A dual-process theory-driven health false information identification system and identification method

The health misinformation identification system driven by the dual-process theory combines knowledge breadth retrieval and deep reasoning modules to simulate the human information processing process and provide post-hoc explanations. This solves the problem of insufficient transparency for the elderly regarding misinformation identification systems and improves the credibility and acceptance of the identification results.

CN118036746BActive Publication Date: 2024-11-19RENMIN UNIVERSITY OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410176844.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-02-08
Publication Date
2024-11-19
Estimated Expiration
2044-02-08

AI Technical Summary

Technical Problem

Existing health misinformation identification systems fail to effectively promote acceptance of their results among the elderly, primarily due to a lack of transparency and an inability to explain the model's internal operational logic and decision-making process.

Method used

A health misinformation identification system driven by a dual-process theory includes a knowledge breadth retrieval module and a knowledge depth reasoning module. It simulates the human information processing process, provides post-event explanations, and achieves two levels of interpretability.

Benefits of technology

It improves the acceptance of misinformation identification results among the elderly. By simulating the human information processing process and providing post-event explanations, it enhances the transparency and credibility of the system, helping the elderly to distinguish correct information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118036746B_ABST
    Figure CN118036746B_ABST
Patent Text Reader

Abstract

The present invention discloses a dual-process theory-driven health false information identification system and identification method. The system includes: a knowledge breadth retrieval module, which is used to retrieve relevant knowledge of the health information to be identified in an external knowledge base; a knowledge depth reasoning module, which is used to perform deep reasoning on the correctness of the health information to be identified based on external relevant knowledge, identify whether the health information is false, and give a post-description of the judgment result. Based on the dual-process theory, the present invention has developed a transparent health false information identification system with dual interpretability, which makes it easier for the elderly to accept the algorithm process, learn the reasons for the judgment, and finally adopt the identification results, which truly helps the elderly resist online health false information and screen correct health information, and has important social significance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of information identification, and in particular to a dual-process theory-driven health false information identification system and identification method. Background Art

[0002] Online health information has become an important source of health care knowledge for the elderly. However, due to the mixed quality of information on the Internet and the current weak information literacy level of the elderly, the elderly often believe in and misrepresent false health information on the Internet, which has a great negative impact on their physical and mental health. The health false information identification system aims to automatically determine the credibility of health information on the Internet, which can help the elderly better integrate into the intelligent era. However, the cognitive bias phenomenon of the "illusory reality effect" often exists in the elderly. They often believe in the cognition they have been indoctrinated to accept, and are unwilling to adopt the judgment results of the health false information identification system. Therefore, it is urgent to design a transparent health false information identification system that is easy for the elderly to adopt from a human-centered perspective, so that the system can truly help the elderly to identify and resist erroneous health information on the Internet.

[0003] In recent years, explainable AI has been one of the important research directions in the field of AI. Some scholars define explainable AI as a "transparent" AI algorithm that users can understand and trust, which specifically requires two levels of explainability: the first level requires the algorithm to be able to extract some feature values ​​from the trained model to form a post-facto explanation of the model's decision results; the second level requires the algorithm to be able to clearly explain the model's internal operation logic.

[0004] However, most of the existing explainable health false information algorithms only achieve the first level of explainability, that is, they can give some reasons for judging the credibility of information. It is still unknown how the model obtains these reasons. Therefore, these models still do not achieve true transparency, which prevents the elderly from accepting the results of the health false information identification system. Summary of the invention

[0005] In order to solve the above-mentioned technical problems, the present invention provides a health false information identification system and identification method driven by a dual-process theory to solve the technical problem that the existing false information identification system cannot promote the elderly to accept the results of the health false information identification system well.

[0006] In the first aspect, the technical solution adopted by the present invention is: a health false information identification system driven by dual-process theory, which includes: a knowledge breadth retrieval module, corresponding to the intuitive process of the dual-process theory, used to retrieve relevant knowledge of the health information to be identified in the external knowledge base; a knowledge depth reasoning module, corresponding to the analysis stage of the dual-process theory, used to perform deep reasoning on the correctness of the health information to be identified based on external relevant knowledge, identify whether the health information is false, and give a post-judgment explanation of the judgment result; wherein, the health false information identification process through the knowledge breadth retrieval module and the knowledge depth reasoning module is consistent with the human information processing process.

[0007] Furthermore, the knowledge breadth retrieval module is used to simulate the first stage of human information processing, the intuitive process; the knowledge breadth retrieval module includes sub-modules: a knowledge extractor, a semantic similarity calculator and a dual filter; the knowledge extractor selects a knowledge graph as an external knowledge base according to the health information to be identified, and quickly retrieves all triples associated with the health information to be identified and containing entity words as head nodes or tail nodes from the external knowledge base through entity word matching, which are defined as a set of candidate triples; the semantic similarity calculator calculates the semantic similarity between each triple and the health information by extracting the semantic features of the health information and the triples in each triple set, and determines the triples similar to the health information; the dual filter distills the determined triples similar to the health information according to the dual principles of more semantic similarity and fairer knowledge distribution to obtain a preset number of triple information.

[0008] Further, the semantic similarity calculator comprises: The triple information retrieved by the knowledge extractor is semantically represented to obtain the semantic vectors corresponding to the health information and triples. and h (h,r,t) ; Calculate the cosine similarity between the health information to be determined and each triple information according to the semantic vector of the two, and determine the triples similar to the health information according to the set high similarity threshold and irrelevant threshold.

[0009] Furthermore, based on the set high similarity threshold and irrelevant threshold, triplets similar to the health information are determined, including: if the cosine similarity between the triplet and the health information to be identified is greater than the high similarity threshold, then the triplet and the health information to be identified are high-similarity triplet; if the cosine similarity between the triplet and the health information to be identified is less than the irrelevant threshold, then the triplet and the health information to be identified are irrelevant triplet, and all irrelevant triplets in the set of alternative triples are eliminated.

[0010] Furthermore, the double filter includes: for each piece of health information, setting a quantity threshold so that at most k triplets can be introduced from the set of alternative triples; within the quantity threshold, selecting triplets in the set of alternative triples whose semantic similarity with the health information is higher than a high similarity threshold, sorting the semantic similarity from high to low, and adding them in sequence to an external knowledge base that assists in the correctness reasoning of health information; within the quantity threshold, continuing to remove the remaining triplets except irrelevant triplets from the set of alternative triples, grouping them according to entities, and sorting each group from high to low according to cosine similarity, as well as the dual principle of fair screening of triplets related to different entity words, adding the triplet currently ranked first from all groups to the external knowledge base that assists in the correctness reasoning of health information, and deleting the triplet from the group until the quantity reaches k or all the remaining triplets are extracted.

[0011] Furthermore, the knowledge depth reasoning module includes: a knowledge feature extraction module, which extracts external entity knowledge representation based on the entity knowledge corresponding to the health information and the triples obtained by the knowledge breadth retrieval module External triplet knowledge representation and semantic representation of health information to be determined Hierarchical attention module, representing external entity knowledge External triplet knowledge representation and semantic representation of health information to be determined The constructed hierarchical attention network is input to fuse the health information semantics with knowledge to obtain the knowledge-aware health information semantics; the credibility prediction module determines whether the knowledge-aware health information semantics is false information and determines its credibility. When the health information obtains the correctness judgment result, the explainable artificial intelligence algorithm can also output the post-event explanation.

[0012] Furthermore, when the health information obtains a correctness judgment result, the explainable artificial intelligence algorithm can also output a post-explanation, including: obtaining a set of external knowledge that assists in reasoning about the correctness of the health information from an external knowledge base as evidence for alternative post-explanations; mining the most important knowledge triples for judging the correctness of the health information based on the attention mechanism; and organizing the key knowledge triples as post-explanations for judging the correctness of the health information.

[0013] Furthermore, it also includes embedding a dual-process theory-driven health false information identification system into social media and search engines: embedding a dual-process theory-driven health false information identification system into the cloud servers of social media and search engines; using the information released by social media and search engines as information to be identified, and judging whether the information to be identified is false information; marking the page of false information with a warning of false information; banning or limiting the flow of false information; and publishing subsequent explanations on the corresponding pages of false information judgments.

[0014] In the second aspect, the technical solution adopted by the present invention is: a method for identifying false health information driven by dual-process theory, which includes: judging whether the information to be identified is false information based on the health information to be identified, using a transparent false health information identification artificial intelligence algorithm with dual interpretability based on the dual-process theory; and mining key external knowledge triples based on the attention mechanism module in the artificial intelligence algorithm as a post-explanation of the health information judgment results.

[0015] In the third aspect, the technical solution adopted by the present invention is: a computer-readable medium, in which a computer program is stored, and the computer program includes program instructions, and when the program instructions are executed by a processor of an electronic device, the processor executes the above-mentioned dual-process theory-driven method for identifying false health information.

[0016] The present invention adopts the above technical solution, which has the following advantages:

[0017] 1. In view of the current situation that "black box" type health false information identification models are difficult to be trusted and adopted by the elderly group, the present invention further promotes users' adoption of information screening results by constructing a transparent detection algorithm that can achieve two levels of explainability (internal logic of the algorithm can be explained and post-explanation can be provided), especially to improve the willingness of the elderly to accept false information identification results.

[0018] 2. Based on the dual-process theory, the corresponding modules are designed to simulate the user's information processing process. The present invention introduces the knowledge graph as a human knowledge memory repository. In addition, the present invention effectively extracts basic knowledge fragments from the knowledge graph by designing two stages of knowledge breadth retrieval (KBR) and knowledge depth reasoning (KDR) corresponding to the heuristic and analytical processes of the dual-process theory, and performs coarse-to-fine reasoning through subtle clues in the knowledge.

[0019] 3. Although existing studies have introduced knowledge graphs as auxiliary information for health information identification, the current solutions have high requirements on the knowledge structure of knowledge graphs and cannot support the dynamic update of knowledge graphs, which limits the practical application of the algorithm. Based on this, the present invention adopts a new method for knowledge graph application, so that the model can support the simultaneous use of multiple knowledge graphs to assist in information determination, support the dynamic update of knowledge graphs, and can simultaneously use the coarse-grained knowledge and fine-grained knowledge contained in the knowledge graph to further enhance the accuracy of health information identification.

[0020] Based on the dual-process theory, the present invention has developed a health false information identification with higher explainability and transparency, which makes it easier for the elderly to accept the algorithm process, understand the reasons for the judgment, and finally adopt the identification results. It can truly help the elderly resist online health false information and screen correct health information, which has important social significance.

[0021] It is to be understood that both 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

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the background technology, the drawings required for use in the embodiments of the present invention or the background technology will be described below.

[0023] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and, together with the specification, are used to illustrate the technical solutions disclosed in the present invention.

[0024] Figure 1 A schematic flow chart of a method for identifying false health information driven by a dual-process theory provided in an embodiment of the present invention;

[0025] Figure 2 This is a framework diagram of a health false information identification system driven by a dual-process theory according to an embodiment of the present invention;

[0026] Figure 3 This is a visualization example diagram of the reasoning process of health false information identification driven by the dual-process theory according to an embodiment of the present invention;

[0027] Figure 4 This is an example diagram of a typical application prototype system interface of an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices.

[0030] The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can represent including any one or more elements selected from the set consisting of A, B, and C.

[0031] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0032] In addition, in order to better illustrate the present invention, numerous specific details are provided in the following specific embodiments. It should be understood by those skilled in the art that the present invention can be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present invention.

[0033] Online health information has become an important source of health care knowledge for the elderly. However, due to the mixed quality of information on the Internet and the current weak information literacy level of the elderly, the elderly often believe in and misrepresent false health information on the Internet, which has a great negative impact on their physical and mental health. The health false information identification system aims to automatically determine the credibility of health information on the Internet, which can help the elderly better integrate into the intelligent era. However, the cognitive bias phenomenon of the "illusory reality effect" often exists in the elderly. They often believe in the cognition they have been indoctrinated to accept, and are unwilling to adopt the judgment results of the health false information identification system. Therefore, it is urgent to design a transparent health false information identification system that is easy for the elderly to adopt from a human-centered perspective, so that the system can truly help the elderly to identify and resist erroneous health information on the Internet.

[0034] In recent years, explainable AI has been one of the important research directions in the field of AI. Some scholars define explainable AI as a "transparent" AI algorithm that users can understand and trust, which specifically requires two levels of explainability: the first level requires the algorithm to extract some feature values ​​from the trained model to form a post-facto explanation of the model's decision results; the second level requires the algorithm to clearly explain the model's internal operation logic. Most of the existing explainable health false information algorithms only achieve the first level of explainability, that is, they can give some reasons for determining the credibility of the information. It is still unknown how the model obtains these reasons. Therefore, these models still do not achieve true transparency, which hinders the elderly from accepting the results of the health false information identification system.

[0035] The present invention discloses a health false information identification system and identification method driven by dual process theory, including: a knowledge breadth retrieval module, corresponding to the intuitive process of the dual process theory, for retrieving relevant knowledge of the health information to be identified in an external knowledge base; a knowledge depth reasoning module, corresponding to the analysis stage of the dual process theory, for performing deep reasoning on the correctness of the health information to be identified based on external relevant knowledge, identifying whether the health information is false, and giving a post-facto explanation of the judgment result. The present invention develops a health false information identification system with higher interpretability based on the dual process theory, which is easier for the elderly to accept the algorithm flow, learn the judgment reasons, and finally adopt the recognition results, which really helps the elderly to resist online health false information and screen correct health information, and has important social significance; the present invention solves the high requirements of health false information detection for external professional knowledge by introducing knowledge graphs, and designs a two-stage model based on the dual process theory, which further improves the reasoning ability of the model and the understandability of the internal operation logic; compared with other published false information detection patents, it is also innovative in research objects and research methods, thereby improving the interpretability, accuracy and practicality of the algorithm.

[0036] Example 1

[0037] In one embodiment of the present invention, a health false information identification system driven by a dual-process theory is provided. Figure 2 In this embodiment, the system includes:

[0038] The knowledge breadth retrieval module, corresponding to the intuitive process of the dual-process theory, is used to retrieve relevant knowledge of the health information to be identified in the external knowledge base;

[0039] The knowledge deep reasoning module corresponds to the analysis stage of the dual-process theory. It is used to conduct deep reasoning on the correctness of the health information to be identified based on external relevant knowledge, identify whether the health information is false, and give a post-judgment explanation of the judgment result;

[0040] Among them, the process of identifying false health information through the knowledge breadth retrieval module and the knowledge depth reasoning module is consistent with the human information processing process.

[0041] In this embodiment, the false health information identification process through the knowledge breadth retrieval module and the knowledge depth reasoning module is consistent with the human information processing process, specifically: a large amount of intuitive knowledge stored in the human brain is modeled using a knowledge graph to obtain a knowledge base model; the first stage of human information processing, the intuitive process, is simulated to quickly obtain knowledge related to the health information to be identified and related to multiple aspects of health information from the knowledge base as an external knowledge set to assist in reasoning about the correctness of health information; the second stage of human information processing, the analysis process, is simulated to first focus on the comprehensive concept of health information-related knowledge entities, and then to mine the detailed information of related knowledge triples, and finally to analyze the correctness of the health information to be identified based on these clues to obtain a judgment result.

[0042] In this embodiment, health false information identification is a sub-branch of the false information identification field, which aims to identify knowledge related to food safety, medical health, care, etc. Compared with research such as false news identification, health false information usually lacks social features such as comments, forwarding, and likes, so it is necessary to mine other auxiliary information to improve the effect of health false information identification.

[0043] According to the purpose of the recognition algorithm, the current research on health false information recognition can be divided into two categories: the first type of research focuses on improving the accuracy of the classification algorithm without considering the user adoption behavior of the algorithm. This type of research usually explores the combination of various auxiliary information with health information content to improve the classification effect of the model. For example, by incorporating user behavior characteristics in social media into machine learning classifiers to improve model performance; considering web page structure and content attributes to assist decision-making in health false information recognition. The second type of research considers both the accuracy and interpretability of the model to improve user trust in the algorithm, which belongs to the field of explainable artificial intelligence research. Explanability is divided into two levels: interpretability and explainability, which refer to the understandability of algorithm logic and the provision of post-explanation respectively. At present, most of the research on health false information detection only focuses on and solves the interpretability of providing post-explanation. For example, a promising interpretable health false information detection method based on medical knowledge graph can identify the importance of different triples in the knowledge graph for information discrimination through attention mechanism, as a post-explanation of the information discrimination result. Interactive contradictions are screened out from the multimodality (text and image) of the article as a post-explanation of information discrimination. This type of model cannot show users how the information discrimination results are obtained, so it is still difficult to convince users. Expert information processing and decision-making are added to the health false information detection model to make the algorithm logic understandable. However, these methods require a lot of manpower and are difficult to promote and apply on a large scale.

[0044] In this embodiment, through the knowledge breadth retrieval module and the knowledge depth reasoning module, on the one hand, the freedom of knowledge graph application is improved, which can adapt to the dynamic update or replacement of knowledge graph; on the other hand, it can more fully call the diverse knowledge elements such as entities, triples, and subgraphs in the knowledge graph to improve the model effect and the feasibility of landing application. At the same time, two levels of interpretable health false information detection are realized, which enhances the user's full trust in this system.

[0045] In an optional embodiment, the specific structures of the knowledge breadth retrieval (KBR) module and the knowledge depth reasoning (KDR) module are as follows.

[0046] Preferably, the knowledge breadth retrieval module is used to simulate the first stage of human information processing, the intuitive process; the knowledge breadth retrieval module includes submodules: a knowledge extractor, a semantic similarity calculator and a dual filter;

[0047] The knowledge extractor selects the knowledge graph as the external knowledge base according to the health information to be identified, and quickly retrieves all triples related to the health information to be identified that contain the entity word as the head node or tail node from the external knowledge base through entity word matching, which is defined as the candidate triple set;

[0048] A semantic similarity calculator, which extracts the semantic features of the health information and the triples in each triple set, calculates the semantic similarity between each triple and the health information, and determines the triples similar to the health information;

[0049] The double filter distills the determined triples similar to the health information into a preset number of triple information based on the dual principles of greater semantic similarity and fairer knowledge distribution.

[0050] Wherein, the optional semantic similarity calculator specifically includes:

[0051] De-identified health information The triple information retrieved by the knowledge extractor is semantically represented to obtain the semantic vectors corresponding to the health information and triples. and h (h,r,t) ;

[0052] The cosine similarity between the health information to be determined and each triple information is calculated according to the semantic vectors of the two, and the triples similar to the health information are determined according to the set high similarity threshold and irrelevant threshold.

[0053] According to the set high similarity threshold and irrelevant threshold, the triples similar to the health information are determined, specifically:

[0054] If the cosine similarity between the triplet and the health information to be identified is greater than the high similarity threshold, the triplet and the health information to be identified are high-similarity triplet;

[0055] If the cosine similarity between the triplet and the health information to be identified is less than the irrelevant threshold, the triplet and the health information to be identified are irrelevant triplet, and all irrelevant triples in the candidate triplet set are eliminated.

[0056] Optional, dual filter, including:

[0057] For each piece of health information, a quantity threshold is set so that at most k triplets can be introduced from the candidate triple set;

[0058] Within the quantity threshold, select the triples in the candidate triple set whose semantic similarity with the health information is higher than the high similarity threshold, sort the semantic similarity from high to low, and add them to the external knowledge base that assists the correctness reasoning of health information in turn;

[0059] Within the quantity threshold, continue to remove the remaining triplets except irrelevant triplets from the set of candidate triples, group them according to entities, and sort each group from high to low according to cosine similarity, as well as the dual principles of fair screening of triplets related to different entity words. Add the triplet currently ranked first from all groups to the external knowledge base that assists in reasoning about the correctness of health information, and delete the triplet from the group until the number reaches k or all remaining triplets have been extracted.

[0060] In this embodiment, the problem of interpretable health false information identification is defined as:

[0061] Given a piece of health information The present invention introduces a rich external knowledge graph Used as auxiliary information to determine the correctness of health information At the same time, two levels of algorithm explainability are achieved. The first level requires that the algorithm's judgment process and judgment logic be understandable to users, and the other level requires that the algorithm provide a certain post-hoc explanation. Help users intuitively understand the algorithm's judgment results. represents a specific knowledge graph, Z represents the number of external knowledge graphs introduced, Represents the correctness judgment result of the information, and i represents the i-th piece of health information.

[0062] Specifically, the two stages of KBR and KDR based on knowledge graph perception are adopted, such as Figure 2 As shown in the figure, the knowledge breadth extraction stage corresponds to the intuitive heuristic process in the dual-process theory, and the knowledge depth reasoning stage corresponds to the deep reasoning process in the dual-process theory. Given an unverified health information, the knowledge breadth retrieval module (KBR) first extracts heuristic knowledge information from the knowledge graph comprehensively, accurately and efficiently through three sub-modules. Then, the knowledge depth reasoning module (KDR) deeply represents and infers the subtle clues in the heuristic information through a coarse-to-fine process to infer the credibility of the information. Finally, the dual-process theory-driven health false information identification model (KG2S) outputs the credibility judgment result and the judgment basis composed of these subtle clues.

[0063] Among them, the knowledge breadth retrieval (KBR) module corresponds to the heuristic process of the dual process theory, which aims to extract as much relevant heuristic knowledge as possible. In this module, the present invention uses the knowledge graph (KG) to simulate the memory storage of the human brain, and simulates the heuristic process of the dual process theory by quickly extracting relevant triple information in the knowledge graph. The knowledge breadth retrieval (KBR) module is refined into the following three sub-modules: Knowledge Extractor (KE), Semantic Similarity Calculator (SC) and Twofold Filter (TF).

[0064] Preferably, the knowledge extractor quickly obtains comprehensive external knowledge from multiple knowledge graphs. This submodule constructs the key-value pairs of "head node-triplet" and "tail node-triplet" to extract the triplets associated with entity words in all information at the fastest speed when a piece of information to be determined is obtained. The specific implementation process is as follows:

[0065] Given a knowledge graph E is an entity set, R is a relation set, and T is a triple set. A triple tr = (h, r, t) ∈ T is a piece of knowledge (e.g.,<jujube,is_a,high potassium foods> ,<jujube,alias,red date> ), where h∈E is the head entity, t∈E is the tail entity, and r∈R is the relationship between h and t.

[0066] The knowledge extractor consists of two key-value structure dictionaries based on backward and forward retrieval rules, namely:

[0067] B_Dict = {h j :{r k ,t k}|(h j ,r k ,t k )∈Ts} and F_Dict={t j :{h k ,r k}|(h k ,r k ,t j )∈Ts},

[0068] Among them, h j , t j Represents a head entity and a tail entity in the knowledge graph set, Ts represents All triples in . Thus, middle The corresponding knowledge triples can be quickly extracted into knowledge sets by the knowledge extractor (KE) in It is health information The set of all entity words in , τ j represents the jth entity.

[0069] In this embodiment, three public knowledge graphs, Cn-DBpedia\s, MedicalKG, and Covid19KG, are selected as external knowledge bases. In practical applications, other domain knowledge graphs can be selected and adapted according to the theme of health information.

[0070] Preferably, a semantic similarity calculator (SC) is used to calculate the retrieved triple information and the health information input by the model The cosine similarity of is used to quickly correct the retrieval results of the knowledge extractor (KE).

[0071] In addition to acquiring comprehensive external knowledge, another goal of the knowledge breadth retrieval (KBR) stage is to achieve the distillation of relevant knowledge and to modify and select the retrieved knowledge. For example, "apple" can refer to the fruit apple or the Apple company. When a piece of dietary information related to apples is input into the knowledge extractor (KE), all knowledge associated with the food "apple" and the company "apple" will be obtained through entity matching. The system needs to filter out the triples that are not related to the input dietary information. Less relevant knowledge may even interfere with the credibility judgment of subsequent information. The present invention solves the above-mentioned problems by adopting a semantic similarity calculator.

[0072] Specifically, the semantic similarity calculator (SC) is a lightweight "bert-as-service" tool that integrates the health information to be judged. And the knowledge set retrieved by the knowledge extractor (KE) The triple information in the candidate triple set is semantically represented to obtain the semantic vector corresponding to the health information and triples and h (h,r,t) ,in

[0073]

[0074]

[0075] Among them, BERT is an encoder that vectorizes text through "bert-as-service".

[0076] Further calculate the cosine similarity between the health information to be determined and each triple information:

[0077]

[0078] Among them, cosine similarity The larger the size, the healthier it is. The higher the semantic similarity with the lth triple.

[0079] Preferably, the double filter (TF) is based on the dual principles of more semantic similarity and fairer knowledge distribution, and finally distills a preset number of triplet information to the next stage for information representation and information discrimination of knowledge gain.

[0080] The semantic similarity calculator (SC) has been able to judge and filter the relevant knowledge triples for the next stage of information reasoning. However, further considering that different entities in a piece of health information have different numbers of triples linked in the knowledge graph, excessive attention to some entities may mislead the direction of information credibility reasoning. For example, when processing a piece of health information "Chronic nephritis patients should eat more red dates", the number of triplets related to "chronic nephritis" may reach hundreds, while the number of triplets related to "red dates" is only a few. Sending these unfair knowledge sets to the next stage for information reasoning will cause the model to pay too much attention to "chronic nephritis" and ignore the entity "red dates", which will interfere with the model's judgment of the correctness of the information; therefore, a double filter (TF) is used in this embodiment.

[0081] Specifically, the double filter (TF) implementation process is as follows:

[0082] For every piece of health information TF limits it to a maximum of Introduce Mt triples to construct the knowledge set for the second stage in KG2S

[0083] Under the condition of the upper limit k, first The high similarity triplets in SC are sorted from high to low according to the similarity calculated in SC and added to Then, the other related triplets except the irrelevant triplets are first grouped according to the entities, and each group is sorted from high to low according to the similarity. The first-ranked triplet is added to the entity group from all the groups in multiple rounds. The triplet is removed from the group until the number reaches k or all the remaining triples are extracted. All irrelevant triples do not participate in the next stage of knowledge reasoning.

[0084] In this embodiment, preferably, the knowledge depth reasoning module is used to simulate the second stage of human information processing, the analysis process; the knowledge depth reasoning module includes:

[0085] The knowledge feature extraction module extracts the external entity knowledge representation based on the entity knowledge corresponding to the health information and the triples obtained by the knowledge breadth retrieval module. External triplet knowledge representation and semantic representation of health information to be determined

[0086] Hierarchical attention module, representing external entity knowledge External triplet knowledge representation and semantic representation of health information to be determined Input the constructed hierarchical attention network to fuse health information semantics with knowledge to obtain knowledge-aware health information semantics;

[0087] The credibility prediction module determines whether the semantics of knowledge-perceived health information is false information and determines its credibility. When the health information obtains the correctness judgment result, the interpretable artificial intelligence algorithm can output the post-event explanation at the same time.

[0088] Specifically, based on the entity knowledge corresponding to the health information and the triple knowledge obtained by distillation in the previous stage, the knowledge is reasoned from coarse to fine layer by layer to infer the correctness of the health information. Entity knowledge refers to coarse-grained concepts. For example, the entity "chronic nephritis" integrates the overall connotation of multiple aspects of knowledge such as "symptoms" and "treatment methods"; triples refer to fine-grained knowledge, such as <chronic nephritis, cannot eat, high potassium food> represents a specific type of food that is taboo for patients with chronic nephritis. In order to realize the knowledge reasoning from coarse to fine for these two types of knowledge elements, the knowledge deep reasoning module includes: knowledge feature extraction and hierarchical attention module (HA).

[0089] Among them, preferably, the knowledge feature extraction is to use the RotatE knowledge graph embedding technology to embed the knowledge entity into a dense low-dimensional space to obtain its feature vector. RotatE defines each relationship as a rotation from the head entity to the tail entity in the vector space, so it can model the symmetric / antisymmetric, inversion and combination relationships between entities. The relationship of these details is very important for the judgment of health information. Taking antisymmetry as an example, "chronic nephritis can lead to hyperkalemia", but "hyperkalemia cannot lead to chronic nephritis", that is, the entities "chronic nephritis" and "hyperkalemia" have an antisymmetric relationship. Therefore, RotatE will effectively promote the knowledge representation of gain information discrimination. Specifically, health information Entities in Can be characterized as:

[0090]

[0091] in, express The corresponding entity e n feature embedding, de is the dimension of the entity feature vector.

[0092] For the extraction of triple knowledge features, the present invention extracts deep semantic features in triples based on BERT model fine-tuning. Specifically, for the triple set extracted by KBR Its deep semantic representation is:

[0093]

[0094] in, Represents the triple tr m The hidden state of the output layer, dtr is the dimension of the triplet feature vector.

[0095] At the same time, BERT fine-tuning is also used to extract the initial representation of health information. Given a piece of health information It is constructed as a word sequence W = {w [CLS] ,w 1 ,w 2 ,…,w O ,w [SEP]}, where O represents [CLS] and [SEP] are two marker words marking the beginning and end of a sentence. The semantic features of can finally be expressed as The calculation process is as follows:

[0096]

[0097] in, Indicates the corresponding word w o The hidden state of the output layer, dw is the dimension of the word feature vector.

[0098] Finally, the external entity knowledge representation External triplet knowledge representation and semantic representation of health information to be determined Together they serve as the input of the next module to perform information credibility reasoning.

[0099] Preferably, the hierarchical attention module (HA) obtains the external entity knowledge representation External triplet knowledge representation and semantic representation of health information to be determined After the semantic representation of these three types of information, the core of the hierarchical attention module is a hierarchical attention network, which realizes the reasoning and fusion of knowledge from coarse to fine through three multi-head self-attention layers (Self-Attention, SA) and two multi-head mutual attention layers (Cross-Attention, CA). Specifically:

[0100] (1) Multi-head self-attention layer (SA)

[0101] This embodiment includes three multi-head self-attention layers (SA), each of which performs further semantic mining on a certain type of information to obtain a better representation of context awareness. Specifically, and Input into three multi-head self-attention layers (SA) respectively, and finally output the enhanced representation and Where L, M, and N represent the number of attention blocks in the three multi-head self-attention layers. Taking the processing flow of coarse-grained entity knowledge representation as an example, the vth attention module of SA calculates the self-attention between entities through the following formula:

[0102]

[0103] in, is the updated representation of the coarse-grained entity knowledge obtained by the first v-1 multi-head self-attention modules, and w is The dimension of Represents different fully connected layers, and obtains the Query, Key, and Value used for each attention head calculation. Each It represents the importance score of the qth entity obtained by the tth head of the vth self-attention block to the pth entity representation. Finally, the entity representation is updated based on the attention score:

[0104]

[0105]

[0106] in, It is a fully connected feed-forward deep neural network, and layer_norm performs normalization.

[0107] (2) Multi-head mutual attention layer (CA);

[0108] The multi-head mutual attention layer (CA) realizes the association and information fusion between different types of information. The present invention sets two multi-head mutual attention layers (CA), in which the first mutual attention layer corrects the representation of fine-grained knowledge through coarse-grained knowledge guidance, so that the semantic representation of fine-grained knowledge perceives the global information of entity words. The second mutual attention layer corrects the representation of the information to be determined by fusing the fine-grained knowledge guidance of coarse-grained knowledge, so that the semantic vector of the information to be determined not only conveys the content of the information itself, but also incorporates the semantics of related knowledge entities and knowledge triples.

[0109] Specifically, the input of the first CA is the coarse-grained knowledge after enhanced representation and fine-grained knowledge The core is to identify the attention weights of different entity knowledge on triple knowledge, and then fuse entity knowledge based on the attention weights to correct the representation of triple knowledge. The formula for calculating the enhanced triple representation after attention and fusion is as follows:

[0110]

[0111]

[0112]

[0113] in, There are three fully connected layers to obtain the Query, Key and Value used for the mutual attention multi-head calculation. The other network structures are consistent with the above SA.

[0114] The input of the second CA is the fine-grained knowledge representation H enhanced by the coarse-grained knowledge C The semantic vector of health information obtained by SA Similarly, the CA identifies the attention weights of different triplet knowledge for information discrimination, and obtains the health information representation of multi-granularity knowledge perception based on the weight fusion, as follows:

[0115]

[0116]

[0117]

[0118] in, It can reflect the importance of different triples to the judgment of information credibility, so it is relatively easy to The most critical pieces of knowledge are selected and presented to the elderly as ex post facto explanations.

[0119] Finally, a pooling layer is used to It is represented as the health information feature vector C, that is, the feature vector corresponding to the [CLS] marker word, which is sent to the final credibility prediction module as the knowledge-aware health information semantic representation.

[0120] In this embodiment, the credibility prediction module obtains its credibility prediction result based on the knowledge-aware health information feature vector C obtained in the previous level attention module through two activation functions (RELU function and softmax function), specifically:

[0121] C′ i =RELU(W f1 C i +b 1 ) (16)

[0122]

[0123] in, Represents the predicted probability of the i-th information. Represents the true label of the i-th information, and uses the FocalLoss function to calculate the loss value:

[0124]

[0125]

[0126] Where N is the number of samples in the training set; As an adjustment factor, the loss function of easy-to-classify samples can be reduced, thereby increasing the loss ratio of difficult-to-classify samples.

[0127] In the above embodiment, when the health information obtains the correctness judgment result, the artificial intelligence algorithm can explain the output of the post-event explanation, including:

[0128] Obtaining a set of external knowledge from an external knowledge base to assist in reasoning about the correctness of health information as evidence for alternative ex post facto explanations;

[0129] The most important knowledge triples for judging the correctness of health information are mined based on the attention mechanism;

[0130] Organized according to key knowledge triads as ex post facto explanations for judging the correctness of health information;

[0131] While ensuring recognition accuracy, a higher level of explainability is achieved, which can promote the adoption of algorithm results by the elderly.

[0132] In this embodiment, for each piece of health information to be determined, KG2S can visualize its reasoning process after obtaining its prediction result, such as Figure 3As shown. The process is as follows: Given health information First, in the knowledge breadth retrieval (KBR) stage, all triples are matched from the knowledge graph through the knowledge extractor (KE), and then the semantic similarity calculator (SC) calculates the similarity between each triple and The semantic similarity of , and finally the double filter (TF) is used to filter and obtain the final six triples ( Figure 3 The six triples in the middle right half) enter the next stage of knowledge deep reasoning (KDR). KDR mainly completes the reasoning of the influence relationship between knowledge entities, triples, and health information. For example, the attention weight of the triple <chronic nephritis, can not eat, high potassium food> on health information is 0.474, which is much higher than other triples and is one of the important bases for judging the credibility of information.

[0133] Preferably, it also includes embedding a dual-process theory-driven health false information identification system into social media and search engines.

[0134] For example, KG2S is embedded into an online chat application as an application DEMO of the present invention, and the interface is designed as follows: Figure 4 As shown. The elderly only need to long press the message in the WeChat chat interface to access the information credibility judgment service. The service returns the judgment results and corresponding basis of health information to the chat interface, and users can make their own comprehensive judgments based on the returned results.

[0135] Preferably, a dual-process theory-driven health misinformation identification system is embedded in social media and search engines, including:

[0136] Embed a dual-process theory-driven health misinformation identification system into cloud servers of social media and search engines;

[0137] The information published on social media and search engines is used as information to be identified, and it is determined whether the information to be identified is false information;

[0138] Put a warning sign on the pages with false information;

[0139] Block or limit the flow of false information;

[0140] Post the post-event explanation to the corresponding page of the false information.

[0141] Example 2

[0142] A dual-process theory-driven approach to identifying health misinformation, ref. Figure 1 , including the following steps:

[0143] S100, corresponding to the intuitive process of the dual process theory, retrieves relevant knowledge of the health information to be identified in the external knowledge base;

[0144] S200, corresponding to the analysis stage of the dual-process theory, mines key external knowledge triples based on the attention mechanism module in the artificial intelligence algorithm as a post hoc explanation of the health information judgment results.

[0145] In this embodiment, an explainable artificial intelligence algorithm is used to perform deep reasoning on health information. Based on external relevant knowledge, deep reasoning is performed on the correctness of the health information to be identified, whether the health information is false, and a post-determination of the judgment result is given.

[0146] In the above step S100, the relevant knowledge of the health information to be identified is retrieved in the external knowledge base, including the following steps:

[0147] S101. According to the health information to be identified, all triples with the entity word contained in the health information as the head node or the tail node are quickly retrieved from the external knowledge graph according to the entity word matching, and defined as a set of candidate triples;

[0148] S102, extracting health information and semantic features of triples in each triple set;

[0149] S103, calculating the semantic similarity between each triple and the health information; setting a high similarity threshold, an irrelevant threshold and a quantity threshold;

[0150] S104, setting a high similarity threshold, an irrelevant threshold, and a quantity threshold;

[0151] S105, removing all triples in the candidate triple set whose semantic similarity with the health information is lower than an irrelevant threshold;

[0152] S106. Within the quantity threshold, select triples from the candidate triple set whose semantic similarity with the health information is higher than the high similarity threshold, and add them to the external knowledge set for assisting the correctness reasoning of the health information;

[0153] S107. Within the quantity threshold, continue to select the remaining triplets from the candidate triple set according to the dual principles of similarity from high to low and fair screening of triplets related to different entity words, and add triplets to the external knowledge set that assists in reasoning about the correctness of health information.

[0154] In the above step S200, an explainable artificial intelligence algorithm is used to perform deep reasoning on false information based on an external knowledge set, identify the correctness of health information, and obtain a post-event explanation, including the following steps:

[0155] S201, extracting semantic features of health information based on the health information;

[0156] S202, extracting semantic features of triples from an external knowledge set that assists in reasoning about correctness of health information;

[0157] S203, using an explainable artificial intelligence algorithm to fuse the triples and health information semantics, update the health information semantic features, and obtain the health information semantic representation of the fused knowledge;

[0158] S204. According to the processing results of the explainable artificial intelligence algorithm, determine whether the health information is false information based on the semantic representation of the health information based on the fused knowledge.

[0159] In the above step S204, when the health information obtains the correctness judgment result, the artificial intelligence algorithm output can be explained and the subsequent explanation includes the following steps:

[0160] S2041. Obtaining a set of external knowledge that assists in reasoning about the correctness of health information from an external knowledge base as evidence for alternative ex post facto explanations;

[0161] S2042. Mining the most important knowledge triples for determining the correctness of health information based on the attention mechanism;

[0162] S2043. Organize the key knowledge triples as a post facto explanation of the correctness of the health information.

[0163] After the above step S100, the method further includes embedding a health false information identification system driven by a dual-process theory into social media and search engines.

[0164] In this embodiment, a dual-process theory-driven health false information identification system is embedded in social media and search engines, including the following steps:

[0165] S300, embedding a dual-process theory-driven health misinformation identification system into cloud servers of social media and search engines;

[0166] S400, taking the information published by social media and search engines as information to be identified, and determining whether the information to be identified is false information;

[0167] S500, marking a warning sign for false information on a page containing false information;

[0168] S600: Block or limit the flow of false information;

[0169] S700. Post the subsequent explanation to the corresponding page of the false information.

[0170] In summary, compared with the prior art, the present invention is the first interpretable health false information detection method. It is based on a people-oriented perspective and aims to promote the acceptance of the health false information detection system by the elderly. It is innovative in terms of research objects and research objectives, and therefore has strong practical significance and application prospects. Compared with the general patent for interpretable false information detection, the present invention designs a two-stage model based on the dual-process theory, which is easy for the elderly to understand and achieves two-level interpretable model transparency; and by introducing a knowledge graph to simulate the knowledge memory storage system of the human brain, it effectively supports the development of the two-stage model, and at the same time solves the high requirements for professional knowledge in health false information detection.

[0171] Furthermore, compared with other published false information detection patents, the present invention, on the one hand, proposes a dual filter to solve the knowledge unfairness problem that arises when external knowledge is introduced into the false information detection system; on the other hand, it points out the simultaneous use of coarse-grained entity knowledge and fine-grained triple knowledge, and designs a coarse-to-fine hierarchical attention module to further improve the reasoning ability of the model.

[0172] Furthermore, through comparison with the baseline model, migration experiments, ablation experiments and user experiments, it is proved that the present invention has high accuracy and interpretability, and thus has high practical value; and based on the dual-process theory, the present invention has developed a transparent health false information identification system with higher interpretability, which is easier for the elderly to accept the algorithm process, understand the reasons for the judgment, and finally adopt the identification results, which truly helps the elderly to resist online health false information and screen correct health information, and has important social significance; the present invention can be easily embedded in search engines, social media (such as WeChat, Douyin) and other applications that are frequently used by the elderly in their daily lives. As a convenient tool for the elderly to identify the credibility of health information, it has considerable economic benefits.

[0173] Example 3

[0174] A computer-readable medium stores a computer program, wherein the computer program includes program instructions. When the program instructions are executed by a processor of an electronic device, the processor executes a method for identifying false health information driven by a dual-process theory.

[0175] Example 4

[0176] An electronic device includes: a processor and a memory, the memory is used to store computer program code, the computer program code includes computer instructions, when the processor executes the computer instructions, the electronic device executes a health false information identification method driven by a dual-process theory.

[0177] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0178] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. Those skilled in the art can also clearly understand that the descriptions of the various embodiments of the present invention have different focuses. For the convenience and brevity of description, the same or similar parts may not be repeated in different embodiments. Therefore, for parts not described or not described in detail in a certain embodiment, refer to the records of other embodiments.

[0179] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0180] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0181] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0182] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrations. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a digital versatile disc (DVD)), or a semiconductor medium (eg, a solid state disk (SSD)).

[0183] A person skilled in the art can understand that to implement all or part of the processes in the above-mentioned embodiments, the processes can be completed by a computer program to instruct the relevant hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned method embodiments. The aforementioned storage medium includes: a read-only memory (ROM) or a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

Claims

1. A dual-process theory-driven health false information identification system, characterized by: include: The knowledge breadth retrieval module, corresponding to the intuitive process of the dual-process theory, is used to retrieve relevant knowledge of the health information to be identified in the external knowledge base; The knowledge deep reasoning module corresponds to the analysis stage of the dual-process theory. It is used to conduct deep reasoning on the correctness of the health information to be identified based on external relevant knowledge, identify whether the health information is false, and give a post-judgment explanation of the judgment result; Among them, the health false information identification process through the knowledge breadth retrieval module and the knowledge depth reasoning module is consistent with the human information processing process; The knowledge breadth retrieval module is used to simulate the first stage of human information processing, the intuitive process; the knowledge breadth retrieval module includes submodules: knowledge extractor, semantic similarity calculator and dual filter; The knowledge extractor selects the knowledge graph as the external knowledge base according to the health information to be identified, and quickly retrieves all triples related to the health information to be identified that contain the entity word as the head node or tail node from the external knowledge base through entity word matching, which is defined as the candidate triple set; A semantic similarity calculator, which calculates the semantic similarity between each triple and the health information by extracting the semantic features of the health information and the triples in each triple set, and determines the triples similar to the health information; Double filter, which distills the determined triples similar to health information into a preset number of triples based on the dual principles of more semantic similarity and fairer knowledge distribution; Knowledge deep reasoning module, including: The knowledge feature extraction module extracts the external entity knowledge representation based on the entity knowledge corresponding to the health information and the triples obtained by the knowledge breadth retrieval module. External triple knowledge representation and semantic representation of health information to be determined Hierarchical attention module, representing external entity knowledge External triple knowledge representation and semantic representation of health information to be determined Input the constructed hierarchical attention network to fuse health information semantics with knowledge to obtain knowledge-aware health information semantics; The credibility prediction module determines whether the semantics of health information perceived by knowledge is false information and determines its credibility. When the health information obtains the correctness judgment result, the artificial intelligence algorithm can be explained and output the post-event explanation at the same time; The dual-process theory-driven health misinformation identification system also includes embedding a dual-process theory-driven health misinformation identification system into social media and search engines: Embed a dual-process theory-driven health misinformation identification system into cloud servers of social media and search engines; The information published on social media and search engines is used as information to be identified, and it is determined whether the information to be identified is false information; Put a warning sign on the pages with false information; Block or limit the flow of false information; Post the post-event explanation to the corresponding page for false information determination.

2. A dual-process theory-driven health false information identification system according to claim 1, characterized in that: Semantic similarity calculator, including: De-identified health information The triple information retrieved by the knowledge extractor is semantically represented to obtain the semantic vectors corresponding to the health information and triples. and h (h, r ,t) ; The cosine similarity between the health information to be determined and each triple information is calculated based on the semantic vectors of the two, and the triples similar to the health information are determined based on the set high similarity threshold and irrelevant threshold.

3. A dual-process theory-driven health false information identification system according to claim 2, characterized in that: According to the set high similarity threshold and irrelevant threshold, triples similar to health information are determined, including: If the cosine similarity between the triplet and the health information to be identified is greater than the high similarity threshold, the triplet and the health information to be identified are high-similarity triplet; If the cosine similarity between the triplet and the health information to be identified is less than the irrelevant threshold, the triplet and the health information to be identified are irrelevant triplet, and all irrelevant triples in the candidate triplet set are eliminated.

4. A dual-process theory-driven health false information identification system according to claim 1, characterized in that: Dual filter, including: For each piece of health information, a quantity threshold is set so that at most k triplets can be introduced from the candidate triple set; Within the quantity threshold, select the triples whose semantic similarity with the health information is higher than the high similarity threshold in the candidate triple set, sort the semantic similarity from high to low, and add them to the external knowledge base for assisting the correctness reasoning of health information in turn; Within the quantity threshold, continue to remove the remaining triplets except irrelevant triplets from the set of candidate triples, group them according to entities, and sort each group from high to low according to cosine similarity, as well as the dual principles of fair screening of triplets related to different entity words. Add the triplet currently ranked first from all groups to the external knowledge base that assists in reasoning about the correctness of health information, and delete the triplet from the group until the number reaches k or all remaining triplets have been extracted.

5. A dual-process theory-driven health false information identification system according to claim 1, characterized in that: When the health information is judged to be correct, the explainable AI algorithm can also output post-event explanations, including: Obtaining a set of external knowledge from an external knowledge base to assist in reasoning about the correctness of health information as evidence for alternative ex post facto explanations; The most important knowledge triples for judging the correctness of health information are mined based on the attention mechanism; Organize according to key knowledge triples as ex post facto explanation for correctness judgment of health information.

6. A dual-process theory-driven method for identifying health false information, characterized in that: include: The step of knowledge breadth retrieval, corresponding to the intuitive process of dual-process theory, is used to retrieve relevant knowledge of the health information to be identified in the external knowledge base; The step of deep knowledge reasoning corresponds to the analysis stage of the dual-process theory, which is used to conduct deep reasoning on the correctness of the health information to be identified based on external relevant knowledge, identify whether the health information is false, and give a post-event explanation of the judgment result; Among them, the process of identifying false health information through knowledge breadth retrieval and knowledge depth reasoning is consistent with the human information processing process; Knowledge breadth retrieval is used to simulate the first stage of human information processing, the intuitive process; knowledge breadth retrieval includes: knowledge extractor, semantic similarity calculator and dual filter; The knowledge extractor selects the knowledge graph as the external knowledge base according to the health information to be identified, and quickly retrieves all triples related to the health information to be identified that contain the entity word as the head node or tail node from the external knowledge base through entity word matching, which is defined as the candidate triple set; A semantic similarity calculator, which calculates the semantic similarity between each triple and the health information by extracting the semantic features of the health information and the triples in each triple set, and determines the triples similar to the health information; Double filter, which distills the determined triples similar to health information into a preset number of triples based on the dual principles of more semantic similarity and fairer knowledge distribution; Knowledge deep reasoning includes: Perform knowledge feature extraction, based on the entity knowledge corresponding to the health information and the triples retrieved from the knowledge breadth, and extract the external entity knowledge representation External triple knowledge representation and semantic representation of health information to be determined Hierarchical attention, representing external entity knowledge External triple knowledge representation and semantic representation of health information to be determined Input the constructed hierarchical attention network to fuse health information semantics with knowledge to obtain knowledge-aware health information semantics; Make credibility predictions, judge whether the semantics of knowledge-perceived health information is false information, determine its credibility, and when the health information obtains the correctness judgment result, the interpretable artificial intelligence algorithm can output the post-event explanation at the same time; The dual-process theory-driven health misinformation identification approach also includes embedding a dual-process theory-driven health misinformation identification system into social media and search engines: Embed a dual-process theory-driven health misinformation identification system into cloud servers of social media and search engines; The information published on social media and search engines is used as information to be identified, and it is determined whether the information to be identified is false information; Put a warning sign on the pages with false information; Block or limit the flow of false information; Post the post-event explanation to the corresponding page for false information determination.

7. A computer-readable medium, characterized in that The computer-readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor of an electronic device, the processor executes a dual-process theory-driven health false information identification method as described in claim 6.