Object category evaluation method and device

Abstract

The present invention relates to the field of artificial intelligence, and specifically provides an object category evaluation method and device, which aims to solve the problem that the existing method directly trains a neural network based on a small-scale positive sample, and the accuracy is low. To this end, the object category evaluation method of the present invention includes obtaining training samples and samples to be analyzed; generating a decision stream based on the training samples; training a generative adversarial network based on the decision stream to obtain a trained generative adversarial network; and evaluating the samples to be analyzed based on the trained generative adversarial network to obtain an object category.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and specifically provides an object category evaluation method and device. Background Art

[0002] At present, the application of artificial intelligence in the public security field is becoming more and more extensive, and the most notable one is the analysis system based on machine learning algorithms. The analysis system based on machine learning algorithms mainly uses small-scale positive samples to train the neural network, and then uses the trained neural network to analyze the text to obtain the analysis results. However, the accuracy of the neural network directly trained based on small-scale positive samples is low. The evaluation of object categories with abnormal activity patterns such as drug-related persons, fugitives, and thieves can only be achieved by experts, which is difficult to meet actual needs.

[0003] Therefore, a new solution is needed in the art to solve the above problems. Summary of the invention

[0004] In order to overcome the above defects, the present invention is proposed to solve or at least partially solve the technical problem that the existing methods directly train the neural network based on small-scale positive samples with low accuracy. The present invention provides an object category evaluation method and device.

[0005] In a first aspect, the present invention provides an object category assessment method, comprising: obtaining training samples and samples to be analyzed; generating a decision stream based on the training samples; training a generative adversarial network based on the decision stream to obtain a trained generative adversarial network; and evaluating the samples to be analyzed based on the trained generative adversarial network to obtain an object category.

[0006] In some embodiments, generating a decision flow based on the training sample further includes: extracting entities and relationships between entities in the training sample; establishing a knowledge base based on the entities and relationships between entities; and generating a decision flow based on the established knowledge base.

[0007] In some embodiments, the knowledge base includes an embedded feature vector and a decision flow sub-model; establishing a knowledge base based on the entities and the relationships between entities further includes: obtaining a subject library corresponding to the entity based on the entity; obtaining a knowledge graph based on the entity and the relationships between entities; extracting and saving the embedded feature vector based on the knowledge graph; obtaining and saving the decision flow sub-model based on the embedded feature vector and the subject library corresponding to the entity.

[0008] In some embodiments, extracting an embedded feature vector based on the knowledge graph further includes: obtaining a community structure in the knowledge graph based on a community discovery algorithm; evaluating each node in the community structure to obtain the feature vector centrality and point degree centrality corresponding to each node; and obtaining an embedded feature vector based on the feature vector centrality and point degree centrality corresponding to each node.

[0009] In some embodiments, obtaining an embedded feature vector based on the eigenvector centrality and point degree centrality corresponding to each node further includes: obtaining the weight of each node and the weight of each edge in the community structure based on the eigenvector centrality and point degree centrality corresponding to each node; constructing an adjacency matrix of the community structure based on the weight of each node and the weight of each edge; decomposing the adjacency matrix to obtain a vector representation of each node in the community structure; performing vector expansion on the vector representation of each node to obtain an expanded vector; and performing attribute simplification on the expanded vector to obtain an embedded feature vector.

[0010] In a second aspect, the present invention provides an object category evaluation device, comprising: an acquisition module, configured to acquire training samples and samples to be analyzed; a generation module, configured to generate a decision stream based on the training samples; a training module, configured to train a generative adversarial network based on the decision stream to obtain a trained generative adversarial network; an evaluation module, configured to evaluate the samples to be analyzed based on the trained generative adversarial network to obtain an object category.

[0011] In some embodiments, the generation module further includes: an entity and relationship extraction module, configured to extract entities and relationships between entities in the training samples; a knowledge base establishment module, configured to establish a knowledge base based on the entities and relationships between entities; and a decision flow generation module, configured to generate a decision flow based on the established knowledge base.

[0012] In some embodiments, the knowledge base includes an embedded feature vector and a decision flow sub-model; the knowledge base establishment module further includes: a subject library acquisition module, configured to obtain a subject library corresponding to an entity based on the entity; a knowledge graph acquisition module, configured to obtain a knowledge graph based on the entity and the relationship between entities; a vector extraction module, configured to extract and save the embedded feature vector based on the knowledge graph; and a decision flow sub-model acquisition module, configured to obtain and save the decision flow sub-model based on the embedded feature vector and the subject library corresponding to the entity.

[0013] In some embodiments, the vector extraction module further includes: a community structure acquisition module, configured to acquire the community structure in the knowledge graph based on a community discovery algorithm; a centrality acquisition module, configured to evaluate each node in the community structure to obtain the feature vector centrality and point degree centrality corresponding to each node; and an embedded feature vector acquisition module, configured to obtain an embedded feature vector based on the feature vector centrality and point degree centrality corresponding to each node.

[0014] In some embodiments, the embedded feature vector acquisition module further includes: a weight acquisition module, configured to acquire the weight of each node in the community structure and the weight of each edge based on the feature vector centrality and point degree centrality corresponding to each node; a construction module, configured to construct the adjacency matrix of the community structure based on the weight of each node and the weight of each edge; a decomposition module, configured to decompose the adjacency matrix to obtain a vector representation of each node in the community structure; a vector expansion module, configured to perform vector expansion on the vector representation of each node to obtain an expanded vector; and an attribute simplification module, configured to perform attribute simplification on the expanded vector to obtain an embedded feature vector.

[0015] In a third aspect, a control device is provided, comprising a processor and a storage device, wherein the storage device is suitable for storing a plurality of program codes, and the program codes are suitable for being loaded and run by the processor to execute any of the object category assessment methods described above.

[0016] In a fourth aspect, a computer-readable storage medium is provided, wherein a plurality of program codes are stored in the computer-readable storage medium, wherein the program codes are suitable for being loaded and run by a processor to execute any of the aforementioned object category assessment methods.

[0017] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

[0018] The present invention processes training samples to generate a decision stream, and then uses the decision stream to train a neural network, thereby obtaining a neural network with higher accuracy, thereby improving the accuracy of subsequent object category evaluation.

[0019] A knowledge base is established by extracting entities and the relationships between entities in the training samples. The decision flow is obtained based on the entities in the knowledge base, the relationships between entities and the decision flow sub-model, and the decision flow is used to train the neural network. Compared with the existing method of directly training the neural network with small-scale positive examples, the accuracy of the neural network is improved, which is beneficial to the later object category evaluation.

[0020] The embedded feature vector is obtained through the feature vector centrality and point degree centrality corresponding to each node. The method is novel and provides a basis for subsequent object category evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In addition, similar numbers in the figures are used to represent similar components, among which:

[0022] Figure 1 is a flowchart of the main steps of an object category evaluation method according to an embodiment of the present invention;

[0023] Figure 2 is a schematic diagram of a syntax tree according to an embodiment of the present invention;

[0024] Figure 3 It is a schematic diagram of the main structure block diagram of an object category evaluation device according to an embodiment of the present invention.

[0025] Reference numerals list :

[0026] 11: Acquisition module; 12: Generation module; 13: Training module; 14: Evaluation module. DETAILED DESCRIPTION

[0027] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0028] In the description of the present invention, "module" and "processor" may include hardware, software or a combination of the two. A module may include hardware circuits, various suitable sensors, communication ports, and memories, and may also include software parts, such as program codes, or a combination of software and hardware. The processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, hardware, or a combination of the two. Non-temporary computer-readable storage media include any suitable medium that can store program codes, such as a magnetic disk, a hard disk, an optical disk, a flash memory, a read-only memory, a random access memory, and the like. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The term "at least one A or B" or "at least one of A and B" has a similar meaning to "A and / or B", and may include only A, only B, or A and B. The singular terms "one" and "the" may also include plural forms.

[0029] Here we first explain some terms involved in the present invention.

[0030] Generative Adversarial Network: Generative Adversarial Network is a kind of generative model. Its main structure includes a generator G (Generator) and a discriminator D (Discriminator). It uses the sample distribution to fit the real sample distribution to achieve the purpose of making the fake look real.

[0031] Eigenvector centrality: measures the value of a node based on the importance of its neighboring nodes.

[0032] Point degree centrality: how many lines are connected to a node, that is, the sum of the in-degree and out-degree of the node.

[0033] At present, the traditional analysis system based on machine learning algorithms is mainly based on small-scale positive sample direct training to obtain the neural network with low accuracy. The evaluation of object categories with abnormal activity patterns such as drug-related persons, fugitives, thieves, etc. can only be achieved by experts, which is difficult to meet actual needs. To this end, this application proposes an object category evaluation method and device, which processes the training samples to generate a decision stream, and then uses the decision stream to train the neural network to obtain a neural network with higher accuracy, thereby improving the accuracy of the later object category evaluation.

[0034] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart showing the main steps of an object category evaluation method according to an embodiment of the present invention. Figure 1 As shown, the object category evaluation method in the embodiment of the present invention mainly includes the following steps S101 to S104.

[0035] Step S101: Obtain training samples and samples to be analyzed, wherein the training samples include training texts, and the samples to be analyzed include texts to be analyzed. Specifically, the training texts here may be historical files or historical data containing object categories, and the object categories may be drug-related persons, fugitives, thieves, etc. The texts to be analyzed may be a series of texts, audios, or videos, etc., which need to be analyzed or processed.

[0036] Step S102: Generate a decision stream based on the training sample. Specifically, this step can be implemented by the following steps S1021-S1023, which are not described here. The decision stream is generated by this step, and then the decision stream is used to train the neural network. Compared with the existing method of directly using a small-scale positive sample to train the neural network, the accuracy of the neural network is improved, which is conducive to improving the accuracy of the later object category evaluation.

[0037] Step S1021: Extract entities and relationships between entities in the training samples. Specifically, in this application, OCR technology and natural language processing technology can be used to extract entities and relationships between entities in the training samples. Furthermore, an entity refers to a combination of data according to the dimension of a controlled object, including an identification ID and entity attributes. The entity attributes may be age, gender, etc. The controlled object includes but is not limited to people, events, organizations, etc. The relationship between entities can be represented by a relationship ID and a relationship attribute.

[0038] Step S1022: Establish a knowledge base based on entities and relationships between entities, where the knowledge base includes embedded feature vectors and decision flow sub-models.

[0039] In some embodiments, the steps of establishing a knowledge base based on entities and the relationships between entities specifically include: first, based on the entity, obtaining a thematic library corresponding to the entity, wherein the thematic library may include a thematic library for personnel call records, a thematic library for personnel financial transactions, a thematic library for personnel travel records, a thematic library for personnel daily activities, and a thematic library for personnel virtual identities, etc. In this application, the method of loss calculation is mainly used to process the entity to obtain the thematic library corresponding to the entity. On the basis of obtaining the thematic library corresponding to the entity, a knowledge graph is further obtained based on the entity and the relationship between the entities. In this embodiment, the knowledge graph is mainly obtained by correspondingly storing the entities and the relationship between the entities. On the basis of obtaining the knowledge graph, the graph embedding algorithm can be further used to extract and save the embedded feature vector. Finally, the decision flow sub-model can be obtained and saved based on the embedded feature vector and the thematic library corresponding to the entity.

[0040] Specifically, the graph embedding algorithm used in this application is a self-developed graph embedding algorithm, but the graph embedding algorithm can also be metapath2vec, random walk and other algorithms, but not limited to this. The steps of extracting embedded feature vectors using the self-developed graph embedding algorithm in this application specifically include: first, using the community discovery algorithm in the social network analysis algorithm to obtain the community structure in the knowledge graph, where the "community discovery algorithm" here can be the KL algorithm, GN algorithm, etc., but not limited to this. Secondly, each node in the community structure is evaluated to obtain the eigenvector centrality and point degree centrality corresponding to each node. On the basis of obtaining the eigenvector centrality and point degree centrality corresponding to each node, the embedded feature vector can be further obtained, which specifically includes the following steps.

[0041] First, according to the eigenvector centrality and point degree centrality corresponding to each node, the weight of each node and the weight of each edge in the community structure are obtained. The weight of each node and the weight of each edge here can be obtained using 1 / (1-Xe -αβ ) is calculated, α is the eigenvector centrality, β is the point degree centrality, and X is the scaling parameter.

[0042] Next, an adjacency matrix of the community structure is constructed based on the weight of each node and the weight of each edge. In this embodiment, the weight of an edge consisting of two nodes is equal to the product of the weight of the edge constituting the shortest path between the two nodes and the weight of the node to obtain the adjacency matrix.

[0043] After obtaining the adjacency matrix, the graph factorization method is used to decompose the adjacency matrix to obtain the vector representation y of each node in the community structure. i , wherein the above-mentioned graph factorization methods include but are not limited to Laplace graph embedding, Cauchy graph embedding, etc.

[0044] The vector representation y for each node i Perform vector expansion to obtain an expanded vector. Specifically, each node is expanded through its vector representation, its own label, and entity attributes to form a new vector Y i , thus obtaining the expanded vector W = [Y 1 ,Y 2 ,…,Y n ].

[0045] The attributes of the expanded vector obtained above are simplified to obtain an embedded feature vector. The attribute simplification here is mainly the attribute simplification of the rough set. Specifically, the first column of attributes of the expanded vector is removed first, and then when the remaining attributes are queried, if the decision attributes of the same conditional attributes do not conflict, it means that the removed first column does not generate inconsistent data, which means that the first column of attributes of the expanded vector can be removed. This cycle is repeated until every attribute in the expanded vector is judged, and the embedded feature vector can be obtained. The embedded feature vector is obtained by the feature vector centrality and point degree centrality corresponding to each node. The method is novel and provides a basis for the subsequent object category evaluation.

[0046] The process of obtaining the decision flow sub-model based on the embedded feature vector and the subject library corresponding to the entity in this application is mainly achieved through supervised models and unsupervised models, wherein the supervised model here may include but is not limited to DNN, Wide-Deep, logistic regression model, etc. Specifically, the decision flow sub-model here includes a personnel anomaly model and a personnel activity dispersion model. In this step, the supervised model is first used to simulate the embedded feature vector and the subject library corresponding to the entity to obtain the personnel anomaly model, and then the unsupervised model is used to obtain the personnel activity dispersion model from the personnel activity trajectory in the subject library corresponding to the entity.

[0047] Step S1023: Generate a decision flow based on the established knowledge base. Specifically, after the knowledge base is established, the user can form a decision flow by dragging or human-computer dialogue. Exemplarily, the following will be explained in the form of human-computer dialogue. Specifically, for a provided audio of drug-related personnel screening, the NLP technology is first used to parse the audio into text, and a syntax tree is generated based on the text. Specifically, for the text obtained after the audio is parsed, "First, find the people who are active at night and hidden during the day in the past month from the subject library and output them; then find the people with abnormal behavior among the people who are active at night and hide during the day, and output the key screening objects", the NLP technology can be used to generate the following Figure 2 The syntax tree shown. NLP technologies include but are not limited to LSTM, BERT, etc. Finally, the syntax tree is parsed by post-order traversal of the tree to obtain the decision flow. Post-order traversal (LRD) is a type of binary tree traversal, which mainly traverses the left subtree first, then the right subtree, and finally visits the root node to obtain the decision flow.

[0048] By processing the training samples to generate a decision stream, and then using the decision stream to train the neural network, a neural network with higher accuracy is obtained, thereby improving the accuracy of subsequent object category evaluation.

[0049] Step S103: training a generative adversarial network based on the decision stream generated in the above steps to obtain a trained generative adversarial network.

[0050] Step S104: Evaluate the sample to be analyzed based on the trained generative adversarial network to obtain the object category. Specifically, before using the trained generative adversarial network to evaluate the sample to be analyzed in this step, the sample to be analyzed is first processed to generate a decision stream, and then the decision stream is input into the trained generative adversarial network to obtain the object category. The object category can be a drug-related person, a fugitive, a thief, etc., wherein the method of "processing the sample to be analyzed to generate a decision stream" here can refer to the aforementioned steps S101-step S102, which will not be repeated here. In addition, the samples to be analyzed and the object category of this step can also be saved in the knowledge base for data enhancement. Using a generative adversarial network to evaluate the sample to be analyzed is convenient and concise, and can effectively improve the efficiency of personnel classification evaluation.

[0051] Based on the above steps S101 to S104, entities and relationships between entities in the training samples are extracted, and then a knowledge base is established. The decision flow is obtained through the established knowledge base, and then the decision flow is used to train the neural network. Compared with the existing method of directly using small-scale positive example samples to train the neural network, the accuracy of the neural network is improved, which is beneficial to the later object category evaluation.

[0052] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art can understand that in order to achieve the effects of the present invention, different steps do not have to be performed in such an order. They can be performed simultaneously (in parallel) or in other orders. These changes are within the scope of protection of the present invention.

[0053] Furthermore, the present invention also provides an object category evaluation device.

[0054] See attached Figure 3 , Figure 3 FIG. 1 is a main structural block diagram of an object category evaluation device according to an embodiment of the present invention. Figure 3 As shown, the object category evaluation device in the embodiment of the present invention mainly includes an acquisition module 11, a generation module 12, a training module 13 and an evaluation module 14. In some embodiments, one or more of the acquisition module 11, the generation module 12, the training module 13 and the evaluation module 14 can be combined into one module. In some embodiments, the acquisition module 11 can be configured to acquire training samples and samples to be analyzed, the training samples include training texts, and the samples to be analyzed include texts to be analyzed. The generation module 12 can be configured to generate a decision stream based on the training samples. The training module 13 can be configured to train a generative adversarial network based on the decision stream to obtain a trained generative adversarial network. The evaluation module 14 can be configured to evaluate the samples to be analyzed based on the trained generative adversarial network to obtain an object category.

[0055] In one embodiment, the generation module further includes an entity and relationship extraction module, a knowledge base establishment module and a decision flow generation module, wherein the entity and relationship extraction module is configured to extract entities and relationships between entities in training samples; the knowledge base establishment module is configured to establish a knowledge base based on entities and relationships between entities; and the decision flow generation module is configured to generate a decision flow based on the established knowledge base.

[0056] In one embodiment, the knowledge base establishment module further includes a subject library acquisition module, a knowledge graph acquisition module, a vector extraction module and a decision flow sub-model acquisition module. The subject library acquisition module is configured to obtain the subject library corresponding to the entity based on the entity; the knowledge graph acquisition module is configured to obtain the knowledge graph based on the entity and the relationship between the entities; the vector extraction module is configured to extract and save the embedded feature vector based on the knowledge graph; the decision flow sub-model acquisition module is configured to obtain and save the decision flow sub-model based on the embedded feature vector and the subject library corresponding to the entity.

[0057] In one embodiment, the vector extraction module further includes a community structure acquisition module, a centrality acquisition module and an embedded feature vector acquisition module. The community structure acquisition module is configured to acquire the community structure in the knowledge graph based on the community discovery algorithm; the centrality acquisition module is configured to evaluate each node in the community structure to obtain the feature vector centrality and point degree centrality corresponding to each node; and the embedded feature vector acquisition module is configured to obtain the embedded feature vector based on the feature vector centrality and point degree centrality corresponding to each node.

[0058] In one embodiment, the embedded feature vector acquisition module further includes a weight acquisition module, a construction module, a decomposition module, a vector expansion module and an attribute simplification module. The weight acquisition module is configured to acquire the weight of each node and the weight of each edge in the community structure based on the feature vector centrality and point degree centrality corresponding to each node; the construction module is configured to construct the adjacency matrix of the community structure based on the weight of each node and the weight of each edge; the decomposition module is configured to decompose the adjacency matrix to obtain the vector representation of each node in the community structure; the vector expansion module is configured to perform vector expansion on the vector representation of each node to obtain an expanded vector; the attribute simplification module is configured to perform attribute simplification on the expanded vector to obtain the embedded feature vector.

[0059] The object category evaluation device is used to perform Figure 1 The object category evaluation method embodiment shown in the figure has similar technical principles, technical problems solved and technical effects produced. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process and related instructions of the object category evaluation device can refer to the contents described in the embodiment of the object category evaluation method, which will not be repeated here.

[0060] It is understood by those skilled in the art that the present invention implements all or part of the processes in the method of the above embodiment, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program code. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electric carrier signals and telecommunication signals.

[0061] Furthermore, the present invention also provides a control device. In one embodiment of the control device according to the present invention, the control device includes a processor and a storage device, the storage device can be configured to store a program for executing the object category evaluation method of the above method embodiment, and the processor can be configured to execute the program in the storage device, which includes but is not limited to the program for executing the object category evaluation method of the above method embodiment. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The control device can be a control device device formed by various electronic devices.

[0062] Furthermore, the present invention also provides a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present invention, the computer-readable storage medium can be configured to store a program for executing the object category evaluation method of the above method embodiment, and the program can be loaded and run by a processor to implement the above object category evaluation method. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0063] Further, it should be understood that since the setting of each module is only for illustrating the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the software in the processor, a part of the hardware, or a part of the combination of software and hardware. Therefore, the number of each module in the figure is only schematic.

[0064] Those skilled in the art will appreciate that the modules in the device can be adaptively split or merged. Such splitting or merging of specific modules will not cause the technical solution to deviate from the principle of the present invention, and therefore, the technical solutions after splitting or merging will fall within the protection scope of the present invention.

[0065] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. An object category assessment method, characterized in that: The steps include: Obtain training samples and samples to be analyzed, where the training samples include training texts, which are historical files or historical data containing object categories; and the samples to be analyzed include texts to be analyzed, which are a series of texts, audios, or videos that need to be analyzed or processed; The categories of subjects are drug-related persons, fugitives or thieves; Generating a decision flow based on the training sample includes: Extracting entities and relationships between entities in the training samples; Establishing a knowledge base based on the entities and the relationships between the entities; Generate a decision flow based on the established knowledge base; Training a generative adversarial network based on the decision stream to obtain a trained generative adversarial network; The sample to be analyzed is evaluated based on the trained generative adversarial network to obtain an object category.

2. The object category evaluation method according to claim 1, characterized in that: The knowledge base includes an embedded feature vector and a decision flow sub-model; Establishing a knowledge base based on the entities and the relationships between the entities further includes: Based on the entity, obtain a subject library corresponding to the entity; Obtain a knowledge graph based on the entities and the relationships between the entities; Extracting and saving embedded feature vectors based on the knowledge graph; A decision flow sub-model is obtained and saved based on the embedded feature vector and the subject library corresponding to the entity.

3. The object category evaluation method according to claim 2, characterized in that: Extracting an embedded feature vector based on the knowledge graph further includes: Acquire the community structure in the knowledge graph based on a community discovery algorithm; Evaluate each node in the community structure to obtain the eigenvector centrality and point degree centrality corresponding to each node; An embedded feature vector is obtained based on the feature vector centrality and point degree centrality corresponding to each node.

4. The object category evaluation method according to claim 3, characterized in that: Obtaining an embedded feature vector based on the feature vector centrality and point degree centrality corresponding to each node further includes: Based on the eigenvector centrality and point degree centrality corresponding to each node, obtaining the weight of each node and the weight of each edge in the community structure; Constructing an adjacency matrix of the community structure based on the weight of each node and the weight of each edge; Decomposing the adjacency matrix to obtain a vector representation of each node in the community structure; Performing vector expansion on the vector representation of each node to obtain an expanded vector; Attribute reduction is performed on the expanded vector to obtain an embedded feature vector.

5. An object category evaluation device, characterized in that: include: The acquisition module is configured to acquire training samples and samples to be analyzed, wherein the training samples include training texts, which are historical files or historical data containing object categories; the samples to be analyzed include texts to be analyzed, which are a series of texts, audios or videos that need to be analyzed or processed; and the object categories are drug-related persons, fugitives or thieves; A generation module, configured to generate a decision flow based on the training sample; The generating module further comprises: An entity and relationship extraction module, configured to extract entities and relationships between entities in the training sample; A knowledge base building module, configured to build a knowledge base based on the entities and the relationships between the entities; A decision flow generation module is configured to generate a decision flow based on the established knowledge base A training module is configured to train a generative adversarial network based on the decision stream to obtain a trained generative adversarial network; An evaluation module is configured to evaluate the sample to be analyzed based on the trained generative adversarial network to obtain an object category.

6. The object category evaluation device according to claim 5, characterized in that: The knowledge base includes embedded feature vectors and decision flow sub-models; the knowledge base building module further includes: The subject library acquisition module is configured to obtain the subject library corresponding to the entity based on the entity, A knowledge graph acquisition module, configured to obtain a knowledge graph based on the entities and the relationships between the entities; A vector extraction module, configured to extract and save an embedded feature vector based on the knowledge graph; The decision flow sub-model acquisition module is configured to obtain and save the decision flow sub-model based on the embedded feature vector and the subject library corresponding to the entity.

7. The object category evaluation device according to claim 6, characterized in that: The vector extraction module further comprises: A community structure acquisition module, configured to acquire the community structure in the knowledge graph based on a community discovery algorithm; A centrality acquisition module is configured to evaluate each node in the community structure to obtain a characteristic vector centrality and a point degree centrality corresponding to each node; The embedded feature vector acquisition module is configured to obtain the embedded feature vector based on the feature vector centrality and point degree centrality corresponding to each node.

8. The object category evaluation device according to claim 7, characterized in that: The embedded feature vector acquisition module further comprises: A weight acquisition module, configured to acquire the weight of each node and the weight of each edge in the community structure based on the feature vector centrality and point degree centrality corresponding to each node; A construction module, configured to construct an adjacency matrix of the community structure based on the weight of each node and the weight of each edge; A decomposition module is configured to decompose the adjacency matrix to obtain a vector representation of each node in the community structure; A vector expansion module is configured to perform vector expansion on the vector representation of each node to obtain an expanded vector; The attribute reduction module is configured to perform attribute reduction on the expanded vector to obtain an embedded feature vector.

9. A control device, comprising a processor and a storage device, wherein the storage device is suitable for storing a plurality of program codes, characterized in that: The program code is suitable for being loaded and executed by the processor to perform the object category assessment method according to any one of claims 1 to 4.

10. A computer-readable storage medium storing a plurality of program codes, characterized in that: The program code is suitable for being loaded and run by a processor to execute the object category assessment method according to any one of claims 1 to 4.

Images