Multi-information clue integration and conclusion generation method and system

Abstract

The invention discloses a multivariate information clue integration and conclusion generation method and system, and the method specifically comprises the steps: S1, carrying out the vectorization of clues through a pre-training language model, and constructing a weighted graph with the clues as nodes and the semantic similarity as an edge weight; clustering the weighted graph by adopting a community discovery algorithm, extracting theme tags, merging clues with similarity exceeding a threshold value through clustering, and retaining metadata; s2, performing text preprocessing on the classification clues, and inputting preconditions and hypotheses into a natural language inference model to obtain a support, inverse or neutral relationship; correcting a model result in combination with a symbol logic matching rule, and outputting a relationship type and confidence; and S3, generating a logic network diagram by taking the clues as nodes and the judged logic relations as edges, and performing weighting according to confidence and removing weak correlation relations through threshold pruning to obtain a global association logic network.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and mainly to a method and system for integrating multiple information clues and generating conclusions. Background Technology

[0002] With the rapid development of big data, artificial intelligence, and information retrieval technologies, information acquisition channels are becoming increasingly diversified. Currently, mainstream information processing systems typically employ methods such as multi-channel retrieval and multi-model fusion to acquire large amounts of raw information from different data sources, algorithms, or perspectives. These technologies are widely used in fields such as search engines, intelligent question answering, recommendation systems, and public opinion analysis.

[0003] In practical applications, common multi-path recall and information fusion technologies include:

[0004] 1. Multi-source data retrieval: Retrieve relevant information in parallel through structured, semi-structured and unstructured data sources.

[0005] 2. Multi-model integration: Utilize different algorithm models to perform independent reasoning on the same problem, thereby improving the comprehensiveness of recall.

[0006] 3. Information summarization and automatic summarization: Automatically summarize the search results, extract key information, and assist users in understanding and decision-making.

[0007] These technologies have greatly enhanced the breadth and depth of information acquisition, providing a solid foundation for the analysis and decision-making of complex problems.

[0008] While current mainstream information processing technologies have enabled the acquisition of information from multiple channels, they have several inherent flaws in the process of fusion and conclusion generation, specifically:

[0009] 1. Information Fusion Distortion and Broken Reasoning Chains: Existing technologies generally adopt a process of "first summarizing each piece of information, then centrally summarizing it." In this process, when the original information is processed independently, its deep-seated contextual logic, causal relationships, and evidence chains are easily disrupted or lost. This leads to the final fused conclusions containing omissions of details, semantic distortions, and even factual errors, and this distortion is irreversible.

[0010] 2. Insufficient handling of conflicts and redundancy: Because each information retrieval channel is processed in isolation, the system lacks a global perspective to compare all original information horizontally. Therefore, it is difficult to effectively identify and handle contradictory, overlapping, or complementary relationships between information from different sources. Furthermore, inconsistent summary standards across channels lead to inconsistencies in the granularity and focus of the final results, reducing the cohesion and comparability of the conclusions.

[0011] 3. Insufficient reliability and transparency of conclusions: The final conclusion is often a "black box" text, making it difficult for users to trace the basis of their judgments, to know whether the conclusion is based on strong or weak evidence, or to understand whether there are opposing viewpoints. This makes it difficult to assess the reliability of the conclusions and limits their application value in serious decision-making scenarios. Summary of the Invention

[0012] To address the above shortcomings, this invention proposes a novel method centered on "clue-based integration," aiming to fundamentally change the information processing paradigm. According to a first aspect of this invention, a method for multi-source information clue-based integration and conclusion generation is proposed, with the following specific steps:

[0013] S1. Vectorize the clues using a pre-trained language model and construct a weighted graph with clues as nodes and semantic similarity as edge weights; use a community detection algorithm to cluster the weighted graph and extract topic tags; merge clues with similarity exceeding a threshold by clustering and retain metadata.

[0014] S2. Perform text preprocessing on the classification clues, input the premises and assumptions into the natural language inference model to obtain support, opposition or neutral relations; combine symbolic logic matching rules to correct the model results, and output the relation type and confidence level;

[0015] S3. Generate a logical network graph using clues as nodes and determined logical relationships as edges. Assign weights based on confidence levels and remove weak correlations through threshold pruning to obtain a global relational logical network.

[0016] Preferably, before step S1, the raw data is further processed into threads, specifically including:

[0017] Receive raw data from the multi-source recall channels;

[0018] In the original data, information of different formats and structures is parsed and extracted into text, and uniformly converted into standardized clue units;

[0019] Each clue unit contains core text, source information, timestamps, and metadata to form a computable and traceable set of clues.

[0020] Preferably, after step S3, a weight evaluation step is performed, in which a comprehensive weight is calculated based on the semantic strength, confidence, and source reliability of the nodes and edges, which serves as the quantitative basis for the conclusion generation.

[0021] Preferably, the formula for calculating the comprehensive weight is as follows:

[0022] W i =αA src,i +βR time,i+γI info,i +δS support,i

[0023] A src,i Indicating authoritative source, R time,i Indicating timeliness, I info,i S represents the uniqueness of information, calculated through inverse frequency or topic distribution. support,i The support level is represented by α, β, γ, and δ, which represent weight adjustment coefficients. W i This represents the weighted score.

[0024] Preferably, after the weight evaluation step, the conclusion generation step is performed, where high-weight nodes are selected from the weighted graph, Dijkstra's algorithm is used to identify supporting paths, and conclusions are generated according to the template of [main viewpoint, key evidence, time source].

[0025] Preferably, during the conclusion generation process, low-weight nodes are added to enhance the contextual description and logical coherence.

[0026] Preferably, in the conclusion generation step, when there is an opposing relationship between high-weight nodes, the MaxSAT conflict detection algorithm is used to verify and mark it as [disputed]; when the weight of a topic branch node is lower than the threshold, it is marked as [insufficient evidence]; when a viewpoint is supported by multiple independent high-authority sources, it is marked as [strong evidence].

[0027] Preferably, in the conclusion generation step, when there is an opposing relationship between high-weight nodes, the MaxSAT conflict detection algorithm is used to verify and mark it as [disputed]; when the weight of a topic branch node is lower than the threshold, it is marked as [insufficient evidence]; when a viewpoint is supported by multiple independent high-authority sources, it is marked as [strong evidence].

[0028] Preferably, by establishing an inverted index of conclusion number and clue node ID, the traceability query from conclusion to evidence node can be realized.

[0029] According to a second aspect of the present invention, a computer program product is provided, on which one or more computer programs are stored, which, when executed by a computer processor, implement the method described above.

[0030] The above-described one or more technical solutions in the embodiments of this application have at least one of the following technical effects:

[0031] 1. Improved accuracy and completeness of conclusions: This invention, through a "clue-based integration" step, fully preserves the details and context of all original information units, fundamentally avoiding the loss of details and semantic distortion caused by premature information compression in the traditional "summarize first, then generalize" model. Therefore, the final conclusions are based on a more comprehensive and accurate evidentiary foundation, significantly improving their accuracy and completeness.

[0032] 2. Effective management of conflicting and redundant information: This invention constructs a global "association network," enabling horizontal comparison and analysis of all information clues within a unified framework. This allows the system to accurately identify and effectively manage duplicate, complementary, and even contradictory information from different sources. This addresses the pain point of existing technologies, which struggle to detect and resolve conflicts due to isolated information processing.

[0033] 3. Enhanced Reliability and Interpretability of Conclusions: The "intelligent weight evaluation" mechanism introduced in this invention enables the generation of conclusions to favor more authoritative and reliable sources, thereby improving the objectivity and reliability of the conclusions. More importantly, the "uncertainty labeling" function transforms the "black box" state of the conclusions into transparency, clearly revealing the points of contention and weak evidence in the conclusions to the user, greatly enhancing the interpretability and credibility of the conclusions, and facilitating users to make prudent secondary judgments and decisions. Attached Figure Description

[0034] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments and, together with the description, serve to explain the principles of the invention. Other embodiments and many anticipated advantages of the embodiments will be readily recognized as they become better understood through reference to the following detailed description. Elements in the drawings are not necessarily to scale. The same reference numerals refer to corresponding similar parts.

[0035] Figure 1 A flowchart illustrating a method for integrating multiple information clues and generating conclusions according to an embodiment of the present invention is shown.

[0036] Figure 2 This is a schematic diagram of the structure of a computer system suitable for implementing the electronic devices of the present application embodiments. Detailed Implementation

[0037] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0038] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0039] Figure 1 A flowchart illustrating a method for integrating multiple information clues and generating conclusions according to an embodiment of the present invention is shown, as follows: Figure 1 As shown:

[0040] Step 1: Clue Processing

[0041] The system receives raw data from multiple recall channels and standardizes information of different formats and structures into "clue units." Each clue unit contains core text, source information, timestamps, and other metadata. This step can be achieved through rule-based parsing and text extraction methods to form a computable and traceable clue set, providing complete input for subsequent analysis.

[0042] Step 2: Building the Connection Network

[0043] A global analysis is performed on all the clues generated in the first step to generate a global association network with clues as nodes and logical relationships as edges, thereby enabling conflict identification, redundancy handling, and logical chain maintenance.

[0044] 1) Topic Classification and Deduplication

[0045] Vectorize the clues:

[0046] V i =f(lineage) i )

[0047] Among them, V i The vector representation of the i-th clue is obtained using a pre-trained language model.

[0048] Construct a weighted graph G = (V, E), where nodes V represent cues, and the weights of edges E are determined by the semantic similarity between cues, sim(v). i ,v j )Decide.

[0049] The Louvain algorithm is used to perform community discovery on the graph, dividing similar clues into topic communities, and performing topic modeling on clues within the community to extract keywords and generate topic tags; clues with similarity exceeding a threshold are further merged using K-Means clustering, while retaining the source and metadata of the merged clues to ensure traceability.

[0050] 2) Logical Relationship Mining

[0051] After categorizing the clues, the text is first preprocessed, including sentence segmentation, entity recognition, and synonym normalization. Then, the premises and assumptions are input into the Natural Language Inference (NLI) model to obtain the probability distributions of support, opposition, and neutrality as preliminary judgments. Subsequently, symbolic logic matching rules are used to convert the sentences into logical expressions, which are corrected by implication (premise leads to conclusion), contradiction (premise and conclusion conflict), or uncertainty (cannot be derived). For ambiguous or polysemous cases, the model probability and logical matching results are combined to make a comprehensive judgment, outputting the final relationship and confidence level, thus balancing linguistic flexibility and logical precision in the construction of the logical network.

[0052] 3) Logical network generation

[0053] The categorized clues and the support, opposition, and neutral relationships determined by logical relations are represented as a graph structure G = (V, E), where nodes V represent clues and edges E represent logical relations. Edge weights are assigned based on confidence and logical strength. The confidence calculation algorithm is as follows:

[0054] Define the prior probability that a clue is judged as "supporting" the conclusion:

[0055]

[0056] in, ω represents the time correlation coefficient. T This represents the time weighting coefficient. Indicates the source's authority rating, ω A The source authority weight is represented by α, which is a coefficient that adjusts the balance between source and time weights.

[0057] Based on the logical and semantic scores output by the model, define the conditional probability:

[0058]

[0059] in, The logical consistency score is represented by ω, which is output by symbolic logic matching. L Represents logical consistency weight. ω represents the semantic similarity between the i-th clue and the target conclusion. S This represents the semantic similarity weight.

[0060] Furthermore, calculate the fusion confidence:

[0061]

[0062] Where ω1 and ω2 are the importance weights of observation and prior, respectively.

[0063] For each validation sample, backpropagation is used to update the weights of each factor based on the error between the predicted result and the manually labeled true value.

[0064]

[0065] Where η represents the learning rate, y i Indicates a true confidence label.

[0066] Output overall confidence level C i It is used for edge weight assignment and threshold pruning in logical networks.

[0067] Threshold pruning is applied to edge weights to highlight the main supporting paths and remove weakly correlated or uncertain relationships.

[0068] The fusion confidence calculation of this invention can maintain stable output when there is inconsistency among multiple sources of evidence. It achieves cross-topic adaptation through dynamic weight adjustment, and integrates four-dimensional features of semantics, logic, source, and time, taking into account both linguistic ambiguity and logical rigor.

[0069] Step 3: Weight Evaluation

[0070] It provides quantitative evidence for the generation of conclusions and supports the judgment of consistency and reliability assessment of multi-source information.

[0071] The formula for calculating the overall weight is as follows:

[0072] W i =αA src,i +βR time,i +γI info,i +δS support,i

[0073] A src,i The score indicates the authority of the source (0-1 standardized score). Different basic credibility scores are set for different information sources, such as 0.9 for academic journals, 0.8 for official media, and 0.5 for ordinary websites.

[0074] R time,i To indicate timeliness, exponential decay is used: R time,i =e -λΔt Based on the information release time, newer leads are given higher weight for topics that require timeliness.

[0075] I info,i The uniqueness of information is represented by inverse frequency or topic distribution.

[0076] S support,i This indicates the level of support, which is accumulated by other high-weight nodes.

[0077] α, β, γ, δ represent weighting adjustment coefficients, which can be adjusted according to requirements.

[0078] By weighting and combining the above dimensions, a quantified weight value is ultimately assigned to each clue node. This method, while ensuring objectivity, also enhances the comprehensive consideration of multi-dimensional factors.

[0079] Step 4: Conclusion Generation and Labeling

[0080] It generates logically clear and traceable conclusions; it provides uncertainty annotation and evidence chain visualization to improve the reliability and transparency of conclusions.

[0081] 1) Conclusion Synthesis

[0082] Extract high-weight nodes V from the weighted graph high ={v i |W i >T}, using Dijkstra's algorithm to identify supporting paths and construct a logical skeleton, generates conclusions through a template of "main viewpoint + key evidence + time source", and supplements low-weight nodes to enhance the contextual description.

[0083] 2) Conclusion labeling

[0084] The clues are evaluated and marked as "controversial," "insufficient evidence," or "strong evidence," using the following methods:

[0085] Dispute detection: If there is an opposing relationship between high-weight nodes, the MaxSAT conflict detection algorithm is used for verification, and a "dispute exists" mark is automatically added to the corresponding paragraph of the conclusion.

[0086] Insufficient evidence detection: If the weight of all nodes under a certain topic branch is lower than the threshold (e.g., 0.3), then add an "insufficient evidence" mark to the corresponding position of the conclusion;

[0087] Strong evidence detection: If a viewpoint is supported by at least three independent sources and nodes with an authority score greater than a preset value, it is confirmed by a multi-source consistency verification algorithm, and a "strong evidence" mark is added to the conclusion.

[0088] 3) Tracing the chain of evidence

[0089] By binding the conclusion number with the clue node ID, an inverted index is constructed to enable fast querying of "conclusion → evidence node".

[0090] In summary, this invention achieves complete preservation of original information, clear reasoning chain, automatic management of conflict redundancy, traceable conclusions and annotation of uncertainty through a closed loop of "clue processing, association network construction, weight evaluation and conclusion generation and annotation", and prioritizes highly authoritative and strongly supporting evidence, thereby significantly improving the accuracy and reliability of the conclusions.

[0091] The main improvements of this invention are as follows:

[0092] 1. Addressing "Information Distortion" -> Employing "Clues-Based Integration" to Preserve Original Information: This invention treats each piece of recalled original information as an independent "clue," fully preserving its content and metadata. This avoids premature compression and summarization of information, ensuring the integrity of the reasoning chain and providing a solid foundation for subsequent accurate analysis and reliable conclusion generation.

[0093] 2. Addressing "Conflict and Redundancy" -> Employing "Relationship Network Construction" for Global Analysis: This invention does not isolate clues but rather places them within a unified framework for global analysis. Through structured classification, semantic deduplication, and logical relationship mining (such as identifying supporting and opposing relationships), a visualized "clue relationship network" is constructed. This allows for the accurate identification and handling of conflicts and redundancies between information, and enables subsequent operations to be performed under a unified standard.

[0094] 3. Addressing "Insufficient Reliability" -> Employing "Intelligent Weighting Evaluation" and "Uncertainty Labeling": To enhance the reliability and transparency of the conclusions, this invention introduces two mechanisms. First, intelligent weighting evaluation is performed on clues based on factors such as the authority and timeliness of the source. Second, when generating conclusions, explicit uncertainty labeling is applied to parts based on conflicting or insufficient evidence. These two points ensure that the final conclusion not only reflects the most credible evidence but also honestly reveals its limitations.

[0095] This invention provides a core capability that efficiently and reliably extracts insights and generates highly credible conclusions from massive, diverse, and even contradictory information, and therefore has extremely broad application prospects.

[0096] Its technology can be widely applied in the following fields:

[0097] 1. Intelligent Search Engine and Question Answering System: For complex and open-ended user questions, this invention can replace the traditional link list, directly integrate information from multiple sources such as news, academic papers, and industry reports, and generate a structured, multi-dimensional, comprehensive answer that includes both positive and negative viewpoints, greatly improving the efficiency and depth of users' knowledge acquisition.

[0098] 2. Market Research and Business Intelligence (BI): When conducting market analysis, competitor tracking, or user needs research, companies need to process a large amount of industry news, financial reports, social media comments, and research reports. This invention can automatically integrate and analyze this information, quickly generate insight reports on market trends, brand reputation, and potential risks, and highlight contradictory signals in the market, providing strong data support for business decisions.

[0099] 3. Public Opinion and Policy Analysis: Government agencies or social organizations need to understand public opinion on a policy or social event in real time. This invention can efficiently aggregate public opinion information from major media and online platforms, automatically sort out the main viewpoints, arguments, and sentiments of all parties, identify and filter low-quality false information, and generate an objective and comprehensive public opinion analysis report.

[0100] 4. Academic Research and Literature Review: When researchers enter a new field, they need to spend a lot of time reading literature. This invention can serve as an intelligent research assistant, automatically capturing relevant papers on specific topics, constructing a knowledge network of the current research status in the field, clearly displaying mainstream academic viewpoints, research branches, remaining controversies, and future research directions, significantly accelerating the process of scientific research and innovation.

[0101] In conclusion, this invention, as a fundamental information processing and cognitive intelligence technology, possesses enormous application potential and commercial value in any scenario requiring in-depth analysis and integration of complex and diverse information to form reliable decisions.

[0102] The following is for reference. Figure 2 It shows a schematic diagram of the structure of a computer system 200 suitable for implementing electronic devices according to embodiments of the present application. Figure 2 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0103] like Figure 2 As shown, the computer system 200 includes a central processing unit (CPU) 201, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 202 or programs loaded from storage section 208 into random access memory (RAM) 203. The RAM 203 also stores various programs and data required for the operation of the system 200. The CPU 201, ROM 202, and RAM 203 are interconnected via a bus 204. An input / output (I / O) interface 205 is also connected to the bus 204.

[0104] The following components are connected to I / O interface 205: an input section 206 including a keyboard, mouse, etc.; an output section 207 including a liquid crystal display (LCD) and speakers, etc.; a storage section 208 including a hard disk, etc.; and a communication section 209 including a network interface card such as a LAN card and a modem, etc. The communication section 209 performs communication processing via a network such as the Internet. A drive 210 is also connected to I / O interface 205 as needed. A removable medium 211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 210 as needed so that computer programs read from it can be installed into storage section 208 as needed.

[0105] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 209, and / or installed from removable medium 211. When the computer program is executed by central processing unit (CPU) 201, it performs the functions defined in the methods of this application. It should be noted that the computer-readable storage medium of this application can be a computer-readable signal medium or a computer-readable storage medium or any combination thereof. The computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable storage medium other than a computer-readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0106] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0107] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0108] The modules described in the embodiments of this application can be implemented in software or in hardware.

[0109] In another aspect, this application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: S1. Vectorize cues using a pre-trained language model, construct a weighted graph with cues as nodes and semantic similarity as edge weights; cluster the weighted graph using a community detection algorithm and extract topic tags, and merge cues with similarity exceeding a threshold by clustering and retaining metadata; S2. Perform text preprocessing on the categorized cues, input the premises and assumptions into a natural language inference model to obtain support, opposition, or neutral relationships; correct the model results by combining symbolic logic matching rules, and output the relationship type and confidence level; S3. Generate a logic network graph with cues as nodes and determined logical relationships as edges, assign weights according to confidence level and remove weakly related relationships through threshold pruning to obtain a global association logic network.

[0110] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for multi-element information cueing integration and conclusion generation, characterized in that, Comprising: S1, vectorizing clues by using a pre-trained language model, constructing a weighted graph with clues as nodes and semantic similarity as edge weights; adopting a community discovery algorithm to cluster the weighted graph and extract topic labels, merging and retaining metadata for clues with similarity exceeding a threshold value through clustering; S2, performing text preprocessing on the classified clues, inputting premises and hypotheses into a natural language inference model to obtain support, opposition or neutral relationships; combining symbolic logic matching rules to correct the model results, and outputting the relationship type and confidence; S3, generating a logical network graph with clues as nodes and determined logical relationships as edges, weighting according to confidence and removing weakly related relationships through threshold pruning to obtain a global associated logical network.

2. The method of claim 1, wherein, The step S1 further comprises clue processing of original data, specifically comprising: receiving original data from multiple source recall channels; in the original data, parsing and text extraction are performed on information of different formats and structures, and are uniformly converted into standardized clue units; wherein each clue unit contains core text, source information, timestamp and metadata to form a computable and traceable clue set.

3. The method of claim 1, wherein, After the step S3, a weight evaluation step is entered, in which the comprehensive weight is calculated according to the semantic strength, confidence and source reliability of nodes and edges, which is used as a quantitative basis for conclusion generation.

4. The method of claim 3, wherein, The calculation formula of the comprehensive weight is as follows: W i = αA src,i + βR time,i + γI info,i + δS support,i A src,i represents source authority, R time,i represents timeliness, I info,i represents information uniqueness, calculated by inverse frequency or topic distribution, S support,i represents support degree, and α, β, γ and δ represent weight adjustment coefficients, W i represents weight score.

5. The method of claim 3, wherein, After the weight evaluation step, a conclusion generation step is entered, in which high-weight nodes are selected from the weighted graph, Dijkstra algorithm is used to identify support paths, and conclusions are generated according to the template of [main point of view, key evidence, time source].

6. The method of claim 5, wherein, During the conclusion generation process, low-weight nodes are supplemented to enhance the context description and logical coherence.

7. The method of claim 5, wherein, In the conclusion generation step, when there is an opposition relationship between high-weight nodes, MaxSAT conflict detection algorithm is used to verify and mark [there is a dispute]; when the weight of a topic branch node is lower than a threshold value, mark [evidence is insufficient]; when a certain point of view is supported by multiple independent high-authority sources, mark [strong evidence].

8. The method according to any one of claims 1 to 7, characterized in that, By establishing an inverted index of conclusion number and clue node ID, traceable query from conclusion to evidence node is realized.

9. A computer program product, characterised in that, A computer program is stored thereon, which implements the method of any one of claims 1-8 when executed by a processor.

10. A computing system, comprising: Comprising a processor and a memory, the processor is configured to execute the method of any one of claims 1-8.

Images