Health rumor detection method, device and equipment and storage medium

Abstract

The application discloses a health rumor detection method and device, equipment and a storage medium, and relates to the technical field of natural language processing. The method comprises the following steps: constructing a text causal graph based on medical entities in a health medical rumor text to be detected and causal relationships between the medical entities; performing strength estimation on the text causal graph, and generating a causal strength graph according to the strength estimation result; determining the causal strength credibility in the causal strength graph through a medical knowledge graph, and generating an enhanced text causal graph based on the causal strength credibility; performing feature coding on the enhanced text causal graph by using an edge-enhanced graph attention network, and obtaining graph structure information; and performing health rumor detection on the health medical rumor text to be detected based on the graph structure information, and obtaining a health rumor detection result. By using the technical scheme, the technical problem that a health rumor detection method in the prior art relies on specific text patterns and is difficult to understand complex medical causal logic is solved.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, and in particular to methods, devices, equipment and storage media for detecting health rumors. Background Technology

[0002] The internet has become an important source of health information. However, while this convenience has lowered the barrier to medical consultation, the deficiencies in the information supervision system have also led to the rapid spread of health rumors. Such false information can easily influence individual health decisions and hinder epidemic prevention and control. Therefore, it is urgent to effectively detect and monitor such rumors.

[0003] Currently, existing methods for detecting health rumors mainly rely on label-specific features trained on known rumor instances to classify texts by capturing shallow semantics and statistical patterns. However, when faced with health rumors that deliberately distort professional terminology and pseudoscientific logic, this detection method struggles to understand complex medical causal logic, resulting in low accuracy in detecting health rumors. Summary of the Invention

[0004] The main purpose of this application is to provide a method, device, equipment and storage medium for detecting health rumors, in order to solve the technical problem that existing methods for detecting health rumors rely on specific text patterns and are difficult to understand complex medical causal logic.

[0005] To achieve the above objectives, this application proposes a method for detecting health rumors, the method comprising:

[0006] Construct a text causal graph based on the medical entities in the text of the health and medical rumors to be detected and the causal relationships between the medical entities;

[0007] Intensity estimation is performed on the text causal graph, and a causal intensity graph is generated based on the intensity estimation results;

[0008] The credibility of causal strength in the causal strength graph is determined by using a medical knowledge graph, and an enhanced text causal graph is generated based on the credibility of causal strength.

[0009] An edge-enhanced graph attention network is used to encode the features of the enhanced text causal graph to obtain graph structure information;

[0010] Based on the graph structure information, the text containing the health and medical rumors to be detected is used to perform health rumor detection, and the health rumor detection results are obtained.

[0011] In one embodiment, the step of constructing a text causal graph based on medical entities in the text of the health and medical rumors to be detected and the causal relationships between the medical entities includes:

[0012] A preset medical language model is used to extract medical entities from the text of the health and medical rumors to be detected, thereby obtaining the medical entities in the text of the health and medical rumors to be detected.

[0013] The causal relationships between the medical entities are determined by a pre-defined greedy fast causal inference algorithm.

[0014] A text causal graph is constructed based on the medical entity and the causal relationship.

[0015] In one embodiment, the step of performing intensity estimation on the text causal graph and generating a causal intensity graph based on the intensity estimation result includes:

[0016] The intensity of the text causal graph is estimated, and the causal association strength corresponding to each edge in the text causal graph is determined based on the intensity estimation results.

[0017] Based on the strength of the causal relationship, the edges are assigned strengths to generate a causal strength graph.

[0018] In one embodiment, the step of determining the causal strength confidence level in the causal strength graph using a medical knowledge graph, and generating an enhanced text causal graph based on the causal strength confidence level, includes:

[0019] Determine the semantic similarity between knowledge graph entities in the medical knowledge graph and causal graph entities in the causal intensity graph;

[0020] Based on the semantic similarity, entity alignment is performed between the knowledge graph entity and the causal graph entity to obtain the aligned causal graph entity.

[0021] Generate an aligned causal graph based on the aligned causal graph entities;

[0022] The causal strength confidence level in the aligned causal graph is determined based on the entity association relationships between entities in the aligned causal graph.

[0023] Determine the comprehensive credibility parameters based on the aforementioned causal strength credibility.

[0024] The causal intensity graph is updated based on the comprehensive credibility parameters to generate an enhanced text causal graph.

[0025] In one embodiment, the step of determining the causal strength confidence level in the aligned causal graph based on the entity association relationships between entities in the aligned causal graph includes:

[0026] If the entity association relationship between the entities in the aligned causal graph is a direct association relationship, then the causal strength confidence level in the aligned causal graph is determined according to the preset direction consistency index.

[0027] Alternatively, if the entity association relationship between the aligned causal graph entities is an indirect association relationship, then the semantic coherence and directional consistency of the entity paths between the aligned causal graph entities are determined, and the causal strength confidence level in the aligned causal graph is determined based on the semantic coherence and the directional consistency.

[0028] In one embodiment, the step of using an edge-enhanced graph attention network to perform feature encoding on the enhanced text causal graph to obtain graph structure information includes:

[0029] An edge-enhanced graph attention network is used to jointly encode the causal strength and knowledge verification credibility in the enhanced text causal graph to obtain edge features;

[0030] Encode the node features and edge features in the augmented text causal graph to determine the attention score corresponding to the augmented text causal graph;

[0031] The attention scores are normalized to generate the attention distribution for all nodes in the enhanced text causal graph.

[0032] Based on the attention distribution and the edge features, the central node in the enhanced text causal graph is aggregated with its neighbor nodes, and graph structure information is generated based on the aggregated nodes.

[0033] In one embodiment, the step of detecting health rumors based on the graph structure information and obtaining the health rumor detection result includes:

[0034] The graph structure information is mapped to a multilayer perceptron;

[0035] When the mapping is complete, the attention weights corresponding to the node embeddings in the enhanced text causal graph are determined through the attention pooling mechanism;

[0036] The attention weights and the aggregated nodes are weighted and summed to generate the graph-level representation corresponding to the node embedding;

[0037] The graph-level representation is input into a classifier so that the classifier outputs health rumor detection results based on the graph-level representation.

[0038] Furthermore, to achieve the above objectives, this application also proposes a health rumor detection device, the device comprising:

[0039] The causal graph construction module is used to construct a text causal graph based on medical entities in the text of health and medical rumors to be detected and the causal relationships between those medical entities.

[0040] The causal intensity estimation module is used to estimate the intensity of the text causal graph and generate a causal intensity graph based on the intensity estimation results.

[0041] A credibility determination module is used to determine the credibility of causal intensity in the causal intensity graph through a medical knowledge graph, and to generate an enhanced text causal graph based on the credibility of causal intensity.

[0042] The feature encoding module is used to perform feature encoding on the enhanced text causal graph using an edge-enhanced graph attention network to obtain graph structure information;

[0043] The rumor detection module is used to perform health rumor detection on the text of the health and medical rumor to be detected based on the graph structure information, and obtain the health rumor detection result.

[0044] In addition, to achieve the above objectives, this application also proposes a health rumor detection device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the health rumor detection method described above.

[0045] Furthermore, to achieve the above objectives, this application also proposes a storage medium that is a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the health rumor detection method described above.

[0046] This application provides a method for detecting health rumors. The method involves constructing a text causal graph based on medical entities and causal relationships within the text containing the health rumor to be detected; estimating the strength of the text causal graph and generating a causal strength graph based on the estimation results; determining the causal strength credibility of the causal strength graph using a medical knowledge graph and generating an enhanced text causal graph based on the causal strength credibility; using an edge-enhanced graph attention network to encode the enhanced text causal graph to obtain graph structure information; and performing health rumor detection on the text containing the health rumor to be detected based on the graph structure information to obtain the detection result. Compared to existing health rumor detection methods that primarily rely on label-specific features trained on known rumor instances, which struggle to understand complex medical causal logic when faced with health rumors that deliberately distort professional terminology and pseudoscientific logic, this invention constructs a text causal graph based on medical entities and their causal relationships within the text containing the health rumor to be detected, and then uses the graph structure information in the enhanced text causal graph corresponding to the text causal graph to detect the health rumor to be detected. This solves the technical problem of existing health rumor detection methods relying on specific text patterns and struggling to understand complex medical causal logic, thereby improving the accuracy of health rumor detection. Attached Figure Description

[0047] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart illustrating an embodiment of the health rumor detection method of this application.

[0050] Figure 2 This is a diagram illustrating the entity causal relationships output by the pre-set greedy fast causal inference algorithm in the health rumor detection method of this application.

[0051] Figure 3 This is a schematic diagram of the causal strength estimation process in the health rumor detection method of this application;

[0052] Figure 4 This is a dual-graph fusion framework of causal graph and medical knowledge graph in the health rumor detection method of this application;

[0053] Figure 5This is a flowchart illustrating Embodiment 2 of the health rumor detection method of this application;

[0054] Figure 6 This is a diagram illustrating the entity alignment strategy and testing status in the health rumor detection method of this application;

[0055] Figure 7 This is a schematic diagram of the minimum model for credibility estimation in the health rumor detection method of this application;

[0056] Figure 8 This is a flowchart illustrating Embodiment 3 of the health rumor detection method of this application;

[0057] Figure 9 This is the framework for an edge-enhanced dual-graph fusion classifier in the health rumor detection method of this application;

[0058] Figure 10 This is a schematic diagram of the module structure of the health rumor detection device according to an embodiment of this application;

[0059] Figure 11 This is a schematic diagram of the hardware operating environment involved in the health rumor detection method in this application embodiment.

[0060] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0061] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0062] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0063] Because existing health rumor detection methods mainly rely on label-specific features trained on known rumor instances, they struggle to understand complex medical causal logic when faced with health rumors that deliberately distort professional terminology and pseudoscientific logic.

[0064] This application provides a solution that can construct a text causal graph based on medical entities and their causal relationships in the text of health and medical rumors to be detected, and use the graph structure information in the enhanced text causal graph corresponding to the text causal graph to detect health and medical rumors. This solves the technical problem that existing health rumor detection methods rely on specific text patterns and have difficulty understanding complex medical causal logic, thereby improving the accuracy of health rumor detection.

[0065] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or health rumor detection device capable of performing the above functions. The following description uses a health rumor detection device (hereinafter referred to as the device) as an example to illustrate this embodiment and the subsequent embodiments.

[0066] Based on this, embodiments of this application provide a method for detecting health rumors, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the health rumor detection method of this application.

[0067] In this embodiment, the health rumor detection method includes steps S10 to S50:

[0068] Step S10: Construct a text causal graph based on the medical entities in the text of the health and medical rumors to be detected and the causal relationships between the medical entities.

[0069] It is understood that the aforementioned health and medical rumor texts to be detected may contain text content that may contain false, misleading, or unverified medical and health information (in accordance with natural laws), such as antibiotics for treating viral infections, bone broth for calcium supplementation, etc. This embodiment does not impose any restrictions on this.

[0070] It should be understood that the aforementioned medical entities can be specific nouns or terms related to medical health in the text of the health-related medical rumors to be detected. They can represent specific concepts, objects, or things in the medical field, such as disease names, treatment methods, biomarkers, body parts or organs, nutrients, or food components. This embodiment does not impose any limitations on this. Correspondingly, the causal relationship between medical entities can be the causal connection between two or more medical entities in the text of the health-related medical rumors to be detected. For example, disease and cause: such as "smoking causes lung cancer"; drug and efficacy: such as "penicillin treats bacterial infections"; treatment method and disease improvement: such as "chemotherapy shrinks tumors"; biomarker and disease risk: such as "high blood sugar increases the risk of diabetes". This embodiment does not impose any limitations on this.

[0071] It should be noted that the above-mentioned text causal graph can be a graph used to represent the logical connections and potential causal relationships between medical entities. It can display medical entities in the text and their causal relationships through nodes and edges, where nodes represent medical entities and edges represent causal relationships between medical entities.

[0072] Specifically, step S10 includes: extracting medical entities from the text of the health and medical rumors to be detected using a preset medical language model to obtain the medical entities in the text of the health and medical rumors to be detected; determining the causal relationship between the medical entities using a preset greedy fast causal inference algorithm; and constructing a text causal graph based on the medical entities and the causal relationship.

[0073] Understandably, the aforementioned pre-defined medical language model can be a model used to identify and extract medical entities from texts containing health and medical rumors, such as a pre-trained BioBERT model. The pre-trained BioBERT model is a pre-trained language model based on the BERT (Bidirectional Encoder-Representation Transformer) architecture. It can be optimized for the biomedical field and its performance in biomedical text mining tasks can be improved by pre-training on large-scale biomedical literature data. In practical applications, the pre-trained BioBERT model can accurately identify key medical entities such as disease names, genes, and drugs in texts containing health and medical rumors, thereby obtaining the medical entities within the texts.

[0074] It should be noted that the aforementioned pre-set greedy fast causal inference algorithm can be used to identify causal relationships between medical entities by analyzing medical data. In practical applications, the device can learn the causal structure from medical data using the pre-set greedy fast causal inference algorithm, and finally construct a text causal graph by using medical entities as nodes and causal relationships as edges. This embodiment, by analyzing the causal graph using the pre-set greedy fast causal inference algorithm, can intuitively see the causal relationships between variables, helping detection personnel to discover possible logical loopholes or irrationalities in rumors, and providing a more in-depth basis for rumor detection.

[0075] In practical applications, automatically extracted causal graphs may contain noisy edges that contradict medical logic. To improve the credibility of the causal graph, this embodiment can further design dual constraint rules to filter out noisy edges, including: a) Medical logic constraint: Some entities and relationships in medicine have explicit logic. In this case, symptom entities are not allowed to point to disease, drug, or treatment entities as causes, as such relationships violate the inherent logic of cause-symptom in the medical knowledge base. b) Temporal constraint: Considering that causal relationships usually precede results in time, and symptom descriptions are usually written in chronological order of events, the temporal order of description can be regarded as a constraint. If factor A appears after B in most cases, an edge from A to B will not be allowed. It should be noted that the temporal constraint does not mean that there is an edge from B to A, because temporal order is not a sufficient condition for causality.

[0076] In this embodiment, refer to Figure 2 , Figure 2This is a diagram illustrating the entity causal relationships output by the greedy fast causal inference algorithm in the health rumor detection method of this application. (Example:) Figure 2 As shown, Figure 2 The text describes four types of uncertain relations contained in the partial ancestor graph output by a pre-defined greedy fast causal inference algorithm: —>, <—>, and Where A—>B means A causes B, and A<—>B means there is an unobserved confounding factor between A and B. This indicates that either A causes B, or there is an unobserved confounding factor. This indicates that either A causes B, or B causes A, or there is an unobserved confounding factor. In this embodiment, definite edges can be retained or removed, while two types of uncertain edges can be handled differently. With a probability of 1 / 2, retain or remove. The decision is made with a 1 / 3 probability of keeping, removing, or reversing the selection.

[0077] Step S20: Perform intensity estimation on the text causal graph and generate a causal intensity graph based on the intensity estimation results.

[0078] It should be noted that the aforementioned causal strength diagram can be a visual graphic used to quantify the strength of causal relationships between medical entities. It can display the strength of causal associations between medical entities (such as diseases, drugs, symptoms, etc.) in the health and medical rumor text to be detected through numerical and visual methods, thereby helping the testers to intuitively assess the credibility of the causal relationship.

[0079] Specifically, step S20 includes: performing intensity estimation on the text causal graph, and determining the causal association strength corresponding to each edge in the text causal graph based on the intensity estimation result; assigning intensity to each edge based on the causal association strength to generate a causal intensity graph.

[0080] It should be understood that the aforementioned causal relationship strength can be considered as the strength of the causal effect between the medical entities corresponding to each edge in the text causal graph.

[0081] In practical applications, since the resulting graph contains noise, the sampled causal graph can be optimized by estimating the strength of the learned causal relationships. This embodiment assigns higher strengths to edges with strong causal effects, and assigns near-zero strengths to edges that do not represent causal relationships or have weak effects. This embodiment considers the strength of T→Y in the text causal graph G as the average treatment effect. And it uses propensity score matching for measurement:

[0082]

[0083] in, t represents the medical instance most similar to i in the opposite group. i y i , z i These are the processing value, result, and confounding factors for instance i, respectively.

[0084] Reference Figure 3 , Figure 3 This is a schematic diagram of the causal strength estimation process in the health rumor detection method of this application. Figure 3 As shown, entities can first be extracted from language text (such as health and medical rumors to be detected), and the causal relationships between these entities can be learned. Then, causal graph sampling is performed to generate several candidate causal graphs. The strength of these causal graphs is estimated, and the strength is assigned based on the causal association strength corresponding to each edge in the causal graph to generate a causal strength graph.

[0085] Step S30: Determine the causal strength confidence level in the causal strength graph through a medical knowledge graph, and generate an enhanced text causal graph based on the causal strength confidence level.

[0086] It should be noted that the aforementioned medical knowledge graph can be a knowledge graph in the biomedical field, which can structure and organize entities in the medical field (such as diseases, symptoms, drugs, treatment methods, genes, proteins, etc.) and the relationships between these entities.

[0087] It should be noted that the aforementioned causal strength confidence level can be a parameter used to characterize the truthfulness of causal strength in the causal strength graph. Correspondingly, the aforementioned enhanced text causal graph can be a causal graph obtained by adding the causal strength confidence level to the causal strength graph.

[0088] Step S40: Use an edge-enhanced graph attention network to encode the features of the enhanced text causal graph to obtain graph structure information.

[0089] In this embodiment, the Edge-enhanced Graph Attention Network (EGAT) is an improved model of Graph Attention Network (GAT). It optimizes information transmission and feature aggregation by enhancing the weights or importance of edges in the graph, thereby improving the accuracy and robustness of health rumor detection. In practical applications, EGAT can dynamically adjust edge weights, allowing the model to focus more on important edges and nodes, thus more accurately capturing key information in the rumor propagation process. This dynamic adjustment mechanism helps reduce noise and interference, improving the robustness of rumor detection. Furthermore, through the edge enhancement mechanism, EGAT can better aggregate feature information from neighboring nodes, generating more discriminative node representations, which helps the model more accurately identify abnormal patterns or inconsistencies in rumor content.

[0090] It should be noted that the above graph structure information can be a holistic expression of information such as node relationships and key features in the enhanced text causal graph, such as node feature representation, edge weight information, subgraph structure features, and global graph features. This embodiment does not limit this.

[0091] Step S50: Based on the graph structure information, perform health rumor detection on the text of the health and medical rumor to be detected, and obtain the health rumor detection result.

[0092] In this embodiment, after obtaining the graph structure information corresponding to the enhanced text causal graph, the graph structure information can be mapped to the classification space, and the text can be classified by a classifier to output the classification result of health rumors, including rumors or non-rumors.

[0093] In the specific implementation, refer to Figure 4 , Figure 4 This is a dual-graph fusion framework of causal graph and medical knowledge graph in the health rumor detection method of this application. For example... Figure 4 As shown, Figure 4 This paper demonstrates the overall architecture of a dual-graph fusion framework (CKDG) based on causal logic and knowledge graphs, aiming to detect health rumors through collaborative reasoning using causal graphs and knowledge graphs. First, it receives input rumor text and uses a pre-trained BioBERT model to perform medical entity recognition, extracting medical entities and key information from the rumor text. Then, the Greedy Fast Causal Inference (GFCI) algorithm is used to determine the causal relationships between these medical entities, constructing a causal graph with medical entities as nodes and causal relationships as edges. Simultaneously, a medical knowledge graph (such as CMeKG) can be introduced to align entities in the causal graph with entities in the knowledge graph. Prior knowledge from the knowledge graph is used to verify the credibility of causal relationships in the causal graph, including direct relationship verification, multi-hop path verification, and no-association checks. Finally, edge-enhanced graph attention network (EGRET) and hierarchical pooling strategies are used to dynamically fuse causal strength and knowledge verification signals, achieving multi-granularity reasoning from local paths to global semantics, ultimately enabling accurate identification of health rumors.

[0094] This embodiment provides a method for detecting health rumors. The method discloses the following steps: constructing a text causal graph based on medical entities and causal relationships within the text containing the health rumor to be detected; estimating the strength of the text causal graph and generating a causal strength graph based on the estimation results; determining the causal strength credibility of the causal strength graph using a medical knowledge graph and generating an enhanced text causal graph based on the causal strength credibility; using an edge-enhanced graph attention network to encode the features of the enhanced text causal graph to obtain graph structure information; and performing health rumor detection on the text containing the health rumor to be detected based on the graph structure information to obtain the health rumor detection result. Compared to existing health rumor detection methods that mainly rely on label-specific features trained on known rumor instances, which struggle to understand complex medical causal logic when faced with health rumors that deliberately distort professional terminology and pseudoscientific logic, this invention constructs a text causal graph based on medical entities and their causal relationships within the text containing the health rumor to be detected, and then uses the graph structure information in the enhanced text causal graph corresponding to the text causal graph to perform health rumor detection on the text containing the health rumor to be detected. This solves the technical problem of existing health rumor detection methods relying on specific text patterns and struggling to understand complex medical causal logic, thereby improving the accuracy of health rumor detection.

[0095] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 5 , Figure 5 This is a flowchart illustrating Embodiment 2 of the health rumor detection method of this application.

[0096] In this embodiment, step S30 includes steps S301 to S306:

[0097] Step S301: Determine the semantic similarity between knowledge graph entities in the medical knowledge graph and causal graph entities in the causal intensity graph.

[0098] It is understood that the aforementioned knowledge graph entities can be the basic units that constitute a medical knowledge image, such as disease entities, symptom entities, drug entities, treatment method entities, etc., and this embodiment does not impose any restrictions on this.

[0099] It should be understood that the above semantic similarity can be the degree of closeness between the medical concepts, meanings, or significance expressed by knowledge graph entities and causal graph entities at the semantic level.

[0100] Step S302: Align the knowledge graph entities and the causal graph entities according to the semantic similarity to obtain aligned causal graph entities.

[0101] It should be noted that the aligned causal graph entities described above can be entities mapped to the knowledge graph. In practical applications, the goal of entity alignment is to align entity e in the causal graph... i With entity e′ in the knowledge graph i Matching is performed. In this embodiment, the semantic similarity matrix between entities can be defined as follows: in Represents entity e in a cause-effect graph i With knowledge graph entity e′ i Semantic similarity between them.

[0102] In practical applications, refer to Figure 6 , Figure 6 This diagram illustrates the entity alignment strategy and testing status in the health rumor detection method of this application. Figure 6 As shown, the matching process in this embodiment can be divided into three steps: exact matching, fuzzy matching, and misalignment marking. In the entity alignment process, exact matching can first be used to directly match entities with exactly the same name using a hash table, making... Then, fuzzy matching is used to calculate the similarity θ between entities that do not pass the exact match using semantic vectors. e,e′ Entity pairs with a matching degree higher than a set threshold θ are retained, and... For θ e,e′ Finally, unmatched entities are marked as "Missing", thus completing entity alignment.

[0103] Step S303: Generate an aligned causal graph based on the aligned causal graph entities.

[0104] It is understandable that the above aligned causal graph can be a causal graph constructed from aligned causal graph entities.

[0105] Step S304: Determine the causal strength confidence level in the aligned causal graph based on the entity association relationships between the entities in the aligned causal graph.

[0106] It should be noted that after entity alignment is completed, each edge e in the causal graph can be verified using a knowledge graph. i →e j The confidence level R of causal strength can be determined by designing a function for R. In this embodiment, we can first focus on a minimal model graph, referring to... Figure 7 , Figure 7 This is a schematic diagram of the minimum model for credibility estimation in the health rumor detection method of this application. (Example:) Figure 7 As shown in the figure, there are four entities: e i ,e j Let e′ be an entity in the causal graph. i ,e′j To align with entities in the knowledge graph, φ(e) is further defined. i →e j ) represents external domain knowledge constraints, ψ(e) i →e j (This is an estimate of the original causal strength) θ e,e ′ represents the semantic similarity of entity alignment.

[0107] In practical applications, based on the gating mechanism concept, we construct a credibility estimation function R(·), whose design follows the principles of knowledge dominance and directionality. Under the principle of knowledge dominance, the external domain knowledge constraint φ, as prior information, should dominate the φ regulation process. When φ = 0, the function should completely suppress the credibility of causal strength. Simultaneously, the function needs to capture the positive and negative correlation characteristics between the knowledge path direction and the causal direction.

[0108] In this embodiment, starting from the product-gated structure, we can establish R = ψ·g(φ, θ) e,e′ The basic form of ) is where the gating function g(·) must satisfy the dominance of φ. Considering the semantic drift problem generated during entity alignment, the two endpoint alignment quality index can be integrated here. and The arithmetic mean of the values ​​is used as the overall alignment confidence score to mitigate the impact of single-endpoint errors on global confidence. To further address the direction issue, this embodiment allows φ to take negative values ​​when the knowledge path direction conflicts with the causal direction, and normalizes it at the end of the calculation. Through the above derivation, the analytical expression of the gating function is finally obtained: g(φ, θ) = avg(θ)·φ.

[0109] This leads to the complete credibility estimation function:

[0110]

[0111] After completing the design of the causal strength confidence estimation function R, it is necessary to further explore the application of φ(e′) in the knowledge graph. i →e′ j (The situation is as follows.)

[0112] Specifically, step S304 includes: if the entity association relationship between the aligned causal graph entities is a direct association relationship, then determine the causal strength confidence level in the aligned causal graph based on a preset directional consistency index; or, if the entity association relationship between the aligned causal graph entities is an indirect association relationship, then determine the semantic coherence and directional consistency corresponding to the entity paths between the aligned causal graph entities, and determine the causal strength confidence level in the aligned causal graph based on the semantic coherence and the directional consistency.

[0113] It should be noted that the above direct association can be the direct association between entities in the aligned causal graph; the above indirect association can be the association between entities in the aligned causal graph that do not have a direct association but have a multi-hop path association.

[0114] In practical applications, for causal edge e i →e j If a knowledge graph triple (e′) exists i ,r,e′ j )∈R k And r is a direct causal chain, that is, entity e. i and e j If a direct correlation exists, a directional consistency index can be defined, and the causal strength confidence level in the aligned causal graph can be determined based on the directional consistency index. In this embodiment, the directional consistency index is as follows:

[0115]

[0116] Where, λ pos and λ neg This is a penalty coefficient used to reflect the difference in credibility between forward and reverse verification.

[0117] When the aligned causal graph entities do not have direct associations in the knowledge graph but have multi-hop paths (such as...), In other words, when the entity relationships between entities in the aligned causal graph are indirect, the semantic coherence and directional consistency of the path can be evaluated, and the causal strength confidence in the aligned causal graph can be determined based on semantic coherence and directional consistency. In this embodiment, to efficiently search for multi-hop paths, a bidirectional breadth-first search algorithm can be used. The steps of this algorithm are as follows:

[0118] a) Path search: from e′ i and e′ j Starting simultaneously, they expand along the relationships in the knowledge graph until the two search domains intersect or the maximum number of hops L is reached. max .

[0119] b) Path filtering: Only retain paths with the same direction (i.e., e′) i →…→e′ j or e′ j ←…←e′ i ). Path φ(e′) i →e′ j Modeling through semantic decay factor:

[0120]

[0121] Where, σ(r) k The relation type weight is r ∈ [0,1].k The relationship type is γ, where γ is the path length attenuation coefficient, |P| is the path hop count, and L is the path length attenuation coefficient. max The maximum allowed path length is 1 - |P| / L max This represents the dilution effect of penalizing long paths on causal strength. Meanwhile, φ path A positive value is taken when the path direction is consistent with the causal direction; otherwise, a negative value is taken.

[0122] Furthermore, if there is no explicit association between the knowledge graph entities corresponding to the causal edges (i.e., the aligned causal graph entities), the causal strength confidence can be dynamically adjusted by the global knowledge graph coverage:

[0123]

[0124] Where Deg(·) represents the entity degree, N k Let φ be the total number of entities in the knowledge graph, and α be the baseline penalty term. This formula reflects unobserved relationships in highly connected subgraphs (e.g., no association between two common disease entities), and is more falsifiable than sparse subgraphs (e.g., rare drugs and symptoms). Specifically, if φ null A low value indicates that the corresponding entity e′ i and e′ j In knowledge graphs, if the data is dense and there are no multi-hop paths between entities, it may indicate that the relationships between entities are very low. Conversely, if φ null A higher value indicates that the corresponding entity e′ i and e′ j This is considered sparse data in knowledge graphs, which is likely due to incomplete data in knowledge graphs, rather than low correlation between entities.

[0125] In this embodiment, the marking of missing entities has been completed by the upstream task; here, the penalty function for the specific missing entities is implemented:

[0126] φ null =exp(-β·ψ)·I missing ;

[0127] Where β is the penalty coefficient for missing entities, I missing This is an indicator function for when there are unaligned entities at the endpoints of a causal edge.

[0128] Step S305: Determine the comprehensive credibility parameter based on the causal strength credibility.

[0129] It should be noted that, considering the possibility that there may be both a direct relationship and multiple multi-hop paths between two entities, we define the final φ(e′) of this type of causal edge as follows: i →e′ jIntegrate the following situations and adjust for contradictions in evidence:

[0130]

[0131] Where, K={φ dir ,φ path} represents the set of credibility for each verification scenario. The variance term Var(·) is used to penalize contradictory evidence and is used by the logarithmic enhancement term log(1+ψ) to amplify the influence of edges with high causal strength.

[0132] Step S306: Update the causal intensity graph based on the comprehensive credibility parameter to generate an enhanced text causal graph.

[0133] In this embodiment, after calculating the comprehensive confidence parameter, φ can be substituted into R(·) and normalized to obtain the confidence of the corresponding edge. Based on this, the credibility parameter (i.e., the comprehensive credibility parameter mentioned above) is added to the original causal strength estimation graph to form an enhanced graph structure representation G. c (i.e., the enhanced text causal graph mentioned above), in

[0134] This embodiment discloses a method for determining the semantic similarity between knowledge graph entities in a medical knowledge graph and causal graph entities in a causal strength graph; aligning the knowledge graph entities and causal graph entities based on semantic similarity to obtain aligned causal graph entities; generating an aligned causal graph based on the aligned causal graph entities; determining the causal strength credibility in the aligned causal graph based on the entity association relationships between the aligned causal graph entities; determining a comprehensive credibility parameter based on the causal strength credibility; and updating the causal strength graph based on the comprehensive credibility parameter to generate an enhanced text causal graph. Since this embodiment can deeply integrate the causal graph with the medical knowledge graph through a knowledge-guided credibility estimation method, and verify the credibility of the relationships in the causal graph through pre-verified knowledge in the medical knowledge graph, the accuracy of health rumor detection can be further improved.

[0135] Based on the first and / or second embodiments of this application, in the third embodiment of this application, the content that is the same as or similar to the above embodiments can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 8 , Figure 8 This is a flowchart illustrating Embodiment 3 of the health rumor detection method of this application.

[0136] In this embodiment, step S40 includes steps S401 to S404:

[0137] Step S401: Use an edge-enhanced graph attention network to perform joint feature encoding on the causal strength and knowledge verification credibility in the enhanced text causal graph to obtain edge features.

[0138] It should be noted that enhanced text causal graphs Graph attention networks (GAT) contain both node and edge features; however, the standard GAT framework does not explicitly incorporate edge features into its computation. To overcome this limitation and fully utilize the structural information encoded by edge features, this embodiment employs a variant of GAT, namely the Edge-Enhanced Graph Attention Network (EGRET). Unlike traditional GAT, EGRET effectively utilizes the structural information encoded by edge features during the aggregation and attention score calculation phases, thereby systematically improving the model's ability to leverage edge feature information.

[0139] In this embodiment, to reduce the dimensionality of subsequent calculations, the causal intensity ψ in the causal intensity graph of the confidence estimation can be used here. ij And knowledge verification credibility R ij Joint encoding as edge feature ξ ij Their dimensions correspond to ψ ij With R ij Then, the two types of features are combined and standardized using the following method:

[0140] ξ ij =σ(α·ψ ij +(1-α)·R ij );

[0141] Where α is a learnable parameter and σ is the Sigmoid function, used to dynamically balance the two types of signals.

[0142] Step S402: Encode the node features and edge features in the augmented text causal graph to determine the attention score corresponding to the augmented text causal graph.

[0143] It should be noted that this embodiment can use the last hidden state of BioBERT as the causal graph node v. i ∈V c initial features This approach inherits BioBERT's semantic understanding capabilities in medical texts. Furthermore, to further enhance the semantic representation of nodes, this embodiment can perform semantic processing on aligned knowledge graph entities e′. i ∈G k The node features described above are obtained by embedding using TransE and fusing it with the representation from BioBERT:

[0144]

[0145] in, It is entity e′ i TransE embedding vector, W i It is a learnable parameter matrix used to align the knowledge graph embeddings into the semantic space of BioBERT.

[0146] In this embodiment, for each edge e ij When calculating the attention score, in addition to the node feature h i h j In addition, the edge feature ξ of the directed edge from node i to node j is also included. ij The attention score, which depends on both nodes and edges, is calculated as follows:

[0147]

[0148] Where, d ij This is the denormalized representation of the attention score, || represents the concatenation operation, and W a W v W e These are all learnable parameters, used to linearly transform the features of nodes and edges, respectively.

[0149] Step S403: Normalize the attention scores to generate the attention distribution corresponding to all nodes in the enhanced text causal graph.

[0150] It should be noted that when calculating the attention score d for each edge... ij After that, you can continue with {d} ij |j∈N i} Perform softmax normalization to generate all nodes j∈N i The probability distribution of (where i is the center node), i.e., the attention distribution {α} ij |j∈N i}:

[0151]

[0152] Step S404: Aggregate the neighbor nodes of the central node in the enhanced text causal graph based on the attention distribution and the edge features, and generate graph structure information based on the aggregated nodes.

[0153] It should be understood that aggregating neighbor nodes for the central node involves aggregating the features of the neighbor nodes corresponding to the central node to the central node. In this embodiment, in order to fully utilize the edge feature ξ... ij The potential of aggregating neighbor node features {h j |j∈N iThe edge features are also included, where node i is the center node:

[0154]

[0155] Edge eigenvector ξ ij Before aggregation, ξ ij Apply a linear transformation. Here, σ(·) is the activation function, and the output of σ(·) is related to h. i The feature representations preceding node i are concatenated to generate a new representation. That is, the graph structure information output by the edge feature enhancement graph attention layer.

[0156] Further, step S50 includes: mapping the graph structure information to a multilayer perceptron; when the mapping is completed, determining the attention weights corresponding to the node embeddings in the enhanced text causal graph through an attention pooling mechanism; performing a weighted summation of the attention weights and the aggregated nodes to generate a graph-level representation corresponding to the node embeddings; and inputting the graph-level representation into a classifier so that the classifier outputs health rumor detection results based on the graph-level representation.

[0157] It should be noted that this embodiment can use a Multilayer Perceptron (MLP) as the classifier. In practical applications, considering the hierarchical features of graph-level semantics in health rumor detection tasks, node-level embeddings need to be aggregated into a graph-level representation first. Furthermore, this embodiment can introduce an attention pooling mechanism to dynamically capture the contribution differences of key nodes. Specifically, for the causal graph G... c Embedded in all nodes Through learnable query vectors Calculate node-level attention weights:

[0158]

[0159] in, and The tanh function, used as the projection parameter, enhances the nonlinear expressive power, and the graph-level representation of h... G Generate by weighted summation:

[0160]

[0161] This mechanism assigns greater weight to nodes with high causal strength, aligning with the task characteristic of rumor detection, which requires focusing on the core causal chain. Subsequently, h GThe input to the MLP undergoes a nonlinear transformation through two fully connected layers. The first layer employs a dimensionality reduction strategy, compressing d-dimensional features to d / 4 dimensions to eliminate redundant information. The second layer maps to a binary classification probability space. Simultaneously, GeLU activation and batch normalization are introduced between layers, mathematically expressed as:

[0162] y pred =Softmax(W2·GeLU(BN(W1h) G +b1)+b2));

[0163] Among them, y ored This represents the predicted probability of rumor detection. and For trainable parameters, BN stands for batch normalization.

[0164] This design enhances the model's ability to discriminate complex causal logic through hierarchical feature abstraction, while avoiding the risk of overfitting.

[0165] In this embodiment, the model's training objective is jointly optimized by multi-task losses, including the main classification loss, knowledge consistency constraint, and sparse regularization term. The main loss employs Label Smoothing Cross Entropy (LS-CE), which mitigates the overconfidence problem caused by class imbalance by introducing a smoothing factor ∈.

[0166]

[0167] Among them, y c For real labels, p c To predict probabilities, and to further enhance the semantic alignment between causal paths and the knowledge graph, a knowledge consistency loss term L is designed. kg Used to constrain attention weight α ij With causal credibility Distribution consistency:

[0168]

[0169] Here, ∈ = le⁻⁸ is the numerical stability term. This loss term forces the model to rely more heavily on high-confidence edges of knowledge verification during the decision-making process. Furthermore, L2 regularization is applied to the MLP parameters. To control model complexity. The final total loss function is a weighted sum of the three factors:

[0170] L total =λ1L cls +λ2L kg +λ3L reg ;

[0171] The weighting coefficients λ1, λ2, and λ3 will be determined through Pareto front analysis to balance classification performance with the robustness of knowledge logic.

[0172] In practical applications, refer to Figure 9 , Figure 9 This is the framework for an edge-enhanced dual-graph fusion classifier in the health rumor detection method of this application. For example... Figure 9 As shown, for graph attention computation with edge enhancement, the causal graph G can be input first. c and knowledge graph G k Input causal graph Among them, V c Let E represent the medical entities in the text containing the health and medical rumors to be detected. c Indicates causal relationships between entities. Indicates the strength of causality. This represents the credibility of knowledge verification; the input knowledge graph G k =(V k ,P k ), where V k P represents a knowledge entity. k This represents the causal relationships between knowledge entities. Then, in this embodiment, node feature encoding and edge feature encoding can be performed on the causal graph. Specifically, the initial features of the causal graph nodes can first be extracted using the BioBERT model. Furthermore, by combining the knowledge graph entity features embedded in TransE, and through the learnable parameter matrix V... align Aligning it to the semantic space of BioBERT generates enhanced node feature representations; simultaneously, causal strength can be... and knowledge verification credibility Joint encoding as edge feature ξ ij As shown by ξ1, ξ2, and ξ3 in the figure. When calculating the attention score, for each edge e... ij It can be combined with node features h i and h j and edge features ξ ij Calculate attention score d ij The LeakyReLU activation function is used, and the attention weights α are obtained by softmax normalization. ij However, the device can incorporate edge features ξ while aggregating neighbor node features. ij New node representations are generated through linear transformation and activation functions. Finally, node-level embeddings can be aggregated into a graph-level representation using a multilayer perceptron. This graph-level representation is then input into a two-layer fully connected network. The first layer employs a dimensionality reduction strategy, and the second layer maps to a binary classification probability space. GeLU activation and batch normalization are used to enhance the model's non-linear expressive power. The final output layer outputs the predicted probability y for rumor detection. pred The output is mapped to the [0, 1] interval using the Softmax function, representing the probability of a rumor and a non-rumor.

[0173] This embodiment discloses a method for jointly encoding the causal strength and knowledge verification credibility in an enhanced text causal graph using an edge-enhanced graph attention network to obtain edge features; encoding the node features and edge features in the enhanced text causal graph to determine the attention score corresponding to the enhanced text causal graph; normalizing the attention score to generate the attention distribution corresponding to all nodes in the enhanced text causal graph; aggregating the neighbor nodes of the central node in the enhanced text causal graph based on the attention distribution and edge features, and generating graph structure information based on the aggregated nodes; since this embodiment can introduce an edge-enhanced attention mechanism and a hierarchical pooling strategy, dynamically fuse causal strength and knowledge verification signals, and combine local path feature aggregation and global semantic distillation, it can improve the model's ability to represent complex causal topologies and its detection robustness.

[0174] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the health rumor detection method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0175] This application also provides a health rumor detection device, please refer to... Figure 10 The health rumor detection device includes:

[0176] Causal graph construction module 10 is used to construct a text causal graph based on medical entities in the text of health and medical rumors to be detected and the causal relationships between the medical entities;

[0177] The causal intensity estimation module 20 is used to estimate the intensity of the text causal graph and generate a causal intensity graph based on the intensity estimation result;

[0178] The credibility determination module 30 is used to determine the credibility of causal intensity in the causal intensity graph through a medical knowledge graph, and generate an enhanced text causal graph based on the credibility of causal intensity.

[0179] Feature encoding module 40 is used to perform feature encoding on the enhanced text causal graph using an edge-enhanced graph attention network to obtain graph structure information;

[0180] The rumor detection module 50 is used to perform health rumor detection on the text of the health and medical rumor to be detected based on the graph structure information, and obtain the health rumor detection result.

[0181] The health rumor detection device provided in this application, employing the health rumor detection method described in the above embodiments, can solve the technical problem that existing health rumor detection methods rely on specific text patterns and are difficult to understand complex medical causal logic. Compared with the prior art, the beneficial effects of the health rumor detection device provided in this application are the same as those of the health rumor detection method provided in the above embodiments, and other technical features in the health rumor detection device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0182] This application provides a health rumor detection device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the health rumor detection method in Embodiment 1 above.

[0183] The following is for reference. Figure 11 The diagram illustrates a structural schematic suitable for implementing the health rumor detection device in the embodiments of this application. The health rumor detection device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 11 The health rumor detection device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of this application.

[0184] like Figure 11As shown, the health rumor detection device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the health rumor detection device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. The communication device 1009 allows the health rumor detection device to communicate wirelessly or wiredly with other devices to exchange data. While the figures show health rumor detection devices with various systems, it should be understood that implementing or having all of the systems shown is not required. More or fewer systems may be implemented alternatively.

[0185] The health rumor detection device provided in this application, employing the health rumor detection method described in the above embodiments, can solve the technical problem of health rumor detection. Compared with the prior art, the beneficial effects of the health rumor detection device provided in this application are the same as those of the health rumor detection method described in the above embodiments, and other technical features of this health rumor detection device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0186] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0187] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0188] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the health rumor detection method described in the above embodiments.

[0189] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0190] The aforementioned computer-readable storage medium may be included in the health rumor detection device; or it may exist independently and not assembled into the health rumor detection device.

[0191] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by the health rumor detection device, the health rumor detection device: constructs a text causal graph based on medical entities and causal relationships between them in the text containing the health and medical rumor to be detected; performs intensity estimation on the text causal graph and generates a causal intensity graph based on the intensity estimation results; determines the causal intensity credibility in the causal intensity graph through a medical knowledge graph and generates an enhanced text causal graph based on the causal intensity credibility; uses an edge-enhanced graph attention network to perform feature encoding on the enhanced text causal graph to obtain graph structure information; and performs health rumor detection on the text containing the health and medical rumor to be detected based on the graph structure information to obtain health rumor detection results.

[0192] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0193] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0194] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described health rumor detection method. This solves the technical problem that existing health rumor detection methods rely on specific text patterns and struggle to understand complex medical causal logic. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the health rumor detection method provided in the above embodiments, and will not be repeated here.

[0195] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for detecting health rumors, characterized in that, The method includes: Construct a text causal graph based on the medical entities in the text of the health and medical rumors to be detected and the causal relationships between the medical entities; Intensity estimation is performed on the text causal graph, and a causal intensity graph is generated based on the intensity estimation results; The credibility of causal strength in the causal strength graph is determined by using a medical knowledge graph, and an enhanced text causal graph is generated based on the credibility of causal strength. An edge-enhanced graph attention network is used to encode the features of the enhanced text causal graph to obtain graph structure information; Based on the graph structure information, the text of the health and medical rumor to be detected is used to detect health rumors, and the health rumor detection results are obtained. The step of determining the causal strength confidence level in the causal strength graph using a medical knowledge graph, and generating an enhanced text causal graph based on the causal strength confidence level, includes: Determine the semantic similarity between knowledge graph entities in the medical knowledge graph and causal graph entities in the causal intensity graph; Based on the semantic similarity, entity alignment is performed between the knowledge graph entity and the causal graph entity to obtain the aligned causal graph entity. Generate an aligned causal graph based on the aligned causal graph entities; The causal strength confidence level in the aligned causal graph is determined based on the entity association relationships between entities in the aligned causal graph. Determine the comprehensive credibility parameters based on the aforementioned causal strength credibility. The causal intensity graph is updated based on the comprehensive credibility parameters to generate an enhanced text causal graph; The step of using an edge-enhanced graph attention network to encode features of the enhanced text causal graph to obtain graph structure information includes: An edge-enhanced graph attention network is used to jointly encode the causal strength and knowledge verification credibility in the enhanced text causal graph to obtain edge features; Encode the node features and edge features in the augmented text causal graph to determine the attention score corresponding to the augmented text causal graph; The attention scores are normalized to generate the attention distribution for all nodes in the enhanced text causal graph. Based on the attention distribution and the edge features, the central node in the enhanced text causal graph is aggregated with its neighbor nodes, and graph structure information is generated based on the aggregated nodes.

2. The method as described in claim 1, characterized in that, The step of constructing a text causal graph based on medical entities in the text of the health and medical rumors to be detected and the causal relationships between the medical entities includes: A preset medical language model is used to extract medical entities from the text of the health and medical rumors to be detected, thereby obtaining the medical entities in the text of the health and medical rumors to be detected. The causal relationships between the medical entities are determined by a pre-defined greedy fast causal inference algorithm. A text causal graph is constructed based on the medical entity and the causal relationship.

3. The method as described in claim 1, characterized in that, The step of performing intensity estimation on the text causal graph and generating a causal intensity graph based on the intensity estimation result includes: The intensity of the text causal graph is estimated, and the causal association strength corresponding to each edge in the text causal graph is determined based on the intensity estimation results. Based on the strength of the causal relationship, the edges are assigned strengths to generate a causal strength graph.

4. The method as described in claim 1, characterized in that, The step of determining the causal strength confidence level in the aligned causal graph based on the entity association relationships between entities in the aligned causal graph includes: If the entity association relationship between the entities in the aligned causal graph is a direct association relationship, then the causal strength confidence level in the aligned causal graph is determined according to the preset direction consistency index. Alternatively, if the entity association relationship between the aligned causal graph entities is an indirect association relationship, then the semantic coherence and directional consistency of the entity paths between the aligned causal graph entities are determined, and the causal strength confidence level in the aligned causal graph is determined based on the semantic coherence and the directional consistency.

5. The method as described in claim 1, characterized in that, The step of detecting health rumors based on the graph structure information and obtaining the detection results includes: The graph structure information is mapped to a multilayer perceptron; When the mapping is complete, the attention weights corresponding to the node embeddings in the enhanced text causal graph are determined through the attention pooling mechanism; The attention weights and the aggregated nodes are weighted and summed to generate the graph-level representation corresponding to the node embedding; The graph-level representation is input into a classifier so that the classifier outputs health rumor detection results based on the graph-level representation.

6. A health rumor detection device, characterized in that, The device includes: The causal graph construction module is used to construct a text causal graph based on medical entities in the text of health and medical rumors to be detected and the causal relationships between those medical entities. The causal intensity estimation module is used to estimate the intensity of the text causal graph and generate a causal intensity graph based on the intensity estimation results. A credibility determination module is used to determine the credibility of causal intensity in the causal intensity graph through a medical knowledge graph, and to generate an enhanced text causal graph based on the credibility of causal intensity. The feature encoding module is used to perform feature encoding on the enhanced text causal graph using an edge-enhanced graph attention network to obtain graph structure information; The rumor detection module is used to perform health rumor detection on the text of the health and medical rumor to be detected based on the graph structure information, and obtain the health rumor detection result; The credibility determination module is further configured to: determine the semantic similarity between knowledge graph entities in the medical knowledge graph and causal graph entities in the causal intensity graph; align the knowledge graph entities and the causal graph entities according to the semantic similarity to obtain aligned causal graph entities; generate an aligned causal graph based on the aligned causal graph entities; determine the causal intensity credibility in the aligned causal graph based on the entity association relationships between the aligned causal graph entities; determine a comprehensive credibility parameter based on the causal intensity credibility; and update the causal intensity graph based on the comprehensive credibility parameter to generate an enhanced text causal graph. The feature encoding module is further configured to use an edge-enhanced graph attention network to jointly encode the causal strength and knowledge verification credibility in the enhanced text causal graph to obtain edge features; encode the node features and the edge features in the enhanced text causal graph to determine the attention score corresponding to the enhanced text causal graph; normalize the attention score to generate the attention distribution corresponding to all nodes in the enhanced text causal graph; aggregate the neighbor nodes of the central node in the enhanced text causal graph based on the attention distribution and the edge features, and generate graph structure information based on the aggregated nodes.

7. A health rumor detection device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the health rumor detection method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the health rumor detection method as described in any one of claims 1 to 5.

Images