A competitive intelligence generation and verification method based on evidence chain constraints

Abstract

The application discloses a competitive intelligence generation and verification method based on evidence chain constraint and belongs to the technical field of artificial intelligence and enterprise information processing. The method comprises the following steps: S1, collecting multi-source competitive intelligence data and cleaning and structurally storing the data; S2, cutting the data to generate evidence units and marking the sources, time, theme and credibility; S3, performing semantic matching according to user questions to form an evidence candidate set; S4, constructing an evidence chain structure comprising a conclusion, evidence, source, time and credibility; S5, generating an initial competitive intelligence conclusion based on the evidence chain; S6, verifying the conclusion in terms of source consistency, time consistency, logical integrity and evidence sufficiency; S7, performing weight reduction, supplementary retrieval and conclusion reconstruction on conflicts or evidence gaps; and S8, outputting a competitive intelligence result comprising a conclusion, evidence items and credibility. The application improves the traceability and consistency of the intelligence result and reduces the risk of information conflicts and model hallucination.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and enterprise information processing technology, specifically to a method for generating and verifying competitive intelligence based on evidence chain constraints, applicable to competitive intelligence service scenarios for micro and small enterprises. Background Technology

[0002] Competitive intelligence generation is a crucial step for enterprises in summarizing, analyzing, and outputting information related to competitor activities, market trends, policy changes, industry sentiment, and operational risks. With the continuous growth of internet information volume, enterprises, especially small and micro-enterprises, face increasingly diverse information sources and faster information updates. How to quickly generate reliable competitive intelligence conclusions from massive, multi-source, and heterogeneous information has become a critical issue that urgently needs to be addressed in the field of enterprise information services.

[0003] For micro and small enterprises, limited staff, insufficient intelligence analysis capabilities, and restricted technical resources often make it difficult to rely on traditional manual methods to continuously conduct high-quality competitive intelligence work. Therefore, utilizing artificial intelligence methods to automatically process and generate intelligence from multi-source data such as news, announcements, bidding information, publicly available corporate data, social media, and industry reports has become an important direction for improving the competitive intelligence service capabilities of micro and small enterprises.

[0004] Most existing competitive intelligence generation methods adopt the approach of "direct generation after retrieval" or "direct output of conclusions after inputting a question," that is, first obtaining some relevant text through a retrieval module, and then directly outputting the analysis results by a generation model. Although this type of method improves information processing speed to some extent, it still has the following problems:

[0005] The relationship between the generated results and the original evidence is unclear, and the output conclusions lack traceable evidence;

[0006] There may be time conflicts, source conflicts, or content contradictions among multi-source data, and existing methods lack a unified verification mechanism;

[0007] Conclusions are often generated based on a single inference, lacking constraints on the sufficiency of evidence and logical consistency, and are prone to factual errors or model illusions.

[0008] When gaps in evidence or conflicts of information occur, existing methods struggle to automatically perform error correction operations such as supplementing evidence, reducing weights, alternative searches, or reconstructing conclusions.

[0009] Therefore, there is an urgent need for a technical method that can structure and organize the original evidence during the competitive intelligence generation process and perform verification and error correction on the generated results, so as to improve the credibility, interpretability and traceability of competitive intelligence results. Summary of the Invention

[0010] This invention discloses a method for generating and verifying competitive intelligence based on evidence chain constraints. The method collects, segments, labels, and structures multi-source competitive intelligence data to construct an evidence chain structure corresponding to the target problem. Under the constraints of this evidence chain, an initial competitive intelligence conclusion is generated, and further, source consistency verification, time consistency verification, logical integrity verification, and evidence sufficiency verification are performed on the initial conclusion. When the verification results indicate the existence of evidence gaps or information conflicts, an error correction process is triggered, low-credibility evidence is downgraded, missing evidence is supplemented, and an updated competitive intelligence result is formed through alternative retrieval or conclusion reconstruction. This method can output traceable and verifiable competitive intelligence results, effectively reducing the illusion risk of the generative model in competitive intelligence applications and improving the credibility and interpretability of the conclusions.

[0011] To solve the above problems, the present invention adopts the following technical solution:

[0012] 1. A method for generating and verifying competitive intelligence based on evidence chain constraints, the method comprising the following steps:

[0013] S1. Acquisition of multi-source competitive intelligence data.

[0014] Collect multi-source data relevant to the target competitive intelligence mission and convert the multi-source data into processable raw evidence data. The multi-source data includes at least one or more of the following: news data, announcement data, bidding information, publicly available corporate data, social media data, and industry report data.

[0015] S2. Evidence unit segmentation and structured annotation.

[0016] The raw evidence data obtained in step S1 is segmented into multiple evidence units, and structured annotation processing is performed on each evidence unit. Each evidence unit includes at least the evidence text content, source identifier, publication time information, topic identifier, and initial credibility value.

[0017] S3. Construction of the candidate evidence set.

[0018] Based on the competitive intelligence question to be answered, multiple evidence units are matched and filtered to obtain a candidate set of evidence related to the target question. The filtering criteria include at least one or more of the following: topic relevance, time range matching degree, target object matching degree, and source validity.

[0019] S4. Construction of the chain of evidence.

[0020] Based on the candidate evidence set obtained in step S3, an evidence chain structure corresponding to the target question is constructed. The evidence chain structure includes at least a conclusion node, an evidence node, a source node, a time node, and a credibility node; in one embodiment, the evidence chain adopts a structured representation of "conclusion-evidence-source-time-credibility".

[0021] S5. Initial competitive intelligence conclusions are generated.

[0022] The evidence chain structure constructed in step S4 is input into the generation model to generate an initial competitive intelligence conclusion. The initial competitive intelligence conclusion includes at least the conclusion text, corresponding evidence citation identifiers, and risk labels.

[0023] S6. Verification of initial conclusions.

[0024] The verification module is invoked to verify the initial competitive intelligence conclusions generated in step S5, and the verification results are obtained. The verification process includes at least one or more of the following: source consistency verification, time consistency verification, logical integrity verification, and evidence sufficiency verification.

[0025] S7. Conflict resolution and error correction.

[0026] When the verification results in step S6 indicate that the initial competitive intelligence conclusion has evidentiary gaps, source conflicts, temporal conflicts, or logical inconsistencies, error correction processing is performed. This error correction processing includes at least de-weighting low-confidence evidence, supplementing missing evidence, performing alternative searches on candidate evidence, and regenerating the conclusion based on the updated evidence chain. Preferably, the conflict resolution employs a combination of time-priority rules, authoritative source priority rules, and cross-support priority rules.

[0027] S8, Output of competitive intelligence results.

[0028] Output verified and corrected competitive intelligence results. These results include at least the conclusion text, evidence entries, source link identifiers, risk tags, and credibility scores.

[0029] The beneficial effects achieved by this invention are:

[0030] (1) Improve the traceability of results. By segmenting multi-source data into structured evidence units and constructing evidence chains, a clear connection is established between competitive intelligence conclusions and their supporting evidence, thereby improving the ability to trace the source of results.

[0031] (2) Improve the ability to judge the consistency of multi-source information. By performing source consistency verification, time consistency verification, logical integrity verification and evidence sufficiency verification on the initial conclusion, conflicts and gaps between information from different sources can be identified more accurately.

[0032] (3) Reduce the risk of erroneous intelligence. When evidence gaps or information conflicts are detected, the system can automatically perform downweighting, supplementary retrieval, alternative retrieval, or regenerate conclusions, effectively reducing the risk of erroneous intelligence to corporate decision-making.

[0033] (4) Reduce the risk of illusion in generative models. This invention changes the competitive intelligence generation process from "direct answering" to a processing mode of "building a chain of evidence first and then generating a conclusion", which can reduce the risk of illusion in direct generation scenarios.

[0034] (5) Improve the interpretability and applicability of the results. The output results are structured and credible results containing evidence entries, source link identifiers, risk labels and credibility scores, which are more suitable for enterprise report generation, intelligence services and risk warning scenarios. Attached Figure Description

[0035] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0036] Figure 1 This is a schematic diagram of the overall process of a competitive intelligence generation and verification method based on evidence chain constraints;

[0037] Figure 2 This is a schematic diagram of the evidence unit segmentation and structured annotation process;

[0038] Figure 3 This is a schematic diagram of the chain of evidence structure;

[0039] Figure 4 This is a schematic diagram of the initial conclusion verification and conflict resolution process;

[0040] Figure 5 This is a schematic diagram of the output structure of competitive intelligence results. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the protection scope of the present invention.

[0042] like Figure 1As shown, the present invention proposes a competitive intelligence generation and verification method based on evidence chain constraints, which mainly includes the following steps: acquiring multi-source competitive intelligence data, segmenting and structuring evidence units, constructing evidence candidate sets, constructing evidence chain structures, generating initial competitive intelligence conclusions, verifying initial conclusions, resolving conflicts and correcting errors, and outputting competitive intelligence results.

[0043] Regarding step S1, which involves acquiring multi-source competitive intelligence data, step S1 specifically includes:

[0044] S11. Obtain multi-source competitive intelligence data from news websites, announcement platforms, bidding systems, enterprise public data platforms, social media platforms, and industry report databases through network collection interfaces or web crawlers;

[0045] S12. Clean the collected raw data, including removing HTML tags, invalid characters and duplicate content, and standardizing the time format;

[0046] S13. Classify and store the cleaned data according to its source type, and add data source identifiers.

[0047] Regarding step S2, which involves the segmentation and structured annotation of evidence units, step S2 specifically includes:

[0048] S21. The text data obtained in step S1 is processed by sentence segmentation or segmentation to divide the long text into multiple text segments.

[0049] S22. Extract keywords or identify topics for each text segment;

[0050] S23. Add structured annotation information to each piece of evidence, including source identifier, publication time, and topic tags;

[0051] S24. Calculate the initial credibility value of the evidence unit.

[0052] In one embodiment, the credibility value of the evidence unit is calculated as follows:

[0053]

[0054] in:

[0055] S indicates the authority of the source;

[0056] T represents time freshness;

[0057] R represents cross-support;

[0058] H represents historical stability;

[0059] α, β, γ, and δ are weighting coefficients.

[0060] Regarding step S3, step S3 specifically includes:

[0061] S31. Extract keywords from the competitive intelligence questions input by the user;

[0062] S32. Calculate the semantic similarity between the evidence unit and the question;

[0063] S33. Filter evidence units with similarity higher than a preset threshold;

[0064] S34. Construct a candidate set of evidence.

[0065] In one implementation, evidence screening employs a cosine similarity calculation method:

[0066]

[0067] Regarding step S4, the system constructs an evidence chain structure based on the evidence candidate set obtained in step S3. Step S4 specifically includes:

[0068] S41. Establish target conclusion nodes;

[0069] S42. Connect the candidate evidence unit as an evidence node;

[0070] S43. Establish the mapping relationship between evidence nodes and source nodes;

[0071] S44. Establish time nodes and record evidence time information;

[0072] S45. Construct a complete chain of evidence, in which the chain of evidence adopts a structured representation of "conclusion - evidence - source - time - credibility".

[0073] Regarding the generation of the initial competitive intelligence conclusion in step S5, step S5 specifically includes:

[0074] S51. Input the evidence chain structure constructed in step S4 into the generation model;

[0075] S52. Generate initial competitive intelligence conclusions based on the chain of evidence;

[0076] S53. The conclusion includes the conclusion text, evidence citation markers, and risk labels.

[0077] For step S6, the system calls the verification module to verify the initial conclusion generated in step S5. Step S6 specifically includes:

[0078] S61. Are there any conflicts in the sources of evidence cited in the test results?

[0079] S62. Are there any temporal inconsistencies among the test evidence?

[0080] S63. Whether the number of evidence samples meets the preset minimum support threshold;

[0081] S64. Determine whether there is a logical inconsistency between the conclusion and the evidence.

[0082] Regarding step S7, conflict resolution and error correction, step S7 specifically includes:

[0083] S71. Reduce the weight of low-credibility evidence;

[0084] S72. Perform a supplementary search for missing evidence;

[0085] S73. Perform replacement processing on conflicting evidence;

[0086] S74. Regenerate competitive intelligence conclusions based on the updated chain of evidence.

[0087] Regarding step S8, the system outputs verified and corrected competitive intelligence results, which include at least the conclusion text, evidence entries, source link identifiers, risk tags, and credibility scores. Compared with existing methods, the results output by this invention are no longer single text conclusions, but rather verifiable, traceable, and interpretable structured results, making them more suitable for competitive intelligence service scenarios for micro and small enterprises.

[0088] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating and verifying competitive intelligence based on evidence chain constraints, characterized in that, Includes the following steps: S1. Collect multi-source competitive intelligence data, and clean and structure the multi-source competitive intelligence data; S2. Divide the original data into multiple evidence units and perform structured annotation on each evidence unit; S3. Based on the competitive intelligence question input by the user, perform semantic matching and filtering on the evidence units to construct a candidate evidence set; S4. Construct an evidence chain structure based on the aforementioned evidence candidate set; S5. Generate initial competitive intelligence conclusions based on the aforementioned chain of evidence structure; S6. Perform consistency and sufficiency verification on the initial competitive intelligence conclusions; S7. When the verification results indicate that there are evidence conflicts or gaps in evidence, conflict resolution and error correction shall be performed. S8. Output the verified and corrected competitive intelligence results.

2. The method according to claim 1, characterized in that, Step S1 specifically includes: S11. Obtain multi-source competitive intelligence data through network collection interfaces or web crawlers; S12. Clean the collected data, including removing invalid characters and duplicate content and standardizing the time format; S13. Classify and store the cleaned data and add source identifiers.

3. The method according to claim 1, characterized in that, Step S2 specifically includes: S21. Process the raw data by dividing it into sentences or segments; S22. Extract keywords or identify topics from text fragments; S23. Add source identifiers, publication dates, and topic tags to the evidence units; S24. Calculate the initial credibility value of the evidence unit.

4. The method according to claim 1, characterized in that, Step S3 specifically includes: S31. Extract keywords for competitive intelligence issues; S32. Calculate the semantic similarity between the evidence unit and the competitive intelligence problem; S33. Filter evidence units with similarity higher than a preset threshold; S34. Construct a candidate set of evidence.

5. The method according to claim 1, characterized in that, In step S4, the chain of evidence structure includes: Conclusion node, evidence node, source node, time node, and credibility node; The relationships between nodes are established using a structured representation of "conclusion-evidence-source-time-credibility".

6. The method according to claim 1, characterized in that, Step S5 specifically includes: S51. Input the evidence chain structure into the generation model; S52. Generate initial competitive intelligence conclusions; S53. The initial competitive intelligence conclusion includes the conclusion text, evidence citation identifiers, and risk labels.

7. The method according to claim 1, characterized in that, Step S6 specifically includes: S61. Are there any conflicts in the sources of evidence cited in the test results? S62. Are there any temporal inconsistencies among the test evidence? S63. Does the number of detected evidences meet the preset threshold? S64. Determine whether there is a logical inconsistency between the conclusion and the evidence.

8. The method according to claim 1, characterized in that, Step S7 specifically includes: S71. Reduce the weight of low-credibility evidence; S72. Perform a supplementary search for missing evidence; S73. Perform replacement processing on conflicting evidence; S74. Regenerate competitive intelligence conclusions based on the updated chain of evidence.

9. The method according to claim 1, characterized in that, The conflict resolution employs one or more of the following rules: Time-priority rule, authoritative source priority rule, and cross-support priority rule.

10. The method according to claim 1, characterized in that, The competitive intelligence results output in step S8 include: Conclusion text, evidence items, source link identifiers, risk labels, and credibility scores.

Images