A method, apparatus, device, and storage medium for fact verification

Abstract

The present application discloses a fact verification method, apparatus, device, and storage medium, which relate to the field of computer technology and are applied to a fact verification system constructed based on a large model and a preset multi-agent falsification system. The method includes: parsing the received information to be processed based on the large model, and determining a hypothesis to be verified corresponding to the information to be processed by using the obtained information parsing result; generating multi-dimensional null hypotheses based on the large model and the hypothesis to be verified, and determining a target falsification strategy by using the determined multiple null hypotheses and data resource information; collecting evidence through the preset multi-agent falsification system and the multiple null hypotheses, and falsifying each of the null hypotheses by using the evidence collection result and the target falsification strategy to obtain a hypothesis falsification result; determining a fact verification result corresponding to the information to be processed based on the evidence collection result, the hypothesis falsification result, and the preset multi-dimensional hypothesis verification quantization rule. The present application can effectively improve the efficiency, comprehensiveness, reliability, and adaptability of fact verification.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and particularly to a fact verification method, device, equipment and storage medium. Background Art

[0002] With the rapid development of information technology, the amount of information faced by humans has increased exponentially, and the information contains a large number of hypotheses, theories or claims that need to be fact-verified.

[0003] However, in the implementation process of traditional verification schemes, on the one hand, since most steps from hypothesis decomposition to evidence collection and then to the formation of verification conclusions require manual intervention and rely on manual analysis and professional judgment, this is not only time-consuming and laborious, but also easily affected by cognitive biases and professional limitations; on the other hand, it usually only verifies a single dimension, making the verification not comprehensive enough and resulting in low accuracy of the conclusions. In addition, this scheme is difficult to adapt to multi-field applications, resulting in poor adaptability. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a fact verification method, device, equipment and storage medium, which can effectively realize the automation of fact verification, thereby greatly reducing the dependence on manual work and the resulting subjective biases, and improving the efficiency, systematicness, comprehensiveness, reliability and adaptability of fact verification. The specific solutions are as follows:

[0005] In a first aspect, the present application provides a fact verification method, which is applied to a fact verification system constructed based on a large model and a preset multi-agent falsification system, and includes:

[0006] Parsing the received information to be processed based on the large model, and using the obtained information parsing result to determine a hypothesis to be verified corresponding to the information to be processed;

[0007] Generating multi-dimensional null hypotheses based on the large model and the hypothesis to be verified, and using the determined multiple null hypotheses and data resource information to determine a target falsification strategy;

[0008] Collecting evidence through the preset multi-agent falsification system and the multiple null hypotheses, and using the evidence collection result and the target falsification strategy to falsify each of the null hypotheses to obtain a hypothesis falsification result;

[0009] Determining a fact verification result corresponding to the information to be processed based on the evidence collection result, the hypothesis falsification result and a preset multi-dimensional hypothesis verification quantization rule.

[0010] Optionally, parsing the to-be-processed information received based on the large model and determining a to-be-verified hypothesis corresponding to the to-be-processed information by using the obtained information parsing result includes:

[0011] Parsing the to-be-processed information received based on the large model to obtain an information parsing result;

[0012] Triggering a hypothesis extraction operation and a meaningless word removal operation based on the information parsing result to determine a to-be-verified hypothesis corresponding to the to-be-processed information.

[0013] Optionally, after determining the to-be-verified hypothesis corresponding to the to-be-processed information by using the obtained information parsing result, it further includes:

[0014] When it is determined that the preset hypothesis splitting condition is satisfied based on the information parsing result, splitting the to-be-verified hypothesis to determine multiple to-be-verified sub-hypotheses, and generating multi-dimensional null hypotheses respectively based on the large model and each to-be-verified sub-hypothesis.

[0015] Optionally, generating multi-dimensional null hypotheses based on the large model and the to-be-verified hypothesis, and determining a target falsification strategy by using the determined multiple null hypotheses and data resource information includes:

[0016] Generating multi-dimensional null hypotheses based on the characteristic information of the large model and the to-be-verified hypothesis to obtain multiple null hypotheses that form a complementary relationship;

[0017] Analyzing the importance degree and verification value of each null hypothesis for the to-be-verified hypothesis to determine a first weight corresponding to each null hypothesis based on the analysis result;

[0018] Determining a target falsification strategy based on the characteristic information, data resource information, and intelligent agent resource information of each null hypothesis.

[0019] Optionally, determining a fact verification result corresponding to the to-be-processed information based on the evidence collection result, the hypothesis falsification result, and a preset multi-dimensional hypothesis verification quantification rule includes:

[0020] Analyzing the source, quality, and relevance of each piece of evidence in the evidence collection result to determine a second weight corresponding to each piece of evidence based on the evidence analysis result;

[0021] Analyzing the falsification degree of each null hypothesis based on the hypothesis falsification result to obtain a falsification degree analysis result corresponding to each null hypothesis;

[0022] Determine the corresponding multi-dimensional hypothesis verification score based on the alternative explanation adjustment factor, the first weights corresponding to the respective null hypotheses, the results of the degree of falsification analysis, and the second weights corresponding to the respective pieces of evidence.

[0023] Optionally, the evidence collection is performed through the preset multi-agent falsification system and the multiple null hypotheses, and each null hypothesis is falsified by using the evidence collection result and the target falsification strategy, including:

[0024] The data collection agent in the preset multi-agent falsification system collects evidence for verifying the multiple null hypotheses from a selected plurality of data sources to obtain an evidence collection result;

[0025] Based on the data collection agent, perform data cleaning operations and format conversion operations on the evidence collection result to obtain the processed evidence collection result;

[0026] Perform a correlation analysis through the data analysis agent in the preset multi-agent falsification system and the processed evidence collection result, and filter the processed evidence collection result based on the correlation analysis result to obtain the filtered evidence collection result;

[0027] Each null hypothesis is falsified through the falsification execution agent in the preset multi-agent falsification system, the filtered evidence collection result, and the target falsification strategy to obtain a hypothesis falsification result.

[0028] Optionally, after determining the fact verification result corresponding to the information to be processed based on the evidence collection result, the hypothesis falsification result, and the preset multi-dimensional hypothesis verification quantization rule, it further includes:

[0029] Perform uncertainty analysis, key evidence extraction, and verification conclusion generation based on the fact verification result and the evidence collection result to determine a fact verification report corresponding to the information to be processed.

[0030] In a second aspect, the present application provides a fact verification device, which is applied to a fact verification system constructed based on a large model and a preset multi-agent falsification system, including:

[0031] A hypothesis determination module, configured to parse the information to be processed received based on the large model, and use the obtained information parsing result to determine a hypothesis to be verified corresponding to the information to be processed;

[0032] A falsification strategy determination module, configured to generate multi-dimensional null hypotheses based on the large model and the hypothesis to be verified, and use the determined multiple null hypotheses and data resource information to determine a target falsification strategy;

[0033] The falsification result determination module is used to collect evidence through the preset multi-agent falsification system and the multiple null hypotheses, and use the evidence collection result and the target falsification strategy to falsify each null hypothesis to obtain the hypothesis falsification result;

[0034] The verification result determination module is used to determine the fact verification result corresponding to the information to be processed based on the evidence collection result, the hypothesis falsification result, and the preset multi-dimensional hypothesis verification quantization rule.

[0035] In a third aspect, the present application provides an electronic device, including:

[0036] A memory for storing a computer program;

[0037] A processor for executing the computer program to implement the steps of the foregoing fact verification method.

[0038] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, and when the computer program is executed by a processor, the steps of the foregoing fact verification method are implemented.

[0039] It can be seen that in the present application, through the fact verification system constructed based on the large model and the preset multi-agent falsification system, first, the information to be processed received is parsed based on the large model, and the verification hypothesis corresponding to the information to be processed is determined by using the obtained information parsing result; multi-dimensional null hypotheses are generated based on the large model and the verification hypothesis, and the target falsification strategy is determined by using the determined multiple null hypotheses and the data resource information; evidence is collected through the preset multi-agent falsification system and the multiple null hypotheses, and the evidence collection result and the target falsification strategy are used to falsify each null hypothesis to obtain the hypothesis falsification result; the fact verification result corresponding to the information to be processed is determined based on the evidence collection result, the hypothesis falsification result, and the preset multi-dimensional hypothesis verification quantization rule. That is to say, in the present application, through the fact verification system constructed based on the large model and the preset multi-agent falsification system, the large model is first used to parse the information to be processed to determine the verification hypothesis, and then the large model is used to generate multi-dimensional null hypotheses for the verification hypothesis, and the corresponding target falsification strategy is determined by using the determined multiple null hypotheses. Then, evidence is collected through the preset multi-agent falsification system, and the evidence collection result and the target falsification strategy are used to falsify each null hypothesis. Then, the fact verification result is determined by using the evidence collection result, the hypothesis falsification result, and the preset multi-dimensional hypothesis verification quantization rule. In this way, the automation of fact verification can be effectively realized, thereby greatly reducing the manual dependence and the resulting subjective biases, and improving the efficiency, systematicness, comprehensiveness, reliability, and adaptability of fact verification. Description of the Drawings

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on the provided drawings.

[0041] Figure 1 It is a flowchart of a fact verification method provided by this application;

[0042] Figure 2 It is a schematic structural diagram of a fact verification device provided by this application;

[0043] Figure 3 It is a structural diagram of an electronic device provided by this application. Specific embodiments

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0045] In the implementation process of traditional verification schemes, on the one hand, since most steps from hypothesis decomposition to evidence collection and then to the formation of verification conclusions require manual intervention and rely on manual analysis and professional judgment, this is not only time-consuming and laborious, but also easily affected by cognitive biases and professional limitations; on the other hand, it usually only conducts verification for a single dimension, making the verification incomplete and resulting in a low accuracy of the conclusion. In addition, this scheme is difficult to adapt to multi-field applications, resulting in poor adaptability. Therefore, this application provides a fact verification scheme, which can effectively realize the automation of fact verification, thereby greatly reducing manual dependence and the resulting subjective biases, and improving the efficiency, systematicness, comprehensiveness, reliability, and adaptability of fact verification.

[0046] See Figure 1 As shown, an embodiment of the present invention discloses a fact verification method, which is applied to a fact verification system constructed based on a large model and a preset multi-agent falsification system, including:

[0047] Step S11: Parse the received information to be processed based on the large model, and use the obtained information parsing result to determine a hypothesis to be verified corresponding to the information to be processed.

[0048] In this embodiment, in addition to the large model and the preset multi-agent falsification system, the fact verification system further includes a hypothesis understanding engine, a falsification strategy planner, and an evidence accumulation and scoring system. Among them, the hypothesis understanding engine is responsible for receiving and parsing the hypotheses expressed in natural language, and extracting the statements or hypotheses to be verified; the falsification strategy planner automatically generates multi-dimensional null hypotheses according to the characteristics of the hypotheses and designs corresponding verification methods; the multi-agent falsification execution system, that is, the preset multi-agent falsification system, is composed of a data collection agent, a data analysis agent, and a falsification implementation agent, which cooperate to execute the falsification task; the evidence accumulation and scoring system is responsible for calculating and updating the verification progress, generating the final multi-dimensional hypothesis verification score (Multi-dimensional Hypothesis Verification Score, MHVS) and report.

[0049] It should be understood that for the information to be processed expressed in natural language, in this embodiment, it is first processed by the hypothesis understanding engine. The hypothesis understanding engine is implemented based on a large language model. Specifically, using the large language model, the information to be processed input into the system is extracted for statements or hypotheses using instructions, and irrelevant tone words and other information are removed. In addition, if there are multiple questions, they are split into multiple sub-questions to facilitate the subsequent generation of null hypotheses. That is, the large model is used to parse the information to be processed received to obtain an information parsing result; based on the information parsing result, a hypothesis extraction operation and a meaningless word removal operation are triggered to determine the hypothesis to be verified corresponding to the information to be processed. When it is determined based on the information parsing result that the preset hypothesis splitting condition is satisfied, multiple sub-hypotheses to be verified are determined by splitting the hypothesis to be verified, so as to generate multi-dimensional null hypotheses based on the large model and each of the sub-hypotheses to be verified.

[0050] Step S12: Generate multi-dimensional null hypotheses based on the large model and the hypothesis to be verified, and determine the target falsification strategy using the determined multiple null hypotheses and data resource information.

[0051] In this embodiment, after obtaining the hypothesis to be verified, relevant processing will be carried out using the falsification strategy planner. The falsification strategy planner is implemented based on a large model and is responsible for automatically generating multi-dimensional null hypotheses according to the characteristics of the hypothesis based on Popper's falsification principle, and designing corresponding verification methods and execution paths. In this way, the authenticity of the main hypothesis can be verified by systematically negating the null hypothesis. Specifically, in this embodiment, multi-dimensional null hypotheses are first generated based on the large model and the characteristic information of the hypothesis to be verified to obtain multiple null hypotheses that form a complementary relationship; by analyzing the importance and verification value of each null hypothesis to the hypothesis to be verified, the first weight corresponding to each null hypothesis is determined based on the analysis results; the target falsification strategy is determined based on the characteristic information, data resource information, and agent resource information of each null hypothesis. That is to say, the specific application steps of the falsification strategy planner are as follows:

[0052] (1) Null hypothesis generation: Based on the main hypothesis, that is, the structured representation of the hypothesis to be verified, multiple complementary null hypotheses are automatically generated to ensure coverage of all key aspects of the main hypothesis. For example, for the main hypothesis of "Factor A causes result B", null hypotheses such as "There is no statistical correlation between A and B", "The temporal relationship between A and B does not conform to causal logic", and "The relationship between A and B disappears after controlling other variables" can be generated;

[0053] (2) Dimension weight assignment: According to the importance and verification value of each null hypothesis to the main hypothesis, the corresponding weight (1-10) is assigned;

[0054] (3) Falsification path planning: According to the characteristics of the null hypothesis and the available data resources and agent resources, the optimal falsification execution path is planned, including data requirements and analysis methods; among them, the data requirements include whether network search is required and the scope of network search; the analysis methods include whether external knowledge obtained needs to be referred to.

[0055] The above steps are all implemented using a large model. By customizing instructions and inputting the hypothesis to be verified into the large model, the large model can output the corresponding one or more null hypotheses, null hypothesis weights, and path planning at one time.

[0056] Step S13: Evidence collection is carried out through the preset multi-agent falsification system and the multiple null hypotheses, and each null hypothesis is falsified using the evidence collection result and the target falsification strategy to obtain the hypothesis falsification result.

[0057] In this embodiment, after obtaining the null hypothesis and the corresponding target falsification strategy, the corresponding processing will be carried out using a preset multi-agent falsification system. The preset multi-agent falsification system consists of a data collection agent, a data analysis agent, and a falsification execution agent. These agents work together to execute the verification plan designed by the falsification strategy planner. Specifically, in this embodiment, first, the data collection agent in the preset multi-agent falsification system collects evidence for verifying the multiple null hypotheses from a selected plurality of data sources to obtain an evidence collection result; based on the data collection agent, data cleaning operations and format conversion operations are performed on the evidence collection result to obtain the processed evidence collection result; correlation analysis is performed on the processed evidence collection result through the data analysis agent in the preset multi-agent falsification system to filter the processed evidence collection result based on the correlation analysis result to obtain the filtered evidence collection result; the falsification execution agent in the preset multi-agent falsification system, the filtered evidence collection result, and the target falsification strategy are used to falsify each of the null hypotheses to obtain a hypothesis falsification result.

[0058] It should be understood that regarding the agents in the preset multi-agent falsification system, the data collection agent is used to obtain the data required for verification, that is, the collection of evidence, from various data sources (such as academic literature databases, public databases, network resources, etc.). And this agent has the ability to access multi-source data and can perform data cleaning and format conversion to unify the data into the JSON (JavaScript Object Notation) format. In addition, the data collection agent can be divided into multiple professional sub-agents according to needs, such as a literature retrieval agent, a web crawler agent, a database query agent, etc. Through different sub-agents, customized processing can be carried out according to each different data source type, including using different interface call methods, retrieval statement construction methods, result return structures, etc., and data can be retrieved and obtained more intelligently. The data analysis agent receives the data provided by the data collection agent, analyzes whether it is relevant to the proposition, and filters out irrelevant information; because the results of network retrieval may be mixed with a large amount of invalid information such as advertisements, irrelevant information will interfere with the operation of the subsequent evidence accumulation scoring system, and this data analysis agent can ensure the relevance of the data. The falsification execution agent, according to the path planned by the falsification strategy planner, uses the data screened by the data analysis agent as data supplementation, and attempts to falsify each sub-null hypothesis in parallel, lists the key evidence item by item, and gives a guiding conclusion. In addition, there is also an agent coordination mechanism in this falsification system, which is used to design the communication protocol and cooperation mechanism between agents to ensure the seamless transfer of data and results and improve the overall execution efficiency.

[0059] Step S14: Determine the fact verification result corresponding to the information to be processed based on the evidence collection result, the hypothesis falsification result, and a preset multi-dimensional hypothesis verification quantification rule.

[0060] In this embodiment, after the falsification is performed, relevant processing will be carried out using an evidence accumulation scoring system. This scoring system is responsible for integrating the hypothesis falsification results of a preset multi-agent falsification system, calculating and updating the verification progress, and finally generating a multi-dimensional hypothesis verification score (MHVS). This scoring system adopts a scientific scoring formula and interpretation standard, and can realize the quantitative expression of verification conclusions and the measurement of uncertainty. Specifically, in this embodiment, first, source analysis, quality analysis, and relevance analysis are performed on each piece of evidence in the evidence collection result to determine the second weight corresponding to each piece of evidence based on the evidence analysis result; falsification degree analysis is performed on each null hypothesis based on the hypothesis falsification result to obtain the falsification degree analysis result corresponding to each null hypothesis; the corresponding multi-dimensional hypothesis verification score is determined based on the alternative explanation adjustment factor, the first weight and the falsification degree analysis result corresponding to each null hypothesis, and the second weight corresponding to each piece of evidence. The relevant steps for scoring are as follows:

[0061] (1) Evidence weight (i.e., the second weight) calculation: Assign weights to each piece of evidence according to factors such as evidence source, quality, and relevance;

[0062] (2) Null hypothesis falsification degree evaluation: Based on the cumulative evidence, evaluate the degree (0 - 1) to which each null hypothesis is falsified;

[0063] (3) Multi-dimensional hypothesis verification score (MHVS) calculation: Calculate the comprehensive score using the following formula:

[0064] ;

[0065] In the formula, Wi represents the weight (1 - 10) of the i-th null hypothesis; Fi represents the falsification degree (0 - 1) of the i-th null hypothesis; Ri represents the reliability coefficient (0.5 - 1) of the i-th piece of evidence; C represents the consistency multiplier (0.8 - 1.2), which is used to evaluate the consistency between evidence; A represents the alternative explanation adjustment factor (0.7 - 1), which is used to reflect possible alternative explanations. Among them, Wi is the value defined by the falsification strategy planner, and the rest are the values assigned after the analysis of the evidence accumulation scoring system.

[0066] (4) Conclusion interpretation standard:

[0067] The higher the MHVS score, the more likely it is to falsify the null hypothesis, and the more likely it is to prove the original main hypothesis to be true. MHVS > 0.8: The main hypothesis is "highly likely" to be true; 0.6 < MHVS ≤ 0.8: The main hypothesis is "very likely" to be true; 0.4 < MHVS ≤ 0.6: There is insufficient evidence to make a judgment; 0.2 < MHVS ≤ 0.4: The main hypothesis is "very likely" to be false; MHVS ≤ 0.2: The main hypothesis is "highly likely" to be false.

[0068] (5)Report generation: Based on the factual verification results obtained from the analysis, generate a structured verification report, including conclusions, key evidence, the basis for scoring each score, uncertainty analysis, and follow-up suggestions. That is, based on the factual verification results and the evidence collection results, perform uncertainty analysis, extract key evidence, and generate verification conclusions to determine the factual verification report corresponding to the information to be processed.

[0069] In summary, the factual verification system provided in this embodiment has the following technical effects:

[0070] 1) Achieved a high degree of automation in the entire process of hypothesis verification, significantly reducing manual intervention and subjective bias, and improving verification efficiency and consistency;

[0071] 2) Through the multi-dimensional null hypothesis and multi-path falsification strategy, improved the systematicness and comprehensiveness of verification, and enhanced the reliability of conclusions;

[0072] 3) Adopted a multi-agent collaboration mode to achieve distributed parallel processing, significantly improving verification efficiency and resource utilization;

[0073] 4) Through the evidence accumulation scoring system, achieved a quantitative expression of verification conclusions and uncertainty measurement, providing a more scientific basis for decision-making;

[0074] 5) Provided efficient and objective factual verification for the content generated by large models, reducing the risk of misinformation dissemination.

[0075] It should be understood that the factual verification system provided in this embodiment can be widely applied to multiple fields such as scientific research, business decision-making, legal evidence analysis, media statement verification, historical event analysis, and artificial intelligence content evaluation.

[0076] It can be seen that in this application, through the fact verification system constructed based on the large model and the preset multi-agent falsification system, the large model is first used to analyze the information to be processed to determine the hypothesis to be verified, and then the large model is used to generate multi-dimensional null hypotheses for the hypothesis to be verified, and the corresponding target falsification strategies are determined using the determined multiple null hypotheses. After that, evidence is collected through the preset multi-agent falsification system, and each of the null hypotheses is falsified using the evidence collection result and the target falsification strategy. Then, the fact verification result is determined using the evidence collection result, the hypothesis falsification result, and the preset multi-dimensional hypothesis verification quantification rule. In this way, the automation of fact verification can be effectively achieved, thereby greatly reducing the manual dependence and the resulting subjective biases, and improving the efficiency, systematicness, comprehensiveness, reliability, and adaptability of fact verification.

[0077] The following specifically describes the technical solutions of the embodiments of this application in combination with the fact verification application examples in specific fields.

[0078] To fully demonstrate the practicality and effectiveness of the fact verification system of the present invention, the following provides two specific application examples, and the data therein is only for illustrative purposes.

[0079] Among them, in the first application example, the fact verification method in this application can be applied to the fact verification of relevant information in the medical field. That is, the information to be processed in the fact verification method of this application can be the medical information to be processed, the corresponding large model can be a medical large model, and the agents in the preset multi-agent falsification system can be used to collect and analyze medical-related data.

[0080] Example 1: Scientific hypothesis verification - "BRCA1 gene mutation increases the risk of breast cancer".

[0081] Input stage: The fact verification system receives the information to be processed "BRCA1 gene mutation increases the risk of breast cancer".

[0082] Hypothesis understanding stage: The hypothesis understanding engine analyzes this hypothesis and extracts the core proposition: There is a causal relationship between the mutation of the BRCA1 gene and the increased risk of breast cancer.

[0083] Strategy planning stage: The following null hypotheses are generated based on the falsification strategy planner:

[0084] 1), Null hypothesis 1: There is no statistical correlation between BRCA1 mutation and breast cancer incidence (weight: 9);

[0085] 2), Null hypothesis 2: The association between BRCA1 mutation and breast cancer risk can be fully explained by other factors (weight: 7);

[0086] 3), Null hypothesis 3: The temporal relationship between BRCA1 mutation and breast cancer occurrence does not conform to causal logic (weight: 6);

[0087] 4) Plan the refutation paths for each null hypothesis, including obtaining data from medical literature databases, clinical research databases, and online searches for reliable sources.

[0088] Execute the refutation phase in parallel: The data collection agent obtains relevant data from clinical trial databases, medical literature databases, and authoritative media respectively; the data analysis agent filters the data and retains data such as highly relevant research results, clinical data, and analysis articles; the refutation implementation agent analyzes the data and attempts to refute each null hypothesis:

[0089] 1) Analysis of null hypothesis 1: It is found that multiple large cohort studies show that the breast cancer risk of BRCA1 mutation carriers is 5-7 times that of the general population, with a p-value < 0.0001;

[0090] 2) Analysis of null hypothesis 2: After controlling variables such as age, race, and lifestyle, the correlation between BRCA1 mutation and breast cancer risk remains significant;

[0091] 3) Analysis of null hypothesis 3: Longitudinal studies show that BRCA1 mutation precedes the occurrence of breast cancer and follows a reasonable biological mechanism.

[0092] Final result synthesis phase:

[0093] 1) Calculate the degree of refutation of each null hypothesis: Null hypothesis 1: F1 = 0.95 (highly refuted); Null hypothesis 2: F2 = 0.85 (strongly refuted); Null hypothesis 3: F3 = 0.90 (strongly refuted).

[0094] 2) Evaluate the data reliability: R1 = 0.95 (high reliability); R2 = 0.85 (relatively high reliability); R3 = 0.80 (relatively high reliability).

[0095] 3) Analyze and determine the consistency multiplier C = 1.1;

[0096] 4) Analyze and determine the alternative explanation adjustment factor A = 0.95 (there are other environmental factors);

[0097] 5) Calculate MHVS: (9×0.95×0.95 + 7×0.85×0.85 + 6×0.90×0.80) / (9 + 7 + 6) × 1.1 × 0.95 = 0.86.

[0098] Output phase: Generate a verification report with the conclusion that "the main hypothesis is highly likely to be true (MHVS = 0.86)". The final report may include a summary of key evidence, the refutation status and weight basis of each null hypothesis, and indicate that this conclusion may vary among different ethnic groups.

[0099] Among them, in the second application example, the fact verification method in the present application can be applied to the fact verification of relevant information in the automotive field. That is to say, the information to be processed in the fact verification method of the present application can be the information related to the automotive to be processed. Correspondingly, the large model can be the large model in the automotive industry, and the agents in the preset multi-agent falsification system can be used to collect and analyze automotive-related data.

[0100] Example 2: Media Statement Verification - "New energy vehicles are more environmentally friendly than traditional fuel vehicles."

[0101] Input stage: The fact verification system receives the information to be processed "New energy vehicles are more environmentally friendly than traditional fuel vehicles."

[0102] Hypothesis understanding stage: The hypothesis understanding engine analyzes the hypothesis and determines that the environmental friendliness comparison to be verified involves the life cycle environmental impact, rather than being limited to the use stage. The rewritten hypothesis is: New energy vehicles are more environmentally friendly than traditional fuel vehicles in terms of the whole life cycle.

[0103] Strategy planning stage: The falsification strategy planner generates the following null hypotheses:

[0104] 1), Null hypothesis 1: The carbon emissions of new energy vehicles in the whole life cycle are not lower than those of traditional fuel vehicles (weight: 8);

[0105] 2), Null hypothesis 2: The environmental burden in the manufacturing process of new energy vehicles offsets the environmental advantages in the use stage (weight: 7);

[0106] 3), Null hypothesis 3: Considering the power source, the operating emissions of new energy vehicles are not better than those of traditional fuel vehicles (weight: 6);

[0107] 4), Null hypothesis 4: The harm caused by the battery disposal of new energy vehicles to the environment is greater than that of traditional fuel vehicles (weight: 5);

[0108] 5), Plan the falsification path, including retrieving environmental science research data, industry report data, and searching for authoritative information sources online.

[0109] Parallel execution of the falsification stage: The data collection agent obtains data from scientific journals, environmental protection agency reports, and industry white papers; the data analysis agent filters and extracts scientifically representative research results, and discriminates and eliminates biased data published by interested parties; the falsification implementation agent analyzes each null hypothesis:

[0110] 1), Analysis of null hypothesis 1: Multiple authoritative life cycle assessments show that new energy vehicles can reduce carbon emissions by an average of 30 - 60%, but there are regional differences;

[0111] 2), Analysis of null hypothesis 2: Although the energy consumption in the manufacturing stage of new energy vehicles is relatively high, the total energy consumption in the whole life cycle is still lower than that of fuel vehicles;

[0112] 3). Analysis of null hypothesis 3: Even in the power grid areas dominated by coal, the overall carbon emissions of electric vehicles are still slightly lower than or equal to those of highly efficient fuel vehicles.

[0113] 4). Analysis of null hypothesis 4: Although significant progress has been made in battery recycling technology, there are indeed environmental concerns in the extraction and recycling of battery materials.

[0114] Final result synthesis stage:

[0115] 1). Calculate the degree of falsification of each null hypothesis: Null hypothesis 1: F1 = 0.75 (strong falsification); Null hypothesis 2: F2 = 0.70 (moderate falsification); Null hypothesis 3: F3 = 0.60 (moderate falsification); Null hypothesis 4: F4 = 0.40 (weak falsification).

[0116] 2). Evaluate the data reliability: R1 = 0.85 (high reliability); R2 = 0.80 (high reliability); R3 = 0.75 (medium reliability); R4 = 0.70 (medium reliability).

[0117] 3). Analyze and determine the consistency multiplier C = 0.9 (there is some controversy).

[0118] 4). Analyze and determine the alternative explanation adjustment factor A = 0.85 (there are other environmental protection factors).

[0119] 5). Calculate MHVS: (8 × 0.75 × 0.85 + 7 × 0.70 × 0.80 + 6 × 0.60 × 0.75 + 5 × 0.40 × 0.70) / (8 + 7 + 6 + 5) × 0.9 × 0.85 = 0.52.

[0120] Output stage: Generate a verification report with the conclusion of "Insufficient evidence to make a clear judgment (MHVS = 0.52)".

[0121] See Figure 2 As shown, the embodiment of the present application also correspondingly discloses a fact verification device, which is applied to a fact verification system constructed based on a large model and a preset multi-agent falsification system, and includes:

[0122] A hypothesis determination module 11, configured to parse the received information to be processed based on the large model, and determine a hypothesis to be verified corresponding to the information to be processed by using the obtained information parsing result;

[0123] A falsification strategy determination module 12, configured to generate multi-dimensional null hypotheses based on the large model and the hypothesis to be verified, and determine a target falsification strategy by using the determined multiple null hypotheses and data resource information;

[0124] The falsification result determination module 13 is configured to collect evidence through the preset multi-agent falsification system and the multiple null hypotheses, and use the evidence collection result and the target falsification strategy to falsify each null hypothesis to obtain the hypothesis falsification result;

[0125] The verification result determination module 14 is configured to determine the fact verification result corresponding to the information to be processed based on the evidence collection result, the hypothesis falsification result, and the preset multi-dimensional hypothesis verification quantization rule.

[0126] Among them, for the more specific working processes of the above-mentioned various modules, reference can be made to the corresponding content disclosed in the foregoing embodiments, and details will not be elaborated herein.

[0127] Thus, in this application, through the fact verification system constructed based on the large model and the preset multi-agent falsification system, the large model is first used to parse the information to be processed to determine the hypothesis to be verified, and then the large model is used to generate multi-dimensional null hypotheses for the hypothesis to be verified, and the corresponding target falsification strategy is determined using the determined multiple null hypotheses. Then, evidence is collected through the preset multi-agent falsification system, and each null hypothesis is falsified using the evidence collection result and the target falsification strategy. Then, the fact verification result is determined using the evidence collection result, the hypothesis falsification result, and the preset multi-dimensional hypothesis verification quantization rule. In this way, the automation of fact verification can be effectively realized, thereby greatly reducing the manual dependence and the resulting subjective biases, and improving the efficiency, systematicness, comprehensiveness, reliability, and adaptability of fact verification.

[0128] In some specific embodiments, the hypothesis determination module 11 may specifically be configured to: parse the received information to be processed based on the large model to obtain an information parsing result; trigger a hypothesis extraction operation and a meaningless word removal operation based on the information parsing result to determine the hypothesis to be verified corresponding to the information to be processed.

[0129] In some specific embodiments, the fact verification device may specifically further be configured to: when it is determined based on the information parsing result that a preset hypothesis splitting condition is satisfied, split the hypothesis to be verified to determine multiple sub-hypotheses to be verified, so as to generate multi-dimensional null hypotheses based on the large model and each sub-hypothesis to be verified respectively.

[0130] In some specific embodiments, the falsification strategy determination module 12 may specifically be configured to: generate multiple null hypotheses in multiple dimensions based on the large model and the characteristic information of the hypothesis to be verified, so as to obtain multiple null hypotheses that form a complementary relationship; determine the first weight corresponding to each null hypothesis based on the analysis result by analyzing the importance and verification value of each null hypothesis for the hypothesis to be verified; determine the target falsification strategy based on the characteristic information, data resource information, and intelligent agent resource information of each null hypothesis.

[0131] In some specific embodiments, the verification result determination module 14 may specifically be configured to: determine the second weight corresponding to each piece of evidence based on the evidence analysis result by performing source analysis, quality analysis, and relevance analysis on each piece of evidence in the evidence collection result; perform falsification degree analysis on each null hypothesis based on the hypothesis falsification result to obtain the falsification degree analysis result corresponding to each null hypothesis; determine the corresponding multi-dimensional hypothesis verification score based on the alternative explanation adjustment factor, the first weight and the falsification degree analysis result corresponding to each null hypothesis, and the second weight corresponding to each piece of evidence.

[0132] In some specific embodiments, the falsification result determination module 13 may specifically be configured to: collect evidence for verifying the multiple null hypotheses from a selected plurality of data sources through the data collection intelligent agent in the preset multi-intelligent agent falsification system to obtain an evidence collection result; perform a data cleaning operation and a format conversion operation on the evidence collection result based on the data collection intelligent agent to obtain the processed evidence collection result; perform a relevance analysis on the processed evidence collection result through the data analysis intelligent agent in the preset multi-intelligent agent falsification system to filter the processed evidence collection result based on the relevance analysis result to obtain the filtered evidence collection result; falsify each null hypothesis through the falsification execution intelligent agent in the preset multi-intelligent agent falsification system, the filtered evidence collection result, and the target falsification strategy to obtain a hypothesis falsification result.

[0133] In some specific embodiments, the fact verification device may specifically further be configured to: perform uncertainty analysis, key evidence extraction, and verification conclusion generation based on the fact verification result and the evidence collection result to determine a fact verification report corresponding to the information to be processed.

[0134] Furthermore, an embodiment of the present application also discloses an electronic device Figure 3 It is a structural diagram of an electronic device 20 shown according to an exemplary embodiment, and the content in the figure cannot be regarded as any limitation on the scope of use of the present application.

[0135] Figure 3 Schematic diagram of the structure of an electronic device 20 provided by an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. Among them, the memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the fact verification method disclosed in any of the foregoing embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0136] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows is any communication protocol applicable to the technical solution of the present application, and specific limitations are not imposed here; the input / output interface 25 is used to obtain external input data or output data to the outside, and its specific interface type can be selected according to specific application needs, and specific limitations are not imposed here.

[0137] In addition, as a carrier for resource storage, the memory 22 may be a read-only memory, a random access memory, a disk, or an optical disc, etc., and the resources stored thereon may include an operating system 221, a computer program 222, etc., and the storage method may be temporary storage or permanent storage.

[0138] Among them, the operating system 221 is used to manage and control each hardware device and the computer program 222 on the electronic device 20, and it may be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the fact verification method executed by the electronic device 20 disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs that can be used to complete other specific tasks.

[0139] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the fact verification method disclosed above is implemented. For the specific steps of this method, reference may be made to the corresponding content disclosed in the foregoing embodiments, and details will not be repeated here.

[0140] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0141] Those skilled in the art may further realize that the units and algorithm steps of each example described in connection with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described according to their functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0142] The steps of the methods or algorithms described in connection with the embodiments disclosed herein can be directly implemented by hardware, software modules executed by a processor, or a combination of the two. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0143] Finally, it should also be noted that in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0144] The technical solutions provided in this application have been introduced in detail above. Specific examples are used herein to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A fact verification method, characterized in that: Applied to fact verification systems built on large models and preset multi-agent falsification systems, including: Parsing the received information to be processed based on the large model, and determining the hypothesis to be verified corresponding to the information to be processed using the obtained information parsing result; Generate multi-dimensional null hypotheses based on the large model and the hypothesis to be verified, and determine a target falsification strategy using the determined multiple null hypotheses and data resource information; Collecting evidence through the preset multi-agent falsification system and the multiple null hypotheses, and falsifying each of the null hypotheses using the evidence collection results and the target falsification strategy to obtain a hypothesis falsification result; Determine a fact verification result corresponding to the information to be processed based on the evidence collection result, the hypothesis falsification result and a preset multi-dimensional hypothesis verification quantitative rule; The step of parsing the received information to be processed based on the large model and determining the hypothesis to be verified corresponding to the information to be processed using the obtained information parsing result includes: Parsing the received information to be processed based on the large model to obtain an information parsing result; triggering a hypothesis extraction operation and a meaningless word removal operation based on the information parsing result to determine a hypothesis to be verified corresponding to the information to be processed; The generating of multi-dimensional null hypotheses based on the large model and the hypothesis to be verified, and determining a target falsification strategy using the determined multiple null hypotheses and data resource information, includes: Based on the large model and the characteristic information of the hypothesis to be verified, multi-dimensional null hypothesis generation is performed to obtain multiple null hypotheses constituting complementary relationships; By analyzing the importance and verification value of each null hypothesis to the hypothesis to be verified, a first weight corresponding to each null hypothesis is determined based on the analysis result; Determining a target falsification strategy based on the characteristic information, data resource information, and agent resource information of each of the null hypotheses; The determining of the fact verification result corresponding to the information to be processed based on the evidence collection result, the hypothesis falsification result and the preset multi-dimensional hypothesis verification quantification rule includes: By performing source analysis, quality analysis and relevance analysis on each piece of evidence in the evidence collection result, a second weight corresponding to each piece of evidence is determined based on the evidence analysis result; Based on the hypothesis falsification results, a falsification degree analysis is performed on each of the null hypotheses to obtain a falsification degree analysis result corresponding to each of the null hypotheses; Determine a corresponding multidimensional hypothesis verification score based on an alternative explanation adjustment factor, the first weights respectively corresponding to each of the null hypotheses, the falsification degree analysis results, and the second weights respectively corresponding to each of the pieces of evidence; The collecting evidence by using the preset multi-agent falsification system and the multiple null hypotheses, and falsifying each null hypothesis by using the evidence collection results and the target falsification strategy, includes: The data collection agent in the preset multi-agent falsification system collects evidences for verifying the multiple null hypotheses from the selected multiple data sources to obtain evidence collection results; Based on the data collection agent, a data cleaning operation and a format conversion operation are performed on the evidence collection result to obtain the processed evidence collection result; Performing correlation analysis on the data analysis agent in the preset multi-agent falsification system and the processed evidence collection results, filtering the processed evidence collection results based on the correlation analysis results to obtain the filtered evidence collection results; Each of the null hypotheses is falsified by using the falsification execution agent in the preset multi-agent falsification system, the filtered evidence collection results and the target falsification strategy to obtain a hypothesis falsification result.

2. The fact verification method according to claim 1, characterized in that: After determining the hypothesis to be verified corresponding to the information to be processed by using the obtained information analysis result, the method further includes: When it is determined based on the information analysis result that the preset hypothesis splitting condition is met, a plurality of sub-hypotheses to be verified are determined by splitting the hypothesis to be verified, so as to perform multi-dimensional null hypothesis generation based on the large model and each of the sub-hypotheses to be verified.

3. The fact verification method according to claim 1, characterized in that: After determining the fact verification result corresponding to the information to be processed based on the evidence collection result, the hypothesis falsification result and the preset multi-dimensional hypothesis verification quantification rule, the method further includes: Uncertainty analysis, key evidence extraction and verification conclusion generation are performed based on the fact verification results and the evidence collection results to determine a fact verification report corresponding to the information to be processed.

4. A fact verification device, characterized in that: Applied to fact verification systems built on large models and preset multi-agent falsification systems, including: A hypothesis determination module, used to parse the received information to be processed based on the large model, and determine the hypothesis to be verified corresponding to the information to be processed using the obtained information parsing result; A falsification strategy determination module, used to generate multi-dimensional null hypotheses based on the large model and the hypothesis to be verified, and determine a target falsification strategy using the determined multiple null hypotheses and data resource information; A falsification result determination module, used to collect evidence through the preset multi-agent falsification system and the multiple null hypotheses, and falsify each of the null hypotheses using the evidence collection results and the target falsification strategy to obtain a hypothesis falsification result; A verification result determination module, used to determine a fact verification result corresponding to the information to be processed based on the evidence collection result, the hypothesis falsification result and a preset multi-dimensional hypothesis verification quantification rule; The hypothesis determination module is used to: parse the received information to be processed based on the large model to obtain an information parsing result; trigger a hypothesis extraction operation and a meaningless word removal operation based on the information parsing result to determine a hypothesis to be verified corresponding to the information to be processed; The falsification strategy determination module is used to: generate multi-dimensional null hypotheses based on the large model and the characteristic information of the hypothesis to be verified, so as to obtain multiple null hypotheses that constitute a complementary relationship; analyze the importance and verification value of each null hypothesis to the hypothesis to be verified, so as to determine the first weight corresponding to each null hypothesis based on the analysis result; determine the target falsification strategy based on the characteristic information, data resource information and agent resource information of each null hypothesis; The verification result determination module is used to: perform source analysis, quality analysis and correlation analysis on each piece of evidence in the evidence collection result, so as to determine the second weight corresponding to each piece of evidence based on the evidence analysis result; perform falsification degree analysis on each null hypothesis based on the hypothesis falsification result, so as to obtain the falsification degree analysis result corresponding to each null hypothesis; determine the corresponding multidimensional hypothesis verification score based on the alternative explanation adjustment factor, the first weight corresponding to each null hypothesis and the falsification degree analysis result, and the second weight corresponding to each piece of evidence; The falsification result determination module is used to: collect evidence to verify the multiple null hypotheses from the selected multiple data sources through the data collection agent in the preset multi-agent falsification system to obtain evidence collection results; perform data cleaning operations and format conversion operations on the evidence collection results based on the data collection agent to obtain the processed evidence collection results; perform correlation analysis through the data analysis agent in the preset multi-agent falsification system and the processed evidence collection results to filter the processed evidence collection results based on the correlation analysis results to obtain the filtered evidence collection results; falsify each null hypothesis through the falsification execution agent in the preset multi-agent falsification system, the filtered evidence collection results and the target falsification strategy to obtain a hypothesis falsification result.

5. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the fact verification method as claimed in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements the fact verification method as claimed in any one of claims 1 to 3.

Images