthermore, since the method disclosed herein processes all articles in a field, the large amount of data is a real problem. If it is processed in a centralized manner, the hardware requirements are very high. Through a batch processing architecture, the requirements for hardware resource processing capabilities can be reduced in batches, and the transmission speed of the hardware when transmitting data can be improved. In addition, GPT can be used to mark the first type of field articles. GPT has relatively low hardware requirements, which can further reduce the hardware requirements of the method and improve computing efficiency.

[0202] In a possible embodiment, for medical articles in the medical field, Figure 6 A flow chart for classifying medical articles is shown. Figure 6 As shown, the following steps are included:

[0203] S602: Acquire multiple medical articles to be classified.

[0204] S604: Determine the department vector of the department-type medical article, the disease vector of the disease-type medical article, and the index vector of the index-type medical article through a network model according to the content identifier and content information of each medical article.

[0205] S606: Match the feature vector of each medical article with the known medical article vectors of drug type, symptom type, category type, and label type in the pre-constructed vector space model to determine the drug vector, symptom vector, category vector, and label vector of the medical article of drug type, symptom type, category type, and label type, respectively.

[0206] S608: Classify multiple medical articles to be classified according to the department vector, disease vector, index vector, drug vector, symptom vector, category vector and label vector.

[0207] Based on some properties of medical articles, we label them in two corresponding ways according to the department type, disease type, and index type included in the first type, and the drug type, symptom type, category type, and label type included in the second type, and obtain department vectors, disease vectors, index vectors, and drug vectors, symptom vectors, category vectors, and label vectors respectively to classify medical articles.

[0208] The entire process is completed automatically, without the need for human intervention to add label vectors and classification. It has high consistency and targeted use of model marking and vector matching to improve marking accuracy and efficiency.

[0209] In a possible embodiment, the marking method by vector matching requires the use of a pre-built vector space model, where: Figure 7 A schematic diagram of the process of constructing a vector space model is shown; Figure 7 As shown, the following steps are included:

[0210] S702: Obtain domain articles of known types, and determine known label vectors corresponding to the domain articles of known types.

[0211] For example, a known label vector can be obtained by generating feature vectors based on domain articles of known types and domain articles of known labels.

[0212] After generating a known label vector, the label vector can be stored in a label database so that it can be directly obtained for subsequent use.

[0213] When constructing a vector space model, we can consider the feature weights corresponding to articles of known domain types. These weights can take into account factors such as the number of word occurrences (term frequency, tf), document frequency (df), and inverse document frequency (IDF). By considering these factors, the weighting of features can characterize the emphasis of different features in domain articles. When labeling through vector matching, calculating similarity for matching can improve accuracy.

[0214] S704: According to the vector matching algorithm, field articles with known label vectors are matched, classification is constructed, and a vector space model is obtained.

[0215] Each category corresponds to a type, and both the first type and the second type include multiple specific types. Vector matching algorithms may include: cosine similarity and Euclidean distance, etc.

[0216] The pre-built vector space model can improve the efficiency and accuracy of user query and analysis of field articles. The vector space model includes different classifications that can be used in the tagging process through vector matching.

[0217] In one possible embodiment, a corresponding vector space model is generated for each field of predefined articles, stored and used separately, and a knowledge graph is constructed for each predefined field. Labeling results for different fields and classification results after article classification are also saved separately for user convenience.

[0218] Based on the same inventive concept as the above method embodiment, the embodiment of the present disclosure also provides an article classification device. Figure 8 A structural diagram of an article classification device provided by an embodiment of the present disclosure is shown.

[0219] The device 80 includes: an acquisition unit 801, used to acquire multiple domain articles to be classified; the domain articles belong to the same preset domain; a first label adding unit 802, used to determine the first vector of each domain article of the first type through a network model based on the content identification and content information of each domain article; the first vector is used to represent the attributes or status of the domain entity in the domain article; a second label adding unit 803, used to match the feature vector of each domain article with the known second type of domain article vector in a pre-constructed vector space model to determine the second vector of the second type of domain article; the second vector is used to represent the event or behavior in the domain article; a classification unit 804, used to classify the multiple domain articles to be classified based on the first vector and the second vector.

[0220] Those skilled in the art will appreciate that various aspects of the present invention may be implemented as systems, methods, or program products. Therefore, various aspects of the present invention may be implemented in the following forms: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits," "modules," or "systems."

[0221] Refer to the following Figure 9 An electronic device 900 according to this embodiment of the present invention will be described. Figure 9 The electronic device 900 shown is merely an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0222] like Figure 9As shown, electronic device 900 is implemented as a general-purpose computing device. Components of electronic device 900 may include, but are not limited to, at least one processing unit 910, at least one storage unit 920, and a bus 930 connecting various system components (including storage unit 920 and processing unit 910).

[0223] The storage unit stores program codes, which can be executed by the processing unit 910, so that the processing unit 910 performs the steps according to various exemplary embodiments of the present invention described in the above “Exemplary Method” section of this specification.

[0224] The storage unit 920 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 9201 and / or a cache memory unit 9202 , and may further include a read-only memory unit (ROM) 9203 .

[0225] The storage unit 920 may also include a program / utility 9204 having a set (at least one) of program modules 9205, such program modules 9205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0226] Bus 930 may represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0227] The electronic device 900 can also communicate with one or more external devices 940 (e.g., a keyboard, a pointing device, a Bluetooth device, etc.), one or more devices that enable a user to interact with the electronic device 900, and / or any device that enables the electronic device 900 to communicate with one or more other computing devices (e.g., a router, a modem, etc.). Such communication can occur via an input / output (I / O) interface 950. Furthermore, the electronic device 900 can communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via a network adapter 960. As shown, the network adapter 960 communicates with other modules of the electronic device 900 via a bus 930. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic device 900, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0228] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

[0229] In exemplary embodiments of the present disclosure, a computer-readable storage medium is also provided, on which is stored a program product capable of implementing the aforementioned methods of this specification. In some possible implementations, various aspects of the present invention may also be implemented in the form of a program product comprising program code. When the program product is executed on a terminal device, the program code is configured to cause the terminal device to execute the steps according to various exemplary embodiments of the present invention described in the "Exemplary Methods" section of this specification.

[0230] A program product for implementing the above-described method according to an embodiment of the present invention is described. The program product may be a portable compact disc read-only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto. In this document, a readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0231] The program product may be implemented in any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0232] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0233] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0234] The program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and the like, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a stand-alone software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0235] It should be noted that although several modules or units of the device for action execution are mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0236] Furthermore, although the steps of the method of the present disclosure are described in a particular order in the accompanying drawings, this does not require or imply that the steps must be performed in this particular order, or that all steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0237] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

[0238] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the appended claims.

Claims

1. A method for article classification, characterized in that: The method comprises: Obtain multiple field articles to be classified; the field articles belong to the same preset field; Determining a first vector for each domain article of the first type through a network model based on the content identifier and content information of each domain article; the first vector is used to represent an attribute or state of a domain entity in the domain article; According to the feature vector of each domain article, the feature vector is matched with the known second type of domain article vectors in the pre-built vector space model to determine the second vector of the second type of domain article; the second vector is used to represent the event or behavior in the domain article; A plurality of articles in a field to be classified are classified according to the first vector and the second vector.

2. The method according to claim 1, characterized in that The field articles include: medical articles in the medical field; the first type includes: department type, disease type and index type; the second type includes: drug type, symptom type, category type and tag type; The method further comprises: Obtain multiple medical articles to be classified; According to the content identifier and content information of each medical article, the department vector of the medical article of department type, the disease vector of the medical article of disease type, and the index vector of the medical article of index type are determined through the network model; Based on the feature vector of each medical article, the feature vector is matched with the medical article vectors of known drug type, symptom type, category type, and label type in the pre-built vector space model to determine the drug vector, symptom vector, category vector, and label vector of the medical article of drug type, symptom type, category type, and label type respectively; The plurality of medical articles to be classified are classified according to the department vector, disease vector, index vector, drug vector, symptom vector, category vector and label vector.

3. The method according to claim 1, characterized in that The determining of the first vector of each field article of the first type through a network model based on the content identifier and content information of each field article includes: Inputting the plurality of field articles into a generative pre-training model; Inputting instruction information for identifying a first type of domain article into the generative pre-training model, so that the generative pre-training model identifies the first type of domain article based on the content identifier and content information of each domain article; Adding a label vector to be verified for the first type of field articles; The tag vector to be checked is checked to obtain a first vector.

4. The method according to claim 3, characterized in that The step of verifying the tag vector to be verified to obtain a first vector includes: Determining a first similarity based on the to-be-verified label vector and the first type of domain article vectors stored in the label vector library; Verifying the first vector according to the first similarity; If the first similarity is greater than the first threshold, the verification is passed, and the passed label vector to be verified is used as the first vector of the first type of field article.

5. The method according to claim 1, characterized in that The step of matching the feature vector of each domain article with the known second type of domain article vectors in the pre-built vector space model to determine the second vector of the second type of domain article includes: Determining a second similarity between the feature vector of each field article and the known field article vector of the second type; For any feature vector of a field article to be classified, if the corresponding second similarity is greater than a second threshold, the type corresponding to the field article to be classified is determined to be the second type, and the second vector of the field article to be classified is determined.

6. The method according to claim 1, characterized in that Construct a vector space model, including: Obtain domain articles of known types and determine the known label vectors corresponding to domain articles of known types; According to a vector matching algorithm, the field articles with the known label vectors are matched, and classifications are constructed to obtain a vector space model; wherein each classification corresponds to a type.

7. The method according to claim 1, characterized in that The acquisition of multiple field articles includes: In response to the user's indication of the classification of the domain articles, under the batch processing architecture, the existing domain articles are read in batches from the existing database in a declarative input and output manner as multiple domain articles to be classified.

8. The method according to claim 1, characterized in that The method further comprises: Pre-process the field articles to obtain content identification and content information; Using the content identifier as a unique identifier, natural language processing technology is used to generate vectors for the content information, and the feature vectors of the obtained field articles are stored in the existing vector database; Among them, after obtaining multiple field articles, it also includes: The content identifier and content information of each field article are obtained, and the feature vector of each field article is obtained from the stock vector database.

9. The method according to claim 8, characterized in that Before obtaining the plurality of field articles, the method further includes: receiving an incremental domain article through a message queue; the incremental domain article and the domain article are in the same preset domain; Preprocessing and natural language processing technology are performed on the incremental field articles to obtain content identification, content information and corresponding feature vectors corresponding to the incremental field articles; Storing the feature vectors corresponding to the incremental field articles in the stock vector database; The content identifiers and content information corresponding to the incremental field articles are stored in the stock database and classified according to the stock field articles.

10. The method according to claim 1, characterized in that After determining the first vector and the second vector, the method further includes: Determine a marking result according to the first vector and the second vector; The formal data and draft data of the marking results are stored in the stock database and the cache database respectively; Different management permissions are set for the official data and the draft data.

11. The method according to claim 1, wherein The method further comprises: Construct a marking record sheet; In the process of classifying the field articles, the marking status of the field articles is recorded in the marking status record table.

12. An article classification device, characterized in that: include: An acquisition unit, used to acquire multiple articles in the field to be classified; The field articles mentioned belong to the same preset field; A first label adding unit is configured to determine a first vector of each field article of the first type through a network model according to the content identifier and content information of each field article; The first vector is used to represent the attributes or states of the domain entities in the domain article; A second label adding unit is configured to match the feature vector of each domain article with a known second type of domain article vector in a pre-built vector space model to determine a second vector of the second type of domain article; the second vector is used to represent an event or behavior in the domain article; A classification unit is used to classify multiple articles in the field to be classified according to the first vector and the second vector.

13. An electronic device, characterized in that: include: processor; and a memory for storing executable instructions for the processor; The processor is configured to perform the method according to any one of claims 1 to 11 by executing the executable instructions.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.

Images