Multi-source information gathering analysis method and system

Abstract

The application discloses a multi-source information gathering and analyzing method and system, and relates to the technical field of information processing and data analysis. The method comprises the following steps: determining a target enterprise and associated entities and constructing an ontology knowledge graph; constructing bilingual search keywords for the target enterprise and the associated entities, searching competitive intelligence from multiple information sources; evaluating the credibility of each information source and filtering out information with credibility lower than a predetermined threshold; using a large language model to extract relationship triples from the verified information, and filtering based on a set of named entities; generating competitive intelligence portraits and competitive situation insights based on the extraction results, and verifying and supplementing; finally generating a competitive value chain (CVC) model. The method can automatically gather competitive intelligence scattered in multiple channels into a structured and verifiable competitive value chain model, improving the credibility and analysis efficiency of the information.

Description

[0001] This invention relates to the field of intelligence processing and data analysis technology, specifically to a multi-source intelligence aggregation and analysis method and system, which can be applied to scenarios such as competitive intelligence analysis, enterprise competitor monitoring, market situation assessment, customer dynamic tracking, and decision support systems. Background Technology

[0002] Competitive intelligence analysis is an important support for enterprises to make competitive decisions. Its data sources usually include multiple heterogeneous channels such as corporate websites, news reports, industry research reports, and public bidding information. Existing competitive intelligence analysis mainly relies on manual or semi-automatic methods to collect and organize the scattered information on enterprises, competitors, markets and customers from the above channels. It has the following technical defects: (1) The sources of intelligence are complicated and of varying quality. There is a lack of a systematic source credibility assessment mechanism. Low-quality intelligence is easy to be mixed into the analysis results, and the credibility of intelligence is difficult to verify; (2) Intelligence mostly exists in unstructured text. There is a lack of unified knowledge representation and structured extraction methods, making it difficult to automatically establish the relationship between enterprises and competitors, products, customers and markets; (3) Manual aggregation is inefficient and has poor timeliness, making it difficult to meet the needs of enterprises to quickly understand the competitive situation; (4) Most existing automated intelligence aggregation solutions are limited to simple retrieval and aggregation. They lack a complete closed loop of collection, verification, structuring and reasoning, and the generated results are difficult to directly support competitive decisions.

[0003] At the theoretical level of competitive analysis, traditional Porter's value chain primarily focuses on the linear value creation process within a firm, making it difficult to directly characterize the value game relationships among multiple entities within a specific market boundary. The value chain from a competitive intelligence perspective, namely the Competitive Value Chain (CVC), refers to a dynamic value network system within a specific market boundary composed of four core entities: customers, market, competitors, and partners. It centers on value competition and acquisition, focusing on the game relationships of value creation, transmission, and distribution among these entities. Current technology lacks methods and systems capable of automatically aggregating multi-source intelligence into the aforementioned dynamic value network model. Summary of the Invention

[0004] This invention aims to overcome the shortcomings of existing technologies and provide a multi-source intelligence aggregation and analysis method and system. It realizes the automatic aggregation of multi-source competitive intelligence, verification of information source credibility, extraction of structured relationships and generation of competitive insights, forming a credible and traceable competitive value chain (CVC) model. This provides decision-making basis for enterprises to identify key control points in the value chain, assess their relative competitive position, predict changes in the landscape, and formulate differentiated competitive strategies.

[0005] The overall technical logic of this invention is as follows: it provides a unified entity and graph foundation for multi-source intelligence through ontology modeling, realizes the convergence of intelligence from multiple sources by relying on competitive intelligence collection, filters out low-credibility intelligence by combining source credibility verification, transforms unstructured intelligence into structured relational data through causal model extraction, and finally generates competitive intelligence profiles, competitive situation insights and competitive value chain models through insight generation and verification, forming a complete closed-loop intelligence processing system of "ontology modeling - intelligence collection - source verification - causal model extraction - insight generation".

[0006] Ontology modeling solution: Identify the target enterprise and entities associated with it, such as products, customers, partners, markets, and competitors; construct an ontology knowledge graph and supplement the attribute information of each entity; generate placeholders for quantitative indicators such as revenue, market share, growth rate, and number of customers, providing a unified entity and graph foundation for subsequent processing.

[0007] Competitive intelligence gathering solution: Construct bilingual (Chinese and English) search keywords for target companies and related entities, and adopt an English-first strategy; extract Chinese and English industry keywords from industry names and combine them with event-related keywords; conduct searches from multiple data sources such as internal intelligence databases, third-party intelligence aggregation services, public news sources, and corporate websites, and filter out irrelevant intelligence after relevance assessment to obtain a candidate intelligence set.

[0008] Source credibility verification scheme: Calculate credibility score by comprehensively considering factors such as the authority level of the source media, link accessibility, attribute completeness and relevance, filter out intelligence with credibility scores below a predetermined threshold, and obtain a set of verified intelligence.

[0009] Causal model extraction scheme: Use a large language model to extract relational triples such as competition, cooperation, investment, bidding, and product release. Filter the extraction results based on a preset named entity set and write the filtered relational triples into the knowledge graph. During model reconstruction, delete the original graph data and then rewrite it to achieve idempotent reconstruction.

[0010] Insight generation and verification solution: Based on the filtered relationship triples, generate enterprise profiles, market profiles, customer profiles, competitor dynamics, market dynamics, customer dynamics and competitive landscape data, and generate graph relationships such as competition, cooperation, market segmentation, and customer groups; use a second language model to verify the credibility of the generated insights, and fill in missing or low credibility fields.

[0011] The system of this invention is divided into six core modules, which work together to achieve intelligent intelligence aggregation and competitive value chain model generation throughout the entire process: The ontology modeling module is responsible for identifying target enterprises and related entities, constructing an ontology knowledge graph, and supplementing entity attribute information; the competitive intelligence collection module is responsible for constructing search keywords, retrieving competitive intelligence from multiple intelligence sources, and filtering out irrelevant intelligence; the source credibility verification module is responsible for credibility assessment and filtering of candidate intelligence; the causal model extraction module is responsible for extracting relation triples, filtering named entities, and writing knowledge graphs; the insight generation and verification module is responsible for generating, verifying, and completing competitive intelligence profiles and competitive situation insights; and the model generation module is responsible for generating a competitive value chain model by integrating the processing results of each module. Attached Figure Description

[0012] The accompanying drawings are provided to further understand the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof.

[0013] In the appendix Figure 1 The diagram illustrates the logical relationships and data flow of the overall system architecture. These are, in order: Front-end Portal 101, Application Routing Layer 102, Model Building Service 103, Queue Work Process 104, Document Database 105, Relational Database 106, Knowledge Graph Database 107, Cache Queue 108, and Ontology Fusion Intelligence Aggregation 109, forming a complete processing chain from intelligence aggregation to model generation and display.

[0014] Figure 2 A flowchart for generating the five stages of the competitive value chain model is provided, showing the complete generation process from ontology modeling 201, competitive intelligence gathering 202, source credibility verification 203, causal model extraction 204, insight generation and verification 205 to competitive value chain model output 206, and marking the task status flow and progress synchronization mechanism.

[0015] Figure 3 This is a schematic diagram of the multi-path retrieval and relevance filtering process in the competitive intelligence gathering phase. The process proceeds sequentially as follows: 301. Target company and related entity identification; 302. Bilingual (Chinese and English) keyword construction; 303. Industry keyword expansion and event combination; 304. Multi-path intelligence retrieval (internal intelligence database 3041, third-party intelligence aggregation service 3042, public news sources 3043, company website 3044); 305. External service accessibility testing; 306. Local intelligence generation as a fallback; 307. Relevance determination. If the determination is negative, the intelligence is discarded and the corresponding source credibility score is adjusted; if the determination is positive, a candidate intelligence set is output.

[0016] Figure 4This is a schematic diagram of the source credibility verification and causal model extraction process. The process involves the following steps in sequence: candidate intelligence set input 401, source credibility scoring 402, threshold filtering 403, intelligence set verification 404, relation triple extraction 405, named entity set filtering 406, knowledge graph writing 407, and finally outputting structured relation triples. Detailed Implementation

[0017] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0018] The Competitive Value Chain (CVC) model generated in this implementation refers to a dynamic value network system composed of four core entities—customers, the market, competitors, and partners—within a specific market boundary. This model focuses on value competition and acquisition, examining the game-theoretic relationships of value creation, transmission, and distribution among these entities: competitors are direct value distributors, customers are the scarce resources being contested and value arbiters, the market is the rule-maker and boundary field, and partners are advantage amplifiers and collaborators. The essence of the competitive value chain model is to use intelligence-driven dynamic analysis to identify key control points in the value chain around competitors, assess relative competitive positions, and predict changes in the market landscape, providing a basis for enterprises to formulate differentiated competitive strategies.

[0019] Figure 1 The overall system architecture of this implementation method is shown below. The modules work together to achieve intelligent intelligence aggregation and competitive value chain model generation throughout the entire process.

[0020] Front-end portal 101: Used to receive users' selections of target companies and related entities, and to display the competitive value chain model and its real-time generation progress to users.

[0021] Application routing layer 102: Used to receive front-end requests and schedule tasks, providing interfaces for model creation, reconstruction, pause, resumption, and progress streaming.

[0022] Model Building Service 103: Used to execute asynchronous in-process tasks and handle the initial model building.

[0023] Queue worker process 104: Used for message queue-based execution model reconstruction and batch scheduling tasks.

[0024] Document Database 105: Used to store competitive intelligence, corporate files, insights, and page data.

[0025] Relational database 106: Used to store model metadata, tenants, users and permission information.

[0026] Knowledge Graph Database 107: Used to store ontology and relation triples.

[0027] Cache queue 108: Used for task queuing and result caching.

[0028] Ontology Fusion Intelligence Aggregation 109: This feature aggregates multi-source intelligence from public news sources, corporate websites, and other third-party channels, and performs fusion processing on ontology to provide a unified fusion intelligence input for subsequent relationship extraction and insight generation.

[0029] As an example, the large language model used in this implementation includes a first language model for core reasoning and a second language model for insight verification and completion. The first language model and the second language model can adopt different series of large language models.

[0030] like Figure 2 As shown, this implementation organizes the generation of the competitive value chain model into five sequentially connected stages.

[0031] Ontology modeling phase 201: Extract entity names from the target company and related entities selected by the user. Related entities include at least four core entities in the competitive value chain model: customers, markets, competitors, and partners. They may also include extended entities such as products. Construct the ontology definition of the ontology knowledge graph and write it into the knowledge graph database 107. Retrieve the archive information of the target company and related entities from the enterprise archive database, and supplement the company's description, industry, official website, financing, key personnel, competitors, and other attributes. Generate placeholders for quantitative indicators such as revenue, market share, growth rate, and number of customers.

[0032] Competitive intelligence gathering phase 202: such as Figure 3As shown, relevant intelligence is retrieved for each company, competitor, market, and customer. First, the target company and related entities are identified 301. The construction of search keywords includes: constructing bilingual (Chinese and English) keywords 302, building bilingual keywords for each entity, and adopting an English-first strategy, placing English keywords before Chinese keywords; expanding industry keywords and combining events 303, extracting Chinese and English industry keywords from the competitive value chain model name, and supplementing with Chinese and English event-related keywords to form a query combining industry terms and event terms; and constructing basic queries and event combination queries for competitors. During the retrieval, multi-source intelligence retrieval is performed 304, which can obtain intelligence from multiple data sources such as internal intelligence databases 3041, third-party intelligence aggregation services 3042, public news sources 3043, and company websites 3044, and external service accessibility testing is performed 305; if external intelligence services are unavailable, an alarm is added to the generated log, and local intelligence generation is triggered as a fallback 306. Then, a relevance determination is performed 307. The relevance of the search results is determined, and low-relevance results that are not related to the target company or related entities are eliminated. If the determination is negative, the intelligence is discarded and the corresponding source credibility score is adjusted 309. If the determination is positive, a candidate intelligence set is output 308.

[0033] Source Trust Verification Phase 203: (e.g.) Figure 4 As shown, each intelligence item in the candidate intelligence set input 401 is scored for source credibility 402. Scoring factors include: assigning high credibility to authoritative media and awarding bonus points to moderately credible media; verifying the accessibility of intelligence links; checking the completeness of attributes such as intelligence title, link, publication time, and category; and verifying the relevance of the intelligence to the target company's events. After threshold filtering 403, only intelligence items with credibility scores not lower than a predetermined threshold and accessible links are retained, resulting in the verified intelligence set 404.

[0034] Causal model extraction stage 204: such as Figure 4 As shown, the intelligence in the verification intelligence set 404 undergoes relation triple extraction 405. A large language model is used to extract relation triples such as competition, cooperation, investment, bidding, and product launch. Each relation triple includes a subject, relational predicate, and object. The extracted relation triples are then filtered through a named entity set 406, retaining only triples related to known named entities, events, or intelligence identifiers. The filtered relation triples are then written into the knowledge graph database 107, and event and intelligence data are stored synchronously. When the competitive value chain model needs to be rebuilt, the original graph data corresponding to the target enterprise in the knowledge graph database 107 is first deleted, and then rewritten, achieving idempotent reconstruction. Graph data can be partitioned according to the target enterprise for isolated management.

[0035] Insight Generation and Verification Phase 205: Utilize a large language model to generate enterprise profiles, market profiles, customer profiles, competitor dynamics, market dynamics, customer dynamics, and competitive landscape data; generate graph relationships such as competition, cooperation, market segmentation, and customer groups; use a second language model to verify the credibility of the generated insights, and fill in missing or low-credibility fields; finally, generate page data, and supplement the page data from the intelligence database when the number of page data is less than the predetermined amount.

[0036] After the above stages are completed, this implementation method generates a competitive value chain model output 206, which summarizes the processing results of each stage, including ontology modeling, competitive intelligence collection, source credibility verification, causal model extraction and insight generation and verification, and outputs the final competitive value chain model.

[0037] Furthermore, in this embodiment, the generation tasks of the competitive value chain model are scheduled and executed asynchronously via a message queue. Tasks are enqueued according to priority, with interactive tasks having higher priority than batch tasks. The worker process retrieves tasks from the queue in a blocking manner and executes them. The system maintains a state machine for the tasks, with states including not generated, generating, completed, failed, and paused. The generation progress is mapped to predetermined progress intervals in stages, and the state and progress are simultaneously written to the relational database 106 and the document database 105 to achieve dual-write persistence. The progress is pushed to the front end through a real-time progress stream interface. The system also implements tenant-level and global-level concurrent slot control, as well as concurrent control, result caching, and timeout protection for language model calls.

[0038] In this embodiment, the data sources for the competitive value chain model include user-selected companies, products, customers, partners, markets, competitors, enterprise files and business registration data, as well as news intelligence collected through third-party intelligence services and profiles and relationships generated by language models. Each type of data is stored in a document database 105, a relational database 106, and a knowledge graph database 107, respectively. For example, the model's main document, enterprise profiles, competitive intelligence, and page data are stored in the document database 105; model metadata (status, progress, current stage, build logs, graph identifiers, etc.) and points balance are stored in the relational database 106; and ontology and relation triples are stored in the knowledge graph database 107.

[0039] In summary, this invention integrates six core technologies: ontology modeling, competitive intelligence gathering, source credibility verification, causal model extraction, insight generation and verification, and model generation. It addresses industry pain points in the field of competitive intelligence, such as severe fragmentation of multi-source data, difficulty in verifying credibility, low degree of structuring, and difficulty in automatically aggregating data into a competitive value chain model. It belongs to the interdisciplinary technology field of intelligence processing and data analysis.

[0040] The specific implementation methods of each module performing the corresponding operation in the apparatus involved in the foregoing embodiments have been described in detail in this method embodiment, and will not be repeated here.

[0041] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the specific structural details of the above embodiments. All simple modifications made to the technical solutions of the present invention within the scope of protection of the inventive concept fall within the protection scope of the present invention.

[0042] It should also be noted that all the technical features described in the foregoing embodiments can be combined with each other in any adaptable way, provided there is no conflict; to avoid redundancy, the present invention will not elaborate on each of the various feasible combinations.

[0043] At the same time, the various embodiments of the present invention can be freely combined. As long as the combination scheme does not deviate from the core technical concept of the present invention, it is all within the scope of the disclosure of the present invention.

Claims

1. A method for multi-source intelligence convergence and analysis, characterized in that, Includes the following steps: The ontology modeling step involves identifying the target company and at least one associated entity among products, customers, partners, markets, and competitors related to the target company, constructing an ontology knowledge graph containing the target company and the associated entities, and supplementing the attribute information of each entity; the competitive intelligence collection step involves constructing search keywords for the target company and the associated entities, retrieving competitive intelligence from multiple intelligence sources, and obtaining a candidate intelligence set. The information source credibility verification step evaluates the credibility of each piece of information in the candidate information set, filters out information with credibility below a predetermined threshold, and obtains a verification information set. The causal model extraction step uses a large language model to extract relation triples from the information in the verification information set, and filters the extracted relation triples based on a preset named entity set. The insight generation and verification step generates a competitive intelligence profile and competitive situation insight of the target enterprise based on the filtered relation triples, and verifies and completes the insight. The model generation step generates a competitive value chain model based on the ontology knowledge graph, the verification intelligence set, the filtered relation triples, and the verified insights.

2. The multi-source intelligence convergence and analysis method according to claim 1, characterized in that, The construction of search keywords specifically includes: constructing bilingual (Chinese and English) keywords for the target company and the associated entities, placing the English keywords before the Chinese keywords, prioritizing the use of the English keywords for retrieval, and using the Chinese keywords as a supplement; extracting Chinese and English industry keywords from the name of the industry to which the target company belongs, and combining the industry keywords with Chinese and English event-related keywords to construct a combined query of industry terms and event terms.

3. The multi-source intelligence convergence and analysis method according to claim 1, characterized in that, The process of retrieving competitive intelligence from multiple intelligence sources specifically includes: retrieving from at least two data sources, such as internal intelligence databases, third-party intelligence aggregation services, public news sources, and corporate websites, and determining the relevance of the search results to filter out intelligence that is irrelevant to the target company or the associated entity.

4. The multi-source intelligence convergence and analysis method according to claim 1, characterized in that, The source credibility assessment specifically includes: calculating the credibility score of each intelligence based on at least one of the following factors: the authority level of the source media, the accessibility of the intelligence link, the completeness of the intelligence attributes, and the relevance of the intelligence to the target enterprise; and retaining intelligence with a credibility score not lower than the predetermined threshold and whose intelligence link is accessible.

5. The multi-source intelligence convergence and analysis method according to claim 1, characterized in that, The relation triple includes a subject, a relation predicate, and an object. The relation predicate includes at least one of competition, cooperation, investment, winning a bid, and product launch. The method also includes a knowledge graph writing step: the filtered relation triple is written into the knowledge graph. When the competitive value chain model needs to be rebuilt, the original graph data corresponding to the target enterprise in the knowledge graph is deleted first, and then rewritten to achieve idempotent reconstruction.

6. The multi-source intelligence convergence and analysis method according to claim 1, characterized in that, The verification and completion of the insights specifically includes: using a second language model to verify the credibility of the generated insights, and completing any missing or low-credibility fields.

7. The multi-source intelligence convergence and analysis method according to claim 1, characterized in that, It also includes an asynchronous scheduling step: the ontology modeling step, the competitive intelligence collection step, the source trust verification step, the causal model extraction step, and the insight generation and verification step are treated as asynchronous tasks and scheduled for execution through a message queue; the state machine and execution progress of the asynchronous tasks are maintained, and the execution progress is pushed to the front end in real time.

8. A multi-source intelligence aggregation and analysis system, characterized in that, include: The ontology modeling module is used to identify the target enterprise and at least one associated entity associated with the target enterprise, construct an ontology knowledge graph, and supplement the attribute information of each entity. The competitive intelligence collection module is used to construct search keywords for the target enterprise and the related entities, retrieve competitive intelligence from multiple intelligence sources, and obtain a candidate intelligence set. The source credibility verification module is used to evaluate the source credibility of each intelligence in the candidate intelligence set, filter out intelligence with credibility below a predetermined threshold, and obtain a verified intelligence set. The causal model extraction module is used to extract relation triples from the intelligence in the verification intelligence set using a large language model, and to filter the extracted relation triples based on a preset named entity set; the insight generation and verification module is used to generate competitive intelligence profiles and competitive situation insights based on the filtered relation triples, and to verify and complete the insights. The model generation module is used to generate a competitive value chain model based on the processing results of the above modules.

9. The multi-source intelligence convergence and analysis system according to claim 8, characterized in that, Also includes: The asynchronous scheduling module is used to asynchronously schedule the processing tasks of each module through a message queue, maintain the task state machine and execution progress, and push the execution progress in real time. The knowledge graph storage module is used to store the ontology knowledge graph and the filtered relation triples, and deletes the original graph data before rewriting it during model reconstruction.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the multi-source intelligence convergence and analysis method as described in any one of claims 1 to 7.

Images