News credibility evaluation method, news credibility evaluation device, storage medium and equipment

Abstract

The embodiment of the invention relates to a news credibility evaluation method and device, a storage medium and electronic equipment, and relates to the technical field of human-computer interaction. The news credibility assessment method comprises the steps of matching to-be-assessed news with a pre-constructed object knowledge graph to obtain source tracing information of a target object described by the to-be-assessed news; the object knowledge graph is used for storing the traceability information of each object in a structured manner; extracting multi-modal content features corresponding to the news to be evaluated; and according to the traceability information and the multi-modal content features in combination with the text expression mode of the to-be-evaluated news, carrying out credibility evaluation on the to-be-evaluated news to obtain a credibility evaluation result. According to the method and the device, more accurate, objective and explainable credibility evaluation can be carried out on the news content, and the accuracy and the intelligent level of news content credibility evaluation are improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the field of human-computer interaction, and more particularly, embodiments of the present disclosure relate to a news credibility evaluation method, a news credibility evaluation device, a computer readable storage medium and an electronic device. BACKGROUND

[0002] This section is intended to provide background or context to the embodiments of the present disclosure and as such description herein can not be construed as admitting that any of the information included in this section is prior art.

[0003] With the rapid development of the Internet and social media, the speed and range of news dissemination have exploded. However, problems such as fake news and misleading information have also arisen, causing many adverse effects on society. SUMMARY

[0004] Traditional news credibility evaluation methods mainly rely on manual review, which is inefficient and difficult to cope with massive news data.

[0005] Existing solutions begin to introduce machine learning and natural language processing technologies to assist in evaluation, but these methods mostly focus on text content analysis, ignoring the value contained in image, video, audio and other modal information. Moreover, it is difficult to deeply mine information about the original source and transmission path of information during the evaluation process.

[0006] Therefore, there is a great need for a news event tracing and credibility evaluation system based on an AIGC (Artificial Intelligence Generated Content) multi-modal large model, which can analyze the credibility of news reports in real time, track information sources, and conduct comprehensive verification of news events.

[0007] In this context, embodiments of the present disclosure aim to provide a news credibility evaluation method, a news credibility evaluation device, a computer readable storage medium and an electronic device.

[0008] According to a first aspect of the embodiments of the present disclosure, a news credibility evaluation method is provided, comprising:

[0009] matching the news to be evaluated with a pre-constructed object knowledge graph to obtain the tracing information of the target object described by the news to be evaluated; the object knowledge graph is used to store the tracing information of each object in a structured manner;

[0010] extracting multi-modal content features corresponding to the news to be evaluated;

[0011] According to the traceability information, the multi-modal content features, and the text expression mode of the news to be evaluated, credibility of the news to be evaluated is evaluated, and a credibility evaluation result is obtained.

[0012] In an optional implementation, the method further includes:

[0013] Obtaining multi-source multi-modal report content for the target object;

[0014] According to the traceability information, the multi-modal content features, the multi-source multi-modal report content, and the text expression mode of the news to be evaluated, credibility of the news to be evaluated is evaluated, and a credibility evaluation result is obtained.

[0015] In an optional implementation, the object knowledge graph is constructed by:

[0016] Obtaining an object context of the object, the object context including at least one of the following: object historical records, object associated person relationships, and object background information;

[0017] Knowledge modeling is performed on the object context, and an object knowledge framework is obtained;

[0018] Knowledge extraction is performed on the object knowledge framework, and multi-source knowledge elements are obtained;

[0019] Knowledge fusion is performed on the multi-source knowledge elements, and the object knowledge graph is obtained.

[0020] In an optional implementation, the knowledge modeling performed on the object context to obtain the object knowledge framework includes:

[0021] Data cleaning processing is performed on the object context, and a standard object context is obtained;

[0022] Knowledge modeling is performed based on the standard object context, and the object knowledge framework is obtained;

[0023] The data cleaning processing includes at least one of the following: time standardization processing, place name remapping processing, and ambiguity elimination processing.

[0024] In an optional implementation, the knowledge extraction performed on the object knowledge framework to obtain the multi-source knowledge elements includes:

[0025] Text information in the object knowledge framework is subjected to knowledge extraction, and text knowledge elements are obtained;

[0026] Image information in the object knowledge framework is subjected to knowledge extraction, and visual knowledge elements are obtained;

[0027] knowledge extraction is performed on video information in the object knowledge framework to obtain audio-visual knowledge elements.

[0028] In an optional implementation, the knowledge fusion is performed on the multi-source knowledge elements to obtain the object knowledge graph, including:

[0029] conflict resolution processing is performed on the multi-source knowledge elements to obtain conflict-free knowledge elements;

[0030] entity alignment processing is performed on candidate entities with potential corresponding relationships in the conflict-free knowledge elements to obtain an entity alignment result;

[0031] knowledge normalization modeling is performed based on the entity alignment result to obtain the object knowledge graph.

[0032] In an optional implementation, the conflict resolution processing is performed on the multi-source knowledge elements to obtain the conflict-free knowledge elements, including:

[0033] a conflict type existing in the multi-source knowledge elements is identified;

[0034] conflict resolution processing is performed on the multi-source knowledge elements according to a conflict resolution strategy corresponding to the conflict type to obtain the conflict-free knowledge elements.

[0035] In an optional implementation, when the conflict type includes a time conflict, the conflict resolution strategy corresponding to the conflict type includes a time decision strategy based on information source credibility;

[0036] when the conflict type includes a location conflict, the conflict resolution strategy corresponding to the conflict type includes a location decision strategy based on multi-modal evidence fusion;

[0037] when the conflict type includes a participant conflict, the conflict resolution strategy corresponding to the conflict type includes a subject decision strategy based on entity credibility evaluation.

[0038] In an optional implementation, the entity alignment processing is performed on the candidate entities with potential corresponding relationships in the conflict-free knowledge elements to obtain the entity alignment result, including:

[0039] entity similarity between any two of the candidate entities is calculated;

[0040] the candidate entities are clustered according to the entity similarity to obtain the entity alignment result.

[0041] In an optional implementation, the news to be evaluated contains at least two of the following: text content, image content, video content, and audio content.

[0042] The extracting the multi-modal content features corresponding to the news to be evaluated comprises:

[0043] The semantic feature extraction is performed on the text content to obtain text features;

[0044] The visual feature analysis is performed on the image content to obtain image features;

[0045] The spatio-temporal feature modeling analysis is performed on the video content to obtain video features;

[0046] The acoustic feature analysis is performed on the audio content to obtain audio features;

[0047] The multi-modal feature fusion is performed on the text features, the image features, the video features and the audio features to obtain the multi-modal content features.

[0048] In an optional implementation, the credibility evaluation of the news to be evaluated according to the provenance information, the multi-modal content features, the multi-source multi-modal report content and the text expression mode of the news to be evaluated comprises:

[0049] The content consistency verification is performed on the provenance information and the news to be evaluated to obtain a first verification result;

[0050] The content consistency verification is performed on the multi-modal content features and a credible data source to obtain a second verification result;

[0051] The content consistency verification is performed on the multi-source multi-modal report content and the news to be evaluated to obtain a third verification result;

[0052] The content self-consistency verification is performed on the news to be evaluated according to the text expression mode of the news to be evaluated to obtain a fourth verification result;

[0053] The credibility evaluation result is determined according to the first verification result, the second verification result, the third verification result and the fourth verification result.

[0054] In an optional implementation, the content self-consistency verification of the news to be evaluated according to the text expression mode of the news to be evaluated to obtain a fourth verification result comprises:

[0055] The text logic of the news to be evaluated is analyzed to obtain a logic analysis result;

[0056] The event causal relationship associated with the news to be evaluated is verified to obtain a causal relationship verification result;

[0057] Integrate the logical analysis result and the causal relationship verification result to obtain the fourth verification result.

[0058] In an optional implementation, the method further includes:

[0059] Obtaining standard record information of the target object in a cross-language trusted data source;

[0060] Translating the news to be evaluated into a target language corresponding to the standard record information;

[0061] Performing content consistency comparison between the translated news to be evaluated and the standard record information to obtain a fifth verification result;

[0062] According to the first verification result, the second verification result, the third verification result, the fourth verification result, and the fifth verification result, determining the credibility evaluation result.

[0063] In an optional implementation, after obtaining the credibility evaluation result, the method further includes:

[0064] According to the credibility evaluation result and key features of the news to be evaluated, generating a news label corresponding to the news to be evaluated;

[0065] Uploading the news to be evaluated and the news label to a blockchain, so that each verification node in the blockchain performs credibility evaluation on the news to be evaluated again based on the news label and real-time dynamic data sources.

[0066] According to a second aspect of the embodiments of the present disclosure, a news credibility evaluation device is provided, including:

[0067] An acquisition module is configured to match news to be evaluated with a pre-constructed object knowledge graph to obtain traceability information of a target object described by the news to be evaluated; the object knowledge graph is configured to store traceability information of each object in a structured manner.

[0068] A feature extraction module is configured to extract multi-modal content features corresponding to the news to be evaluated.

[0069] A credibility evaluation module is configured to perform credibility evaluation on the news to be evaluated according to the traceability information, the multi-modal content features, and a text expression mode of the news to be evaluated to obtain a credibility evaluation result.

[0070] According to a third aspect of the embodiments of the present disclosure, a computer readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the news credibility evaluation method of the first aspect is implemented.

[0071] According to a sixth aspect of the embodiments of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the news credibility evaluation method of the first aspect described above via execution of the executable instructions.

[0072] According to the news credibility evaluation method, the news credibility evaluation device, the computer readable storage medium and the electronic device of the embodiments of the present disclosure, on the one hand, the scheme can accurately obtain the traceability information of the target object described in the news by matching the news to be evaluated with the pre-constructed object knowledge graph, and provides strong support for news authenticity evaluation. Further, the multi-modal features corresponding to the news, including text, image, video, etc., can comprehensively capture news information, enrich information dimensions, avoid the limitations of relying only on text analysis in related technologies, and provide more intuitive and richer evidence for news authenticity evaluation, which facilitates verification of the authenticity of news content from multiple angles and improves the credibility of the evaluation result. On the other hand, by combining the traceability information, the multi-modal features and the text expression mode of the news to evaluate the credibility of the news, the key news and potential correlations in the news can be identified on the basis of deep semantic understanding, ensuring the objectivity and fairness of the evaluation result. Overall, the method has significant advantages in improving the comprehensiveness, accuracy and intelligent level of news credibility evaluation, provides a powerful tool for combating fake news, and has good practical application value and wide popularization prospect. BRIEF DESCRIPTION OF DRAWINGS

[0073] The above and other objects, features and advantages of the exemplary embodiments of the present disclosure will be more apparent from the following detailed description taken in conjunction with the accompanying drawings, in which:

[0074] Figure 1 A flowchart of a news credibility evaluation method in an embodiment of the present disclosure is shown;

[0075] Figure 2 A flowchart of how to construct an object knowledge graph in an embodiment of the present disclosure is shown;

[0076] Figure 3 A schematic diagram of an object knowledge framework obtained by modeling in an embodiment of the present disclosure is shown;

[0077] Figure 4 A flowchart of how to perform knowledge extraction on the object knowledge framework to obtain multi-source knowledge elements in an embodiment of the present disclosure is shown;

[0078] Figure 5A flowchart showing how knowledge fusion is performed on multi-source knowledge elements in the embodiments of the present disclosure to obtain an object knowledge graph is shown.

[0079] Figure 6 A flowchart showing how entity alignment is performed on candidate entities with potential corresponding relationships in the deconfliction knowledge elements in the embodiments of the present disclosure to obtain an entity alignment result is shown.

[0080] Figure 7 A flowchart showing how multi-modal content features corresponding to the news to be evaluated are extracted in the embodiments of the present disclosure is shown.

[0081] Figure 8 A flowchart showing how the credibility evaluation result is determined in the embodiments of the present disclosure is shown.

[0082] Figure 9 A flowchart showing how the news to be evaluated is evaluated in terms of credibility according to the traceability information, the multi-modal content features, the multi-source multi-modal report content, and the text expression mode of the news to be evaluated to obtain the credibility evaluation result in the embodiments of the present disclosure is shown.

[0083] Figure 10 A flowchart showing how the content self-consistency of the news to be evaluated is verified according to the text expression mode of the news to be evaluated to obtain the fourth verification result in the embodiments of the present disclosure is shown.

[0084] Figure 11 A flowchart showing how the credibility evaluation result is determined in another embodiment of the present disclosure is shown.

[0085] Figure 12 A flowchart showing how a transparent, credible, and tamper-proof news dissemination system is constructed in the embodiments of the present disclosure is shown.

[0086] Figure 13 A schematic diagram of a news credibility evaluation device according to an embodiment of the present disclosure is shown.

[0087] Figure 14 A structural diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0088] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts. DETAILED DESCRIPTION

[0089] The principles and spirits of the present disclosure will be described below with reference to a number of exemplary embodiments. It should be understood that the embodiments are given only so that those skilled in the art can better understand and implement the present disclosure, and in no way limit the scope of the present disclosure. On the contrary, these embodiments are provided so that the present disclosure is more thorough and complete, and the scope of the present disclosure is fully conveyed to those skilled in the art.

[0090] Those skilled in the art know that the embodiments of the present disclosure can be implemented as a system, device, apparatus, method or computer program product. Therefore, the present disclosure can be embodied in the form of a complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0091] According to embodiments of the present disclosure, a news credibility evaluation method, a news credibility evaluation device, a computer readable storage medium and an electronic device are provided.

[0092] In this document, the number of any elements in the drawings is used for example and not limitation, and any naming is only used for distinction and does not have any limiting meaning.

[0093] The principles and spirits of the present disclosure will be described below with reference to a number of exemplary embodiments. It should be understood that the embodiments are given only so that those skilled in the art can better understand and implement the present disclosure, and in no way limit the scope of the present disclosure. On the contrary, these embodiments are provided so that the present disclosure is more thorough and complete, and the scope of the present disclosure is fully conveyed to those skilled in the art. SUMMARY

[0095] The present inventors found that the existing news credibility evaluation method mainly relies on manual review, which is inefficient and not accurate enough.

[0096] In view of the above, the basic idea of the present disclosure is to provide a news credibility evaluation method, a news credibility evaluation device, a computer readable storage medium and an electronic device. On the one hand, by matching the news to be evaluated with the pre-constructed object knowledge graph, the traceability information of the target object described in the news can be accurately obtained, providing strong support for news authenticity evaluation. Further, the multi-modal features corresponding to the news, including text, image, video, etc., can be extracted to comprehensively capture news information, enrich information dimensions, avoid the limitations of relying only on text analysis in related technologies, and provide more intuitive and richer evidence for news authenticity evaluation, which facilitates verification of the authenticity of news content from multiple angles and improves the credibility of the evaluation results. On the other hand, by combining the traceability information, multi-modal features and text expression mode of the news, the credibility of the news can be evaluated on the basis of deep semantic understanding, the key news and potential associations in the news can be identified, and the objectivity and fairness of the evaluation results can be ensured. Overall, this method has significant advantages in improving the comprehensiveness, accuracy and intelligent level of news credibility evaluation, provides a powerful tool for combating fake news, and has good practical application value and wide application prospects.

[0097] After introducing the basic principles of the present disclosure, various non-limiting embodiments of the present disclosure will be specifically introduced below.

[0098] Overview of Application Scenarios

[0099] It should be noted that the following application scenarios are only shown for the purpose of facilitating understanding of the spirit and principles of the present disclosure, and the embodiments of the present disclosure are not limited in this respect. On the contrary, the embodiments of the present disclosure can be applied to any applicable scenario.

[0100] The embodiments of the present disclosure support multi-dimensional credibility evaluation of news to be evaluated. Specifically, the multi-modal content features corresponding to the news to be evaluated can be extracted, and then the news to be evaluated is matched with the pre-constructed object knowledge graph to obtain the traceability information of the target object associated with the news to be evaluated. Thereafter, the news to be evaluated can be evaluated in terms of credibility according to the multi-modal content features, the traceability information and the text expression mode of the news to be evaluated, and a credibility evaluation result is obtained to realize more accurate, objective and interpretable credibility evaluation of news content and improve the accuracy and intelligent level of news content credibility evaluation.

[0101] Exemplary Method

[0102] The exemplary embodiments of the present disclosure first provide a news credibility evaluation method.

[0103] It should be noted that the present disclosure can adopt a client-server architecture, the client can be deployed on the user's work device (for example: mobile phone, tablet computer and personal desktop computer) to facilitate the user to upload news links or news content. The server end can deploy multiple source information collection module (responsible for collecting multi-source multi-modal reporting content from multimedia platform), object knowledge graph (used to provide traceability information of target object), multi-modal content analysis engine (used to compare multi-modal content features with credible data sources), traceability verification module (used to compare the news to be evaluated with traceability information) and credibility evaluation system (used to integrate credibility evaluation results of multiple dimensions to determine the final credibility evaluation result) and other core components responsible for specific news analysis and credibility evaluation tasks.

[0104] Figure 1 The flowchart of the news credibility evaluation method in the embodiment of the present disclosure is shown, which is executed by the server end described above, and includes the following steps S110 to S130:

[0105] Before step S110, it should be noted that the present disclosure can construct an object knowledge graph in advance, and the object knowledge graph is used to store the traceability information of each object in a structured manner. Specifically, the "object knowledge graph" in the present disclosure is a knowledge representation system that takes objects (such as persons, events, organizations, places, policies, and technological achievements) as core entities, describes their attributes, evolution processes, and associated relationships, and their sources of evidence in a structured manner. This graph converts unstructured or semi-structured information scattered in texts, archives, and databases into graph structure data with clear semantics, queryability, and reasoning capability, thereby supporting in-depth understanding and traceability analysis of objects.

[0106] Specifically, referring to Figure 2 , Figure 2 The flowchart of how to construct the object knowledge graph in the embodiment of the present disclosure is shown, which includes steps S201 to S204:

[0107] In step S201, the object context of each object is obtained.

[0108] In this step, the object context of each object can be obtained, which can include at least one of the following: object history record, object associated person relationship, and object background material.

[0109] The object history record is used to describe the evolution process of the object in the time dimension, and is the core basis for constructing the life cycle track. It includes but is not limited to the following contents:

[0110] Origin and development stage: the time and place of the first appearance of the object and its initial state;

[0111] Key event nodes: significant events experienced by the object during its lifetime (e.g., establishment, merger, restructuring, dissolution, etc.);

[0112] State change records: changes in object attributes over time (e.g., name changes, affiliation adjustments, functional positioning changes, etc.);

[0113] Time series modeling support: organizing object history records through time series graph structure to facilitate subsequent trend analysis, event reasoning, and evolution path prediction.

[0114] The above data can come from various data sources, such as Wikidata, electronic versions of historical yearbooks, academic papers, news archives, government bulletins, etc., which can be set according to actual conditions, and the present disclosure does not make special limitations on this.

[0115] Object-related person relationship refers to defining and storing the relationships between different entities (objects) in a knowledge graph, especially the various relationships between people and other entities (such as other people, events, places, organizations, etc.). This relationship describes how these entities interact or connect with each other and provides a foundation for understanding complex information networks. It includes but is not limited to the following: affiliation, cooperation, opposition, causality, influence, etc., which can be set according to actual conditions, and the present disclosure does not make special limitations on this. The above relationship information can come from unstructured or multi-modal data such as Wikipedia text, historical literature PDF (Portable Document Format), scanned copies, documentary subtitles, news reports, etc., which can be set according to actual conditions, and the present disclosure does not make special limitations on this.

[0116] Object background information covers basic information related to the object, knowledge in the field to which it belongs, policy and regulatory basis, cultural or technical background, etc., which is key support information for understanding the macro environment and deep logic of the existence of the object. It includes but is not limited to the following:

[0117] Basic attribute information: such as the category of the object, identity, industry to which it belongs, functional positioning, etc.;

[0118] Field knowledge mapping: through international patent classification (IPC) and subject classification code, etc., the object is attributed to a specific technical or subject field;

[0119] Policy and regulatory basis: relevant legal provisions, administrative orders, policy documents, etc. involved by the object;

[0120] Cultural and technical background: the era background, level of technological development, characteristics of social system, etc. in which the object is located;

[0121] Geospatial information: With the help of tools such as GeoNames API (Application Programming Interface), ancient place names, historical administrative divisions are mapped into modern geographical coordinates, enhancing the spatiotemporal perception ability;

[0122] The above background information can come from DBpedia, YAGO (Yet Another Great Ontology), historical map databases, government public data, and other structured or semi-structured data sources, which can be set by the actual situation, and the present disclosure does not make special limitations on this.

[0123] By integrating the above multi-dimensional object context information, the semantic features and background associations of the object can be fully described, providing a structured and traceable data foundation for subsequent content enhancement, authenticity verification, knowledge reasoning, and intelligent recommendation tasks.

[0124] In step S202, the object context is knowledge modeled to obtain an object knowledge framework.

[0125] In this step, the object context can be knowledge modeled to obtain an object knowledge framework. Among them, the knowledge modeling aims to convert the unstructured or semi-structured object context information into a standardized knowledge framework with unified structure, semantic expression ability, and support for reasoning and query. Through this process, the system can more efficiently understand the internal relationship between objects and provide technical support for its application in news enhancement, event tracing, intelligent recommendation and other tasks.

[0126] Optionally, the object context can also be data cleaned to obtain a standard object context, and then knowledge modeling is performed based on the standard object context to obtain an object knowledge framework. Among them, the data cleaning includes at least one of the following: time standardization processing (for example: converting the lunar calendar to the solar calendar), place name remapping processing (for example: using GeoNames API to map ancient place names to modern coordinates), and ambiguity elimination processing (for example: a disambiguation model can be established using BERT (Bidirectional Encoder Representations from Transformers) + CRF (Conditional Random Fields)).

[0127] Exemplarily, the specific process of the above knowledge modeling can be: first, defining the ontology model of the knowledge framework, clearly defining the object category, attribute and relationship system; then extracting and uniformly identifying objects from multi-source data through entity recognition and linking technology; subsequently generating "subject-relation-object" triples by using relationship extraction method, constructing a semantic relationship network; at the same time, modeling the time and space information, restoring the historical trajectory and geographical association of the object; then integrating multi-source information and solving conflicts through knowledge fusion and consistency checking, ensuring the accuracy of the knowledge; finally, generating a structured object knowledge framework, and storing and displaying it in the form of a graph database or visualization to support subsequent intelligent applications.

[0128] Exemplarily, reference can be made to Figure 3 , Figure 3 A schematic diagram of an object knowledge framework modeled in the embodiments of the present disclosure is shown, specifically, the diagram shows an entity relationship model of an event (EVENT), where each event is defined by its unique identifier (event_id), name (name), description (description) and type (type). There is an association relationship between the event and the participant (PARTICIPANT), and the participant is described by its unique identifier (entity_id), name (name) and type (type), and each participant plays a specific role (ROLE) in the event. In addition, the location (LOCATION) of the event is associated with its geographic coordinates (COORDINATE), providing spatial information of the event. The event is also associated with a time span (TIMESPAN), which records the time range of the event. Finally, the source (DOCUMENT) of the event provides the document or material source about the event. The entire model comprehensively describes the multi-dimensional information of the event, including participants, locations, times and related documents, through these entities and the relationships between them.

[0129] In step S203, knowledge extraction is performed on the object knowledge framework to obtain multi-source knowledge elements.

[0130] In this step, knowledge extraction can be performed on the above object knowledge framework to obtain multi-source knowledge elements. The multi-source knowledge elements can include text knowledge elements, visual knowledge elements and audio-visual knowledge elements.

[0131] Exemplarily, reference can be made to Figure 4 , Figure 4 A flowchart showing how to perform knowledge extraction on the object knowledge framework to obtain multi-source knowledge elements in the embodiments of the present disclosure is shown, including steps S401-S403:

[0132] In step S401, knowledge extraction is performed on the text information in the object knowledge framework to obtain text knowledge elements.

[0133] In this step, knowledge extraction can be performed on the textual information in the object knowledge framework to obtain textual knowledge elements. Specifically, SPARQL can be used to extract factual information from the existing object knowledge framework. Furthermore, the BERT model, which has been trained specifically for historical text, can be used to perform relationship extraction on unstructured historical text, thereby obtaining structured semantic knowledge that can be used to construct an event knowledge graph.

[0134] In step S402, knowledge extraction is performed on the image information in the object knowledge framework to obtain visual knowledge elements.

[0135] In this step, knowledge extraction can be performed on the image information in the object knowledge framework to obtain visual knowledge elements. Specifically, a variety of image processing and computer vision technologies can be used to perform multi-level analysis and knowledge mining on the image content. For example:

[0136] Optical Character Recognition (OCR) extracts ancient text content: For image materials containing historical text (such as ancient memorials, rubbings of inscriptions, and handwritten manuscripts), the Tesseract OCR engine is integrated and loaded with a model trained on historical fonts to achieve high-precision recognition of special characters such as traditional Chinese characters, variant characters, seal script, and official script, converting the text content in the image into structured text information;

[0137] CNN (Convolutional Neural Network) identifies routes and regional boundaries in maps: For historical map images (such as ancient territorial maps, transportation route maps, and war situation maps), convolutional neural network models analyze visual features such as lines, colors, and shapes in the map to automatically identify spatial information such as major routes, administrative boundaries, and spheres of influence, and further convert them into spatial coordinates or geographic markers;

[0138] Image metadata extraction and entity association: In addition to the image content itself, image metadata information (such as shooting time, source organization, image description, etc.) can also be extracted. Through natural language processing and entity linking technology, semantic associations are established between the image content and related entities in the object knowledge framework (such as people, events, and places);

[0139] Text content, geographic boundaries, route information, image descriptions, etc. extracted from images can serve as the above-mentioned visual knowledge elements.

[0140] Through the above-mentioned image knowledge extraction process, not only the information integrity and cross-modal expression capabilities of the object knowledge framework are improved, but also richer data support is provided for subsequent content enhancement, credibility verification, intelligent recommendation and other functions.

[0141] In step S403, knowledge extraction is performed on the video information in the object knowledge framework to obtain audio-visual knowledge elements.

[0142] In this step, knowledge extraction can be performed on the video information in the object knowledge framework to obtain audio-visual knowledge elements. Specifically, speech recognition can be performed on the video information to extract commentary, and key frame extraction and scene analysis can be performed on the video information to obtain the above audio-visual knowledge elements.

[0143] Next reference Figure 2 ,In step S204, knowledge fusion is performed on multi-source knowledge elements to obtain an object knowledge graph.

[0144] In this step, the multi-source knowledge elements can be fused to obtain the object knowledge graph. Figure 5 , Figure 5 The flowchart of how to fuse multi-source knowledge elements to obtain an object knowledge graph in the embodiment of the present disclosure includes steps S501 to S503:

[0145] In step S501, conflict resolution is performed on multi-source knowledge elements to obtain conflict-free knowledge elements.

[0146] In this step, conflict resolution can be performed on the multi-source knowledge elements to obtain conflict-free knowledge elements. Specifically, the conflict types existing in the multi-source knowledge elements can be identified, and conflict resolution can be performed on the multi-source knowledge elements according to the conflict resolution strategy corresponding to the conflict type to obtain conflict-free knowledge elements. By identifying and strategically resolving conflict types, the inconsistency problem in multi-source knowledge fusion is effectively resolved, significantly improving the quality and efficiency of knowledge integration, and providing key technical support for the construction of an object knowledge graph with clear structure, accurate semantics, and strong scalability.

[0147] Optionally, the conflict type described above can include time conflict, when the conflict type is time conflict, the time decision strategy based on information source credibility can be used to resolve the conflict of the multi-source knowledge elements. Specifically, first, the time description differences about the same event or object from different data sources can be identified, such as inconsistent event time, duration, time node sequence, etc. Then, according to the historical accuracy, authority, publishing time and other factors of each information source, the credibility weight is evaluated, and the time logic consistency rules (such as rationality of sequence, rationality of time span) are combined to determine which time expression is more consistent with the objective fact. For the time information provided by the high credibility source, the system gives higher priority and adopts it in the knowledge fusion process. At the same time, low credibility or contradictory information is marked and recorded for subsequent manual review or dynamic updating mechanism, so as to realize the automatic identification and intelligent resolution of time conflict.

[0148] Optionally, the conflict type described above can include location conflict, when the conflict type is location conflict, the location decision strategy based on multi-modal evidence fusion can be used to resolve the conflict of the multi-source knowledge elements. Specifically, first, the geographical location description differences about the same entity or event in different sources can be detected, such as inconsistent place name expression, coordinate position deviation, different administrative division attribution, etc. Then, the unified space semantic representation model is constructed by integrating multi-modal information such as text description, geographic coordinates, historical map image, spatial semantic relationship, etc. On this basis, the location data provided by each source can be semantically aligned and coordinate matched by using geographic information system (GIS) technology, place name standardization tool and spatial boundary information extracted by image recognition, and the credibility evaluation results of the information sources are combined to weight the evidence support degree from different modalities. For the conflicting location information, the system optimizes the most possible accurate location information as the fusion result by calculating the comprehensive matching score of each candidate location in the time dimension, context, spatial rationality, etc.

[0149] Optionally, the conflict type described above can include participant conflict, when the conflict type is participant conflict, the subject decision strategy based on entity credibility evaluation can be used to resolve the conflict of the multi-source knowledge elements. Specifically, first, the inconsistencies between the participant information about the same event or object in different sources can be identified, such as differences in person name, role identity, affiliation organization, participation degree, etc. Then, based on the entity knowledge base in the knowledge graph, combined with external authoritative databases (such as historical literature, biographies, organization archives, etc.), the participant entities involved are evaluated in multiple dimensions, including but not limited to:

[0150] Entity Source Authority: Determine whether the participant information comes from a high-trust data source (such as official archives, academic research results) or a low-trust data source (such as network user-generated content, non-professional publications);

[0151] Entity Association Strength: Analyze the semantic association between the participant and related events, places or other entities, such as frequent co-occurrence or explicit role description;

[0152] Entity Historical Consistency: Evaluate the rationality of the participant in the timeline and spatial distribution, such as whether their birth and death years, and activity areas are consistent with the time and place of the event;

[0153] Multi-modal Evidence Support: Combine image, text, structured data and other multi-modal information to verify whether there is evidence supporting the participant's involvement in the relevant event (such as photos, signatures, literature records, etc.).

[0154] On this basis, a credibility score can be assigned to each participant entity, and the conflict information can be prioritized according to the score. For participant information with high credibility, it is retained and used as the dominant information in the knowledge fusion process; while for information with low credibility or contradictions, it is marked, isolated and stored, and supports subsequent manual review or dynamic update mechanism.

[0155] In step S502, the candidate entities with potential corresponding relationships in the de-conflicted knowledge elements are processed for entity alignment to obtain an entity alignment result.

[0156] In this step, the candidate entities with potential corresponding relationships in the de-conflicted knowledge elements can be processed for entity alignment to obtain an entity alignment result. The potential corresponding relationship refers to that although two entities are not completely the same in surface form (such as name, expression method, language, code, etc.) in different data sources or knowledge expressions, they may point to the same object or concept in the real world at the semantic level. This semantic consistency has not been explicitly confirmed and needs to be discovered through semantic analysis and reasoning.

[0157] Specifically, features can be extracted for each candidate entity, and then the entity similarity between any two candidate entities can be calculated according to the extracted features. The candidate entities can be clustered according to the entity similarity to obtain the entity alignment result. Optionally, after clustering to obtain the entity alignment result, manual verification can be performed to determine the final entity alignment result.

[0158] Reference Figure 6 , Figure 6A flowchart of how to perform entity alignment processing on candidate entities with potential corresponding relationships in the conflict-removed knowledge elements in the embodiments of the present disclosure is shown, and the entity alignment result is obtained, which includes steps S601-S605:

[0159] In step S601, candidate entities are discovered.

[0160] In step S602, feature extraction is performed on the candidate entities.

[0161] In step S603, similarity calculation is performed on the candidate entities according to the extracted features.

[0162] In step S604, entity clustering analysis is performed according to the similarity.

[0163] In step S605, manual verification is performed to obtain the final entity alignment result.

[0164] Next, referring to Figure 5 In step S503, knowledge normalization modeling is performed based on the entity alignment result to obtain an object knowledge graph.

[0165] In this step, knowledge normalization modeling can be performed based on the above-mentioned entity alignment result to obtain an object knowledge graph. Specifically, first, a unique identifier can be assigned to the aligned entities, and their names and aliases can be unified; second, attribute information from different sources can be standardized in format and weighted and fused in credibility to ensure the consistency and accuracy of the attribute values; meanwhile, diversified semantic relationships can be mapped to a unified relationship system to realize the standardized expression of the relationships; finally, the entity semantics can be enhanced in combination with multi-modal information, and the graph structure can be optimized to improve the overall consistency and computability, so that a semantic clear, structure reasonable and extensible object knowledge graph is obtained.

[0166] By constructing the object knowledge graph, the structured, semantic and visual expression of knowledge can be realized, the semantic association between multi-source heterogeneous information can be broken through, and solid support can be provided for the subsequent news credibility evaluation process.

[0167] Optionally, in addition to relying on the data collection and arrangement of the system itself to construct the object knowledge graph, a more authoritative and comprehensive knowledge graph data can be introduced by cooperating with a professional knowledge graph construction institution, or a crowdsourcing mode can be adopted to invite field experts and scholars to participate in the construction and improvement of the knowledge graph. The present disclosure does not make special limitations on this.

[0168] Next, referring to Figure 1 In step S110, the to-be-evaluated news is matched with the pre-constructed object knowledge graph to obtain the traceability information of the target object described in the to-be-evaluated news.

[0169] In this step, the user can upload the news link or news content of the news to be evaluated to the server through the client, so that the server can obtain the news to be evaluated.

[0170] Afterwards, the server can match and associate the news to be evaluated with the pre-built object knowledge graph to obtain the traceability information of the target object described in the news to be evaluated. For example, the traceability information can include the target object's past time records, character relationships, and object background information.

[0171] In step S120 , multimodal content features corresponding to the news to be evaluated are extracted.

[0172] In this step, the multimodal content features corresponding to the news to be evaluated can be extracted. Specifically, the news to be evaluated can contain at least two of the following: text content, image content, video content and audio content. Figure 7 , Figure 7 A flow chart showing how to extract multimodal content features corresponding to news to be evaluated in an embodiment of the present disclosure includes steps S701 to S705:

[0173] In step S701, semantic features are extracted from the text content to obtain text features.

[0174] In this step, semantic features can be extracted from the text content to obtain text features. Specifically, AIGC technology can be used to perform semantic analysis and sentiment analysis on the text content to extract text features such as keywords, entities, and semantic vectors.

[0175] In step S702, visual feature analysis is performed on the image content to obtain image features.

[0176] In this step, the image content can be analyzed for visual features to obtain image features. Specifically, computer vision technology can be used to perform object recognition, scene analysis, image quality assessment, etc. on image and video content, extracting image features such as color histograms, texture features, and object recognition results.

[0177] In step S703, spatiotemporal feature modeling analysis is performed on the video content to obtain video features.

[0178] In this step, the video content may be subjected to spatiotemporal feature modeling and analysis to obtain video features. Specifically, the video content may be subjected to spatiotemporal feature modeling and analysis to extract video features such as frame sequence features and motion trajectory features.

[0179] In step S704, acoustic feature analysis is performed on the audio content to obtain audio features.

[0180] In this step, the audio content can be subjected to acoustic feature analysis to obtain audio features. Specifically, the audio content can be subjected to speech recognition, voiceprint analysis, etc. through audio analysis techniques to extract audio features.

[0181] In step S705, the text features, image features, video features, and audio features are subjected to multi-modal feature fusion to obtain multi-modal content features.

[0182] In this step, the text features, image features, video features, and audio features can be subjected to multi-modal feature fusion to obtain multi-modal content features, thereby comprehensively characterizing various forms of expression of the target object and forming a comprehensive understanding of the target object.

[0183] Reference is then made to Figure 1 In step S130, the credibility of the news to be evaluated is evaluated according to the traceability information, the multi-modal content features, and the text expression mode of the news to be evaluated, to obtain a credibility evaluation result.

[0184] In this step, in an optional embodiment, the credibility of the news to be evaluated can be evaluated according to the traceability information, the multi-modal content features, and the text expression mode of the news to be evaluated, to obtain a credibility evaluation result. Specifically, first, the traceability information and the news to be evaluated can be subjected to content consistency verification to obtain a first verification result, second, the multi-modal content features and the credible data source can be subjected to content consistency verification to obtain a second verification result, third, the news to be evaluated can be subjected to content self-consistency verification according to the text expression mode of the news to be evaluated to obtain a fourth verification result, and then the credibility evaluation result can be determined according to the first verification result, the second verification result, and the fourth verification result.

[0185] In another optional embodiment, reference can be made to Figure 8 , Figure 8 A flowchart showing how the credibility evaluation result is determined in the embodiments of the present disclosure is shown, comprising steps S801-S802:

[0186] In step S801, multi-source multi-modal reporting content for a target object is obtained.

[0187] In this step, multi-source multi-modal reporting content for a target object can be obtained. Multi-source multi-modal reporting content refers to a collection of information obtained from different sources and in various forms of expression. This concept emphasizes the diversity and richness of information, which can be understood from two dimensions:

[0188] Multi-source refers to information from multiple different channels or platforms. These sources can include but are not limited to:

[0189] Traditional media: such as newspapers, magazines, TV stations, etc.

[0190] New media: such as websites, blogs, social media platforms (Weibo, WeChat, Twitter, Facebook, etc.);

[0191] Professional databases: academic paper databases, government-issued statistical data, corporate annual reports, etc.

[0192] Internet of Things devices: real-time data provided by sensor networks, surveillance cameras, smart wearable devices, etc.

[0193] Each source may provide a unique perspective or specialized data. Integrating information from different sources can provide a more comprehensive understanding of an event or topic.

[0194] Multimodal refers to information existing or being expressed in multiple forms, including but not limited to the following modalities:

[0195] Text: written descriptions such as articles, reviews, reports, etc.

[0196] Images: visual materials such as photos, illustrations, charts, etc.

[0197] Video: dynamic images such as news clips, documentaries, user-generated content, etc.

[0198] Audio: sound recordings such as radio programs, podcasts, interview recordings, etc.

[0199] Geospatial data: geographic location-related information such as maps, GPS (Global Positioning System) tracks, etc.

[0200] Time series data: data that changes over time, such as stock price changes, weather forecasts, etc.

[0201] Each modality has its unique advantages, for example, images can directly show the scene, while text is more suitable for conveying detailed analysis and explanation. Combining information from multiple modalities can provide users with a richer and more in-depth understanding experience.

[0202] Optionally, in the aspect of multi-source multi-modal reporting content collection, in addition to automatically scraping media platform reporting content, users can be encouraged to manually upload and share relevant news content, and incentives and rewards can be provided to users for providing information. To enrich information sources, you can set it yourself according to actual conditions, and this disclosure does not make special limitations.

[0203] In step S802, according to the traceability information, the multi-modal content features, and the multi-source multi-modal reporting content, the text expression mode of the news to be evaluated is combined to evaluate the credibility of the news to be evaluated, and the credibility evaluation result is obtained.

[0204] In this step, the credibility of the news to be evaluated can be evaluated according to the traceability information, the multi-modal content features, the multi-source multi-modal reporting content, and the text expression mode of the news to be evaluated, and a credibility evaluation result is obtained. Specifically, reference is made to Figure 9 , Figure 9 A flowchart for illustrating how the credibility of the news to be evaluated is evaluated according to the traceability information, the multi-modal content features, the multi-source multi-modal reporting content, and the text expression mode of the news to be evaluated, and a credibility evaluation result is obtained in the embodiment of the present disclosure is shown, including steps S901-S905:

[0205] In step S901, the traceability information and the news to be evaluated are verified for content consistency, and a first verification result is obtained.

[0206] In this step, the traceability information and the news to be evaluated can be verified for content consistency, and a first verification result is obtained. Specifically, the news to be evaluated and its traceability information can be compared in multiple dimensions in terms of entities, time, location, and event description, and whether they are consistent can be determined through semantic analysis and structured information extraction, so as to identify whether there is a content deviation or tampering, and finally generate a verification result reflecting the consistency strength, providing a preliminary basis for news authenticity evaluation.

[0207] In step S902, the multi-modal content features and the credible data source are verified for content consistency, and a second verification result is obtained.

[0208] In this step, the multi-modal content features (such as image scenes, video frame information, speech-to-text, text semantics, etc.) extracted from the news can be compared with the original data (such as real pictures officially released, surveillance videos, government announcements, mainstream media reports, authoritative databases, and credible news archives, etc.) from authoritative sources in cross-modal content comparison, verifying their consistency in terms of time, location, characters, event description, etc., identifying whether there are problems such as image tampering, video splicing, and selective quotation, and finally generating a second verification result as a further judgment basis for the authenticity, integrity, and credibility of the news content.

[0209] In step S903, the multi-source multi-modal reporting content and the news to be evaluated are verified for content consistency, and a third verification result is obtained.

[0210] In this step, the to-be-evaluated news can be compared and analyzed in terms of semantics and facts with multi-source and multi-modal reporting content from different media platforms, different information channels (such as mainstream media, social media, government releases, etc.), and various expression forms (such as text, pictures, videos, etc.), to analyze the consistency degree of the core elements such as time, place, person, and event process, and to identify whether there are problems such as exaggeration, distortion, or isolated quotation, so as to generate a third verification result for further enhancing the objectivity and comprehensiveness of the news authenticity judgment.

[0211] In step S904, content self-consistency verification is performed on the to-be-evaluated news according to the text expression mode of the to-be-evaluated news, and a fourth verification result is obtained.

[0212] In this step, content self-consistency verification can be performed on the to-be-evaluated news according to the text expression mode of the to-be-evaluated news, and a fourth verification result is obtained. Specifically, reference is made to Figure 10 , Figure 10 A flowchart showing how content self-consistency verification is performed on the to-be-evaluated news according to the text expression mode of the to-be-evaluated news in the embodiment of the present disclosure is shown, including steps S1001-S1003:

[0213] In step S1001, the text logic of the to-be-evaluated news is analyzed, and a logic analysis result is obtained.

[0214] In this step, the text logic of the to-be-evaluated news can be analyzed in depth by means of a large prediction model, so as to judge whether the narration is reasonable, the arrangement is clear, and there is misleading writing or logical loopholes. Specifically, this process can include the following aspects:

[0215] Overall architecture analysis: Check whether the news article conforms to the standard news writing specifications, such as whether it adopts the "inverted pyramid structure" - that is, the most important information (such as the subject of the event, time, place, and result) is in the front, and the details and background information are supplemented in the back, to ensure that the reader can quickly obtain the core content;

[0216] Paragraph logical relationship identification: Analyze the semantic connection and logical sequence between paragraphs to determine whether they progress layer by layer around the central theme, whether there are problems such as jump narration, contradiction between the front and back, or irrelevant content insertion, etc.;

[0217] Sentence logical coherence evaluation: Use a language model to identify the logical relationships between sentences such as causality, transition, parallelism, progression, etc., to evaluate whether the sentences are natural and fluent, and whether there are broken or forced splicing situations;

[0218] Reasoning chain completeness judgment: If the news involves reasoning or conclusions, the model can identify whether it has clear premises, and whether there are logical fallacies such as "generalizing from a small sample" and "unjustified conclusion".

[0219] Thus, the logical analysis result in the present disclosure can include but is not limited to: article logical structure score, logical breakpoint or abnormal paragraph mark, reasoning rationality evaluation, and judgment suggestion on whether it meets the news writing specification.

[0220] In step S1002, the event causal relationship related to the news to be evaluated is verified, and a causal relationship verification result is obtained.

[0221] In this step, the causal logic between the events described in the news can be analyzed in depth by means of a large language model, combined with existing authoritative data sources (such as policy documents, statistical data, official reports, etc.) and common sense logical reasoning, to determine whether the "cause-effect" relationship in the report is real, reasonable and complete. Specifically, it includes: verifying whether the event cause is consistent with the known facts, whether the causal chain exists or is forced, and whether the conclusion has sufficient premise support, so as to identify potential misleading narratives or logical flaws, and output structured causal relationship verification results to provide key basis for news authenticity evaluation.

[0222] In step S1003, the logical analysis result and the causal relationship verification result are integrated to obtain a fourth verification result.

[0223] In this step, the logical analysis result and the causal relationship verification result can be integrated to obtain the fourth verification result. Specifically, the logical analysis result of the news text (such as whether the article structure is reasonable and the paragraph connection is coherent) and the causal relationship verification result (such as whether the causal chain between events is real and whether there is a logical fallacy) can be analyzed comprehensively, combined with weight distribution and consistency judgment mechanism, to identify the logical contradictions or information biases, and finally generate a more comprehensive and reliable fourth verification result, which is used to further improve the depth and accuracy of news authenticity evaluation.

[0224] Next, reference is made to Figure 9 In step S905, the credibility evaluation result is determined according to the first verification result, the second verification result, the third verification result, and the fourth verification result.

[0225] In this step, in an optional embodiment, the first verification result, the second verification result, the third verification result, and the fourth verification result can be integrated to determine the credibility evaluation result.

[0226] In another optional embodiment, reference can be made to Figure 11 , Figure 11 Another flowchart for determining the credibility evaluation result is shown, which includes steps S1101-S1104:

[0227] In step S1101, the standard record information for the target object in the cross-language trusted data source is obtained.

[0228] In this step, the standard record information for the target object in the cross-language trusted data source can be obtained. The cross-language trusted data source refers to a source that can provide accurate, objective and authoritative information in multiple different language environments. These data sources typically include but are not limited to news media, government announcements, academic publications, professional databases, multilingual versions of internationally renowned news media, authoritative translations by professional translation agencies, multilingual academic literature and research reports, etc. They are distributed in different countries and regions, use their own languages to publish content, but all have high credibility and reliability.

[0229] In step S1102, the news to be evaluated is translated into the target language corresponding to the standard record information.

[0230] In this step, in order to achieve cross-language content consistency verification, the news to be evaluated can be translated into the target language used by the standard record information, such as English, French, Spanish, etc. The translation process can use high-quality machine translation models (such as Transformer-based models, multilingual large models, etc.) to ensure grammatical correctness, accurate terminology, and support for subsequent content comparison and logical analysis in a unified language dimension while preserving the original semantics.

[0231] In step S1103, the translated news to be evaluated and the standard record information are compared for content consistency, and a fifth verification result is obtained.

[0232] In this step, after the news to be evaluated is translated into the target language, it is compared with the standard record information (i.e. authoritative and reliable news reports or official releases) from the cross-language trusted data source for multi-dimensional content consistency. The comparison content includes but is not limited to key entities (persons, places, organizations), event timelines, cause-effect relationships, core fact descriptions, and other semantic information. Through natural language processing techniques such as text similarity calculation, event extraction, and semantic role labeling, it is determined whether there are factual biases, information omissions or inconsistent expressions between the two, and finally a fifth verification result is generated to judge the accuracy and reliability of the news in a cross-language context.

[0233] In step S1104, the first verification result, the second verification result, the third verification result, the fourth verification result and the fifth verification result are combined to determine the credibility evaluation result.

[0234] In this step, the five types of verification results described above can be weighted and fused and synergistically analyzed, combined with the structured output information of the credibility score of each dimension, the difference point identification, the logical abnormality marking, etc. The rule engine or machine learning model (such as an integrated classifier, a confidence fusion algorithm) is used to model the overall credibility, and finally a unified news credibility evaluation result is generated. The result can be expressed as a comprehensive score, a credibility level (such as high / medium / low), or a risk prompt label, which is used to assist users, platforms or regulatory departments in making scientific judgments on the authenticity, objectivity and dissemination value of news content.

[0235] After obtaining the credibility evaluation result for the above-mentioned news to be evaluated, the present disclosure can also build a transparent, credible and tamper-proof news dissemination system with the aid of a blockchain system. Specifically, reference can be made to Figure 12 , Figure 12 A flowchart showing how to build a transparent, credible and tamper-proof news dissemination system in the embodiment of the present disclosure is shown, including steps S1201-S1202:

[0236] In step S1201, a news label corresponding to the news to be evaluated is generated according to the credibility evaluation result and the key features of the news to be evaluated.

[0237] In this step, the news label (also referred to as a digital fingerprint, which can be set according to actual conditions, and the present disclosure does not make special limitations thereon) corresponding to the news to be evaluated can be generated according to the credibility evaluation result and the key features of the news to be evaluated. The above-mentioned news key features can include the creation timestamp of the news, the author identity information, the editing modification record, the link of the cited original material, etc., which can be set according to actual conditions, and the present disclosure does not make special limitations thereon.

[0238] In step S1202, the news to be evaluated and the news label are uploaded to the blockchain, so that each verification node in the blockchain performs credibility evaluation on the news to be evaluated again based on the news label and real-time dynamic data sources.

[0239] In this step, the news to be evaluated and news tags can be uploaded to the blockchain, so that each verification node in the blockchain can re-evaluate the credibility of the news to be evaluated based on the news tags and real-time dynamic data sources. Specifically, after completing the preliminary credibility evaluation, the news to be evaluated and its corresponding news tags (such as event type, keyword, source identification, content structure feature, credibility level, and other meta-information) can be uploaded to the blockchain system. Multiple verification nodes in the blockchain (such as mainstream media agencies, third-party audit organizations, government regulatory departments, and intelligent contract automatic verification modules) can combine real-time dynamic data sources (such as the latest official announcements, authoritative database updates, public opinion monitoring platform data, etc.) to perform secondary collaborative verification on the authenticity, timeliness, and consistency of the news content according to these tag information. Each verification result will be recorded on the blockchain to form an unalterable evaluation log, thereby realizing multi-party participation, process transparency, dynamic updating, and traceability of news credibility evaluation. This mechanism not only enhances the objectivity and robustness of news content credibility evaluation, but also provides technical support for subsequent content recommendation, false information identification, and media credit system construction.

[0240] Based on the above technical solutions, the present disclosure has at least the following technical effects:

[0241] First, the multi-source information collection module can collect reporting content from a variety of media platforms (such as mainstream news websites, social media, government announcement platforms, etc.), covering different perspectives and expression forms, effectively ensuring the comprehensiveness and diversity of news analysis. This avoids the evaluation bias that may be caused by relying on a single information source in traditional methods, providing rich and reliable data support for subsequent traceability analysis and authenticity evaluation, and significantly improving the accuracy and representativeness of the evaluation results.

[0242] Second, by constructing an object knowledge graph, current news events can be correlated with historical events, character relationships, background information, etc. for correlation analysis, achieving deep understanding and context tracing of news content. For example, when analyzing news involving political figures, the system can quickly retrieve the historical behavior trajectory, social network, and key events of the figure to assist in identifying whether the current news is consistent with known facts, thereby improving the depth and accuracy of news authenticity evaluation.

[0243] Third, the multi-modal content analysis engine integrates image, video, audio, and other multi-modal information, making up for the technical limitations of traditional text analysis only. By deeply analyzing the multimedia content in the news, such as detecting whether the pictures have been tampered with, whether the video has been spliced, and whether the audio has been forged, the system can verify the authenticity of the news from multiple dimensions, enhancing the intuitiveness and evidence support of the evaluation, and further improving the effectiveness of the credibility evaluation.

[0244] Fourth, the traceability verification module uses the reasoning ability of large language models to construct the transmission chain of news events and accurately identify the origin and diffusion path of information. This mechanism not only helps to find the source of fake news and tampering traces in the transmission process, but also provides a powerful tool for combating rumors and misleading information. At the same time, users can understand the transmission range and influence of news through the traceability results, enhance their information discrimination ability, and reduce the risk of being misled.

[0245] In summary, the present disclosure constructs a comprehensive, in-depth, traceable, and multi-angle verification news credibility evaluation system, significantly improving the scientificity, objectivity, and reliability of news authenticity judgment.

[0246] Exemplary Apparatus

[0247] After introducing the news credibility evaluation method of the exemplary embodiments of the present disclosure, next, with reference to Figure 13 The news credibility evaluation device of the exemplary embodiments of the present disclosure is described.

[0248] Figure 13 The structure schematic diagram of the news credibility evaluation device in the exemplary embodiments of the present disclosure is shown. As shown in Figure 13 The news credibility evaluation device 1300 can include an acquisition module 1310, a feature extraction module 1320, and a credibility evaluation module 1330. Among them:

[0249] The acquisition module 1310 is configured to match the news to be evaluated with a pre-constructed object knowledge graph to obtain traceability information of a target object described by the news to be evaluated. The object knowledge graph is used to store the traceability information of each object in a structured manner.

[0250] The feature extraction module 1320 is configured to extract multi-modal content features corresponding to the news to be evaluated.

[0251] The credibility evaluation module 1330 is configured to evaluate the credibility of the news to be evaluated according to the traceability information, the multi-modal content features, and the text expression mode of the news to be evaluated, and obtain the credibility evaluation result.

[0252] In an optional embodiment, the credibility evaluation module 1330 is configured to:

[0253] Obtain multi-source multi-modal reporting content for the target object;

[0254] According to the traceability information, the multi-modal content features, the multi-source multi-modal reporting content, and the text expression mode of the news to be evaluated, the credibility of the news to be evaluated is evaluated, and the credibility evaluation result is obtained.

[0255] In an optional implementation, the obtaining module 1310 constructs the object knowledge graph in the following manner:

[0256] obtaining an object context of each object, the object context including at least one of the following: an object history record, an object associated person relationship, and an object background material;

[0257] performing knowledge modeling on the object context to obtain an object knowledge framework;

[0258] performing knowledge extraction on the object knowledge framework to obtain multi-source knowledge elements;

[0259] performing knowledge fusion on the multi-source knowledge elements to obtain the object knowledge graph.

[0260] In an optional implementation, the obtaining module 1310 performs knowledge modeling on the object context to obtain an object knowledge framework, including:

[0261] performing data cleaning processing on the object context to obtain a standard object context;

[0262] performing knowledge modeling based on the standard object context to obtain the object knowledge framework;

[0263] wherein the data cleaning processing includes at least one of the following: time standardization processing, place name remapping processing, and ambiguity elimination processing.

[0264] In an optional implementation, the obtaining module 1310 performs knowledge extraction on the object knowledge framework to obtain multi-source knowledge elements, including:

[0265] performing knowledge extraction on text information in the object knowledge framework to obtain text knowledge elements;

[0266] performing knowledge extraction on image information in the object knowledge framework to obtain visual knowledge elements;

[0267] performing knowledge extraction on video information in the object knowledge framework to obtain audio-visual knowledge elements.

[0268] In an optional implementation, the obtaining module 1310 performs knowledge fusion on the multi-source knowledge elements to obtain the object knowledge graph, including:

[0269] performing conflict resolution processing on the multi-source knowledge elements to obtain conflict-free knowledge elements;

[0270] performing entity alignment processing on candidate entities having potential corresponding relationships in the conflict-free knowledge elements to obtain an entity alignment result;

[0271] model knowledge normalization based on the entity alignment result to obtain the object knowledge graph.

[0272] In an optional implementation, the obtaining module 1310 performs conflict resolution processing on the multi-source knowledge elements to obtain conflict-removed knowledge elements, including:

[0273] identifying a conflict type existing in the multi-source knowledge elements;

[0274] performing conflict resolution processing on the multi-source knowledge elements according to a conflict resolution strategy corresponding to the conflict type to obtain the conflict-removed knowledge elements.

[0275] In an optional implementation, when the conflict type includes a time conflict, the conflict resolution strategy corresponding to the conflict type includes a time decision strategy based on information source credibility;

[0276] when the conflict type includes a location conflict, the conflict resolution strategy corresponding to the conflict type includes a location decision strategy based on multi-modal evidence fusion;

[0277] when the conflict type includes a participant conflict, the conflict resolution strategy corresponding to the conflict type includes a subject decision strategy based on entity credibility evaluation.

[0278] In an optional implementation, the obtaining module 1310 performs entity alignment processing on candidate entities having potential corresponding relationships in the conflict-removed knowledge elements to obtain an entity alignment result, including:

[0279] calculating an entity similarity between any two of the candidate entities;

[0280] performing clustering processing on the candidate entities according to the entity similarity to obtain the entity alignment result.

[0281] In an optional implementation, the news to be evaluated includes at least two of the following: text content, image content, video content, and audio content;

[0282] The feature extraction module 1320 extracts multi-modal content features corresponding to the news to be evaluated, including:

[0283] performing semantic feature extraction on the text content to obtain text features;

[0284] performing visual feature analysis on the image content to obtain image features;

[0285] performing spatio-temporal feature modeling analysis on the video content to obtain video features;

[0286] performing acoustic feature analysis on the audio content to obtain an audio feature;

[0287] performing multi-modal feature fusion on the text feature, the image feature, the video feature, and the audio feature to obtain the multi-modal content feature.

[0288] In an optional implementation, the credibility evaluation module 1330 performs credibility evaluation on the news to be evaluated according to the provenance information, the multi-modal content feature, the multi-source multi-modal report content, and a text expression mode of the news to be evaluated, to obtain the credibility evaluation result, including:

[0289] performing content consistency verification on the provenance information and the news to be evaluated to obtain a first verification result;

[0290] performing content consistency verification on the multi-modal content feature and a credible data source to obtain a second verification result;

[0291] performing content consistency verification on the multi-source multi-modal report content and the news to be evaluated to obtain a third verification result;

[0292] performing content self-consistency verification on the news to be evaluated according to a text expression mode of the news to be evaluated to obtain a fourth verification result;

[0293] determining the credibility evaluation result according to the first verification result, the second verification result, the third verification result, and the fourth verification result.

[0294] In an optional implementation, the credibility evaluation module 1330 performs content self-consistency verification on the news to be evaluated according to a text expression mode of the news to be evaluated to obtain a fourth verification result, including:

[0295] performing analysis on a text logic of the news to be evaluated to obtain a logic analysis result;

[0296] verifying an event causal relationship associated with the news to be evaluated to obtain a causal relationship verification result;

[0297] integrating the logic analysis result and the causal relationship verification result to obtain the fourth verification result.

[0298] In an optional implementation, the credibility evaluation module 1330 is configured to:

[0299] obtain standard record information of the target object in a cross-language credible data source;

[0300] translate the news to be evaluated into a target language corresponding to the standard record information;

[0301] performing content consistency comparison between the translated news to be evaluated and the standard record information, to obtain a fifth verification result;

[0302] determining the credibility evaluation result according to the first verification result, the second verification result, the third verification result, the fourth verification result, and the fifth verification result.

[0303] In an optional implementation, after obtaining the credibility evaluation result, the credibility evaluation module 1330 is configured to:

[0304] generating a news label corresponding to the news to be evaluated according to the credibility evaluation result and a key feature of the news to be evaluated;

[0305] uploading the news to be evaluated and the news label to a block chain, so that each verification node in the block chain performs credibility evaluation on the news to be evaluated again based on the news label and real-time dynamic data sources.

[0306] In addition, other specific details of the embodiments of the present disclosure have been described in detail in the above method embodiments, and will not be repeated here.

[0307] Exemplary Storage Medium

[0308] The storage medium of the exemplary embodiments of the present disclosure is described below.

[0309] In the exemplary embodiments, the above method can be implemented by a program product, for example, a portable compact disc read-only memory (CD-ROM) including program codes and can be run on a device, for example, a personal computer. However, the program product of the present disclosure is not limited to this, and in this document, the readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or apparatus.

[0310] The program product can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, be but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0311] A computer readable signal medium can include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated signal can take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium can be any computer readable medium that can be involved in

[0312] The code can be transmitted according to any media, including but not limited to wireless, wire line, optical fiber cable, Rf, etc. or any suitable combination of the above.

[0313] The code for carrying out operations for embodiments of the disclosure can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++, etc., or conventional procedural programming languages, such as the "C" programming language or similar programming languages. The code can execute entirely on the user's computing device, partly on the user's computing device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device or entirely on the remote computing device or server. In the latter scenario, the remote computing device can be connected to the user's computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computing device, such as through the Internet using an Internet Service Provider (ISP). The communication between the computing device and the external computing device can be facilitated via a communication interface.

[0314] Exemplary Electronic Device

[0315] Reference Figure 14 An electronic device according to an exemplary embodiment of the disclosure is described.

[0316] Figure 14 The electronic device 1400 shown is merely an example and should not limit the scope of functionality or use of any embodiments of the disclosure.

[0317] As Figure 14 shown, the electronic device 1400 is in the form of a general computing device. Components of the electronic device 1400 can include, but are not limited to, at least one processor 1410, at least one memory 1420, a bus 1430 connecting different system components, including the memory 1420 and the processor 1410, a display unit 1440.

[0318] The memory stores program codes which can be executed by the processor 1410, so that the processor 1410 performs steps according to various exemplary embodiments of the disclosure described in the above "Exemplary Methods" section of the specification. For example, the processor 1410 can perform steps as described in the above "Exemplary Methods" section of the specification, by executing the program codes stored in the memory.Figure 1 The method steps and the like shown.

[0319] The memory 1420 can include volatile memory, such as random access memory (RAM) 1421 and / or cache memory 1422, and can further include nonvolatile memory, such as read-only memory (ROM) 1423.

[0320] The memory 1420 can also include a program / utility 1424 having a set (at least one) of program modules 1425, including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which can give rise to an implementation of a network environment, individually or in some combination.

[0321] The bus 1430 can include a data bus, an address bus, and a control bus.

[0322] The electronic device 1400 can also communicate with one or more external devices 1500, such as a keyboard or a pointing device, via an input / output (I / O) interface 1450. The electronic device 1400 can further include a display unit 1440, which is connected to the input / output (I / O) interface 1450, for displaying. The electronic device 1400 can also communicate with one or more networks, such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet, via a network adapter 1460. As depicted, the network adapter 1460 communicates with the other modules of the electronic device 1400 via the bus 1430. It should be appreciated that although not shown, other hardware and / or software modules could be used in conjunction with the electronic device 1400. These include, but are not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.

[0323] It should be noted that although several modules or sub-modules of an apparatus are mentioned in the foregoing detailed description, such a division is merely exemplary and not mandatory. Indeed, according to an embodiment of the disclosure, features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, a feature or function of one unit / module described above can be further divided into several units / modules.

[0324] Furthermore, although the operations of the methods of the present disclosure are described in a particular, sequential order, this order is not mandatory and is merely illustrative. Additionally or alternatively, certain steps can be performed in parallel, in a different order, or omitted, combined, or split into further steps.

[0325] While the spirit and principles of the present disclosure have been described with reference to several specific embodiments, it is to be understood that the present disclosure is not limited to the specific embodiments disclosed and that the division of the aspects is not meant to imply that features from the aspects cannot be combined to benefit from the disclosure, but is merely for convenience of presentation. The present disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the claims appended hereto.

Claims

1. A news credibility evaluation method, characterized in that: include: Match the news to be evaluated with the pre-built object knowledge graph to obtain the traceability information of the target object described in the news to be evaluated; The object knowledge graph is used to structuredly store the traceability information of each object; Extracting multimodal content features corresponding to the news to be evaluated; The credibility of the news to be evaluated is evaluated based on the tracing information, the multimodal content features and the text expression mode of the news to be evaluated to obtain the credibility evaluation result.

2. The method according to claim 1, characterized in that The method further comprises: Acquiring multi-source and multi-modal reporting content for the target object; The credibility of the news to be evaluated is evaluated based on the tracing information, the multimodal content features, the multi-source multimodal reporting content and the text expression mode of the news to be evaluated to obtain the credibility evaluation result.

3. The method according to claim 2, characterized in that The object knowledge graph is constructed in the following way: Obtaining an object context for each object, wherein the object context includes at least one of the following: object history, object-associated person relationship, and object background information; Performing knowledge modeling on the object context to obtain an object knowledge framework; Performing knowledge extraction on the object knowledge framework to obtain multi-source knowledge elements; The multi-source knowledge elements are subjected to knowledge fusion to obtain the object knowledge graph.

4. The method according to claim 3, characterized in that The performing knowledge modeling on the object context to obtain an object knowledge framework includes: Performing data cleaning on the object context to obtain a standard object context; Performing knowledge modeling based on the standard object context to obtain the object knowledge framework; The data cleaning process includes at least one of the following: time standardization, place name remapping and ambiguity elimination.

5. The method according to claim 3, characterized in that The step of extracting knowledge from the object knowledge framework to obtain multi-source knowledge elements includes: Performing knowledge extraction on the text information in the object knowledge framework to obtain text knowledge elements; performing knowledge extraction on the image information in the object knowledge framework to obtain visual knowledge elements; Knowledge extraction is performed on the video information in the object knowledge framework to obtain audio-visual knowledge elements.

6. The method according to claim 3, characterized in that The performing knowledge fusion on the multi-source knowledge elements to obtain the object knowledge graph includes: Performing conflict resolution processing on the multi-source knowledge elements to obtain conflict-free knowledge elements; Performing entity alignment processing on candidate entities with potential corresponding relationships in the de-conflicting knowledge elements to obtain entity alignment results; Knowledge normalization modeling is performed based on the entity alignment results to obtain the object knowledge graph.

7. The method according to claim 5, characterized in that The performing conflict resolution processing on the multi-source knowledge elements to obtain conflict-free knowledge elements includes: identifying conflict types existing in the multi-source knowledge elements; Conflict resolution processing is performed on the multi-source knowledge elements according to a conflict resolution strategy corresponding to the conflict type to obtain the conflict-free knowledge elements.

8. A news credibility assessment device, characterized in that: include: An acquisition module is used to match the news to be evaluated with a pre-built object knowledge graph to obtain the traceability information of the target object described in the news to be evaluated; The object knowledge graph is used to structuredly store the traceability information of each object; A feature extraction module, configured to extract multimodal content features corresponding to the news to be evaluated; The credibility evaluation module is used to perform credibility evaluation on the news to be evaluated based on the traceability information, the multimodal content features and the text expression mode of the news to be evaluated to obtain the credibility evaluation result.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: processor; as well as a memory for storing executable instructions of the processor; The processor is configured to perform the method according to any one of claims 1 to 7 by executing the executable instructions.

Images