A conflict-driven incremental retrieval method, system, and electronic device

By employing conflict clusters, fact fingerprints, and evidence gap vectors in the retrieval system to dynamically adjust the retrieval scope, the problems of resource waste and inaccurate conflict identification in existing technologies are solved, achieving efficient and accurate incremental retrieval and verification closed loop.

CN123115298APending Publication Date: 2026-11-17CHINA ELECTRONICS CLOUD DIGITAL INTELLIGENCE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611497185.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing retrieval systems suffer from low resource utilization, inability to accurately locate conflicts, disconnect between retrieval and verification, lack of interpretable stopping conditions, and absence of cross-source evidence profiles, resulting in low retrieval efficiency and reduced credibility.

Method used

Using conflict clusters as retrieval units, the retrieval scope is dynamically adjusted through fact fingerprints, conflict graphs, and evidence gap vectors. Combined with expected information gain and cost constraints, adaptive focusing and closed-loop supplementary evidence are achieved.

Benefits of technology

It improves the accuracy of conflict detection, reduces duplicate recall, enhances retrieval efficiency and reliability, provides interpretable stopping conditions, and facilitates manual review.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115298A_ABST
    Figure CN123115298A_ABST
Patent Text Reader

Abstract

This application discloses a conflict-driven incremental retrieval method and system. The method includes: acquiring the initial retrieval results and parsing them into candidate fact units; generating fact fingerprints including core fact fingerprints, contextual fingerprints, and source lineage fingerprints; constructing a conflict graph and identifying conflict clusters; generating evidence gap vectors for the conflict clusters; selecting incremental retrieval actions based on the evidence gap vectors and the utility of the retrieval actions and generating an incremental retrieval execution graph; executing the incremental retrieval and writing back the conflict graph; and outputting resolved conclusions or unresolved conflicts based on the conflict status. This application uses conflict as the retrieval driver and aims for a minimum sufficient evidence set. It can distinguish between genuine conflicts and superficial contradictions, avoid redundant weighting of information from the same source, automatically supplement evidence, and provide interpretable stopping conditions, thereby improving the accuracy and efficiency of conflict resolution with a relatively low retrieval cost.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the fields of information retrieval and trusted intelligence analysis technology, specifically to a conflict-driven incremental retrieval method, system, and electronic device. Background Technology

[0002] With the continuous generation of massive amounts of heterogeneous information from various data sources such as the internet, industry databases, social media, news feeds, public reports, and internal business systems, intelligence analysis work targeting individuals, organizations, events, equipment, products, markets, and risks typically requires retrieving, aggregating, and cross-validating relevant facts from multiple data sources. However, existing retrieval systems generally organize searches using methods such as keyword expansion, fixed search templates, periodic full crawling, or one-time recall based on ranking scores. For multiple sources on the same topic, the system usually simply aggregates the query results, and then the aggregated results are summarized manually or by a large language model.

[0003] The existing technology has at least the following shortcomings:

[0004] (1) Low utilization of retrieval resources. Fixed and broad queries will repeatedly obtain a large amount of known and consistent information, making it impossible to allocate retrieval budgets specifically for the uncertain parts of the information; and full review will further waste interface quotas, proxy resources and computing resources.

[0005] (2) Conflicts cannot be accurately located. Descriptions of the same fact from different sources often differ in terms of subject aliases, time granularity, units of measurement, event stages, and citation chains. Existing systems lack the ability to identify and normalize these differences, often directly judging content such as "80 aircraft vs. 60 aircraft" or "2026 contract vs. 2025 framework agreement" as contradictory, thus generating a large number of false conflicts. Conversely, for content with the same wording but highly homogeneous sources of evidence, existing systems are prone to misjudging them as mutually corroborating each other, forming false consistency.

[0006] (3) The retrieval and judgment are separated. The existing process generally adopts a serial approach of "retrieve first, then analyze": when information conflicts are found in the analysis stage, the query terms can only be reconstructed and entered manually. The system cannot automatically generate the next round of retrieval tasks for the identified conflict gaps, which leads to the interruption of the verification chain and makes it difficult to close the loop between retrieval and verification.

[0007] (4) Lack of interpretable stopping conditions. Existing methods usually use a fixed number of search results, a fixed number of search rounds, or no new results as the conditions for ending the search. This cannot prove that the conflict has been fully resolved, nor can it clearly indicate what evidence is missing when the conflict cannot be resolved. The auditability of the search process is insufficient.

[0008] (5) Lack of cross-source evidence spectrum. For information that has been reproduced, cited, translated or reprocessed, the existing system usually counts it as independent evidence, which leads to the misjudgment of content from the same source as mutual corroboration from multiple independent sources, thereby reducing the accuracy of credibility assessment.

[0009] In summary, existing technologies are insufficient to meet the needs for targeted, incremental, and auditable retrieval based on actual conflicts. Therefore, it is necessary to provide a conflict-driven incremental retrieval method to overcome these shortcomings. Summary of the Invention

[0010] To overcome the aforementioned deficiencies in existing technologies, this application proposes a novel conflict-driven incremental retrieval method and system. More specifically, this invention proposes an incremental retrieval method and system that uses semantic conflicts between multi-source information as triggering conditions and aims at a minimum sufficient evidence set. It aims to solve problems existing in current retrieval systems, such as indiscriminate expansion of the retrieval scope, inaccurate identification of factual conflicts, inability to automatically supplement evidence after a conflict occurs, duplicate weighting of homologous information, and unexplainable retrieval termination. The solution of this invention constructs fact fingerprints, conflict graphs, evidence gap vectors, and incremental retrieval execution graphs, enabling the retrieval scope to dynamically converge with the conflict state, thereby obtaining sufficient heterogeneous evidence to support the conclusion with a relatively small retrieval cost.

[0011] To achieve the above objectives, the present invention employs the following technical strategies:

[0012] 1. It proposes to use "conflict clusters" as the smallest retrieval unit, rather than documents, topics or single keywords, so as to achieve adaptive focusing of retrieval resources on high-risk and uncertain areas.

[0013] 2. A three-layer association mechanism is proposed, consisting of factual core fingerprint, contextual fingerprint, and source lineage fingerprint, which simultaneously solves the problems of factual comparability, pseudo-conflict identification, and duplicate weighting in homologous propagation.

[0014] 3. Propose a computable evidence gap vector to resolve conflicts into operational dimensions such as authority, independence, time anchor, statistical caliber, original documents, and object disambiguation.

[0015] 4. An incremental retrieval utility function is proposed that combines expected information gain, source independence, data source health, acquisition cost, and duplication penalty to achieve dynamic retrieval orchestration in a multi-backend environment.

[0016] 5. A joint mechanism of counter-evidence retrieval and version chain tracing is proposed, enabling the system to not only search for supporting evidence, but also to actively search for evidence that may overturn or limit the current conclusion.

[0017] 6. Propose an interpretable cessation mechanism for unresolved conflicts, which can output a clear reason for "indeterminability" and the type of evidence that still needs to be supplemented, rather than misjudging the lack of evidence as the absence of conflict.

[0018] Specifically, this application provides the following technical solutions:

[0019] The first aspect of this application provides a conflict-driven incremental retrieval method, such as... Figure 1 As shown, it includes the following steps:

[0020] S1. Receive the topic to be analyzed, the initial query, and the data source constraints. Perform the first round of retrieval according to the initial query and the data source constraints, and parse each result in the first round of retrieval results into a candidate fact unit. Normalize the candidate fact units to generate a fact fingerprint. The fact fingerprint includes a fact core fingerprint, a context fingerprint, and a source lineage fingerprint.

[0021] S2. Construct a conflict graph based on the fact fingerprint. The conflict graph uses normalized candidate fact units as nodes and the relationships between facts as edges. The conflict connected subgraphs in the conflict graph are determined as conflict clusters. The conflict type, conflict intensity and resolvability score of each edge in the conflict graph are also determined.

[0022] S3. Generate an evidence gap vector for each conflict cluster. The evidence gap vector represents the dimension of the supplementary evidence required to resolve the corresponding conflict cluster. Evaluate candidate incremental retrieval actions based on the evidence gap vector and the utility of the retrieval action, select incremental retrieval actions and generate an incremental retrieval execution graph. Execute the incremental retrieval execution graph to obtain incremental evidence. After deduplication, genealogy identification, fact extraction and confidence labeling of the incremental evidence, write back to the conflict graph.

[0023] S4. Based on the updated conflict status, when the conflict resolution conditions are met, output the resolved conclusion and evidence spectrum; when the retrieval budget is reached or there is no incremental retrieval action that satisfies the preset conditions, output the unresolved conflict, missing evidence type and risk level.

[0024] Furthermore, in the method of this application, in step S1, the candidate fact unit includes subject, predicate, object, time, location, value, unit, limiting condition, source and citation relationship; the normalization process includes entity alignment, time normalization, value unit conversion, event stage identification and source lineage tracing.

[0025] Furthermore, in the method of this application, in step S1, the core fact fingerprint consists of a subject, a predicate, an object, and a normalized value; the context fingerprint consists of at least two of time interval, region, statistical caliber, and event stage; the source lineage fingerprint is generated based on at least one of publishing subject, first publication time, citation relationship, text similarity, and propagation path; the reprint or derivation relationship between candidate fact units is identified based on the source lineage fingerprint, and candidate fact units belonging to the same source lineage are weighted together in evidence counting and confidence update to reduce the impact of homologous propagation on fact confidence.

[0026] Furthermore, in the method of this application, in step S2, the edges in the conflict graph include consistent edges, direct conflict edges, context conflict edges, undetermined edges, and edges propagating from the same source; the conflict intensity is determined by the following formula:

[0027] ;

[0028] Where C is the conflict intensity, D_v is the numerical or attribute difference, D_t is the temporal incompatibility, D_c is the context incompatibility, D_s is the semantic relationship conflict, A_p is the source lineage overlap, and w1 to w5 are weight coefficients.

[0029] Furthermore, in the method of this application, in step S3, the evidence gap vector includes at least the authoritative source gap, independent source gap, time anchor gap, caliber definition gap, original document gap, geographical or object disambiguation gap, and version chain gap.

[0030] Furthermore, in the method of this application, in step S3, the utility U(a) of the retrieval action is determined by the following formula:

[0031] ;

[0032] Wherein, EIG(a) is the expected information gain, AUTH(a) is the authority of the target source, IND(a) is the independence from existing evidence, FRESH(a) is the timeliness matching degree, COST(a) is the acquisition cost, RED(a) is the expected repetition degree with existing results, and α, β, γ, δ, λ and μ are weight parameters; the incremental retrieval action with the highest comprehensive utility and marginal repetition degree below the preset threshold is selected.

[0033] Furthermore, in the method of this application, in step S3, the incremental retrieval action includes at least one of the following: original announcement location, authoritative institution site retrieval, time anchor supplementary inspection, terminology and unit expansion retrieval, cross-language equivalence retrieval, object alias disambiguation retrieval, citation chain reverse tracing, counter-evidence retrieval, data source cross-retrieval, and version difference retrieval; wherein, the query of the counter-evidence retrieval is constructed based on the current claim and includes at least one of the following: opposite numerical range, alternative event stage, cancellation or revision status, and adjacent time window;

[0034] The incremental retrieval execution graph is a directed acyclic graph, whose nodes include retrieval statements, target data sources, evidence gaps to be filled, budget limits, timeout policies, and success criteria. Its edges indicate that the subsequent node is executed after the evidence of the corresponding preceding node satisfies the success criteria.

[0035] Furthermore, in the method of this application, in step S4, the conflict resolution conditions include at least one of the following: authoritative original evidence that meets the independence threshold appears in the same conflict cluster; the conflicting parties can be compatible with time windows, statistical calibers or event phase rules; the source lineage of the evidence on one side of the conflict is identified as a reproduction or derivation of the evidence on the other side; the posterior probability of the conflict is lower than a preset threshold.

[0036] The conditions for stopping incremental retrieval include at least one of the following: the posterior confidence of the factual conclusion reaches the target confidence and there are at least two independent chains of evidence; the expected information gain of all candidate incremental retrieval actions is lower than the minimum marginal information gain; or the retrieval budget or maximum retrieval depth is exhausted.

[0037] A second aspect of this application provides a conflict-driven incremental retrieval system, wherein the system implements the steps of the aforementioned conflict-driven incremental retrieval method during operation, and the system includes:

[0038] The initial retrieval module is used to obtain initial retrieval results and generate candidate fact units;

[0039] The fact normalization module is used to normalize the candidate fact units and generate fact fingerprints, which include fact core fingerprints, context fingerprints, and source lineage fingerprints.

[0040] The source genealogy module is used to identify original sources, reprinted sources, cited sources, and derived sources, and to calculate source independence;

[0041] The conflict graph construction module is used to construct and update the conflict graph based on the fact fingerprint;

[0042] The evidence gap diagnosis module is used to generate evidence gap vectors for conflict clusters in the conflict graph.

[0043] The incremental orchestration module is used to select incremental retrieval actions and generate an incremental retrieval execution graph based on the evidence gap vector and the utility of the retrieval action.

[0044] The retrieval execution module is used to execute the incremental retrieval execution graph to obtain incremental evidence, and to update the conflict graph using the incremental evidence;

[0045] The conclusion generation module is used to output resolved conclusions or pending conflict results based on the updated conflict status.

[0046] Furthermore, the system in this application also includes a data source adaptation layer for connecting to at least one of a search engine, a news database, government and corporate announcements, an internal knowledge base, and a web page collection interface;

[0047] The source lineage module counts multiple pieces of evidence belonging to the same source lineage as a single chain of evidence.

[0048] The conclusion generation module maintains a retrieval budget, maximum retrieval depth, target confidence level, and minimum marginal information gain for each conflict cluster. It terminates incremental retrieval based on the condition of stopping incremental retrieval. When the conflict is not resolved, it outputs the unresolved conflict, missing evidence type, and risk level for manual review.

[0049] A third aspect of this application provides an electronic device, including: a memory and a processor;

[0050] Memory: Used to store computer programs;

[0051] Processor: Used to execute the computer program to implement the steps of the aforementioned conflict-driven incremental retrieval method.

[0052] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned conflict-driven incremental retrieval method.

[0053] In summary, compared with the prior art, the solution of the present invention has the following advantages:

[0054] (1) Using conflict rather than keyword popularity as the search driver can concentrate the search budget on uncertain facts that affect the results and reduce duplicate recall.

[0055] (2) By using the three-layer fingerprint of fact core, context and source lineage, the real contradiction can be distinguished from the superficial contradiction caused by differences in time, caliber, unit and event stage, thereby improving the accuracy of conflict judgment.

[0056] (3) By using the evidence gap vector, “what else needs to be checked” is transformed into a machine-executable task, and the most suitable retrieval source and retrieval method can be automatically selected to achieve a closed loop from discovering conflicts to supplementing evidence.

[0057] (4) By deduplicating source lineages and calculating independence, we can avoid reprinting, translation and secondary abstracts being used as independent evidence, thereby improving the authenticity of multi-source corroboration.

[0058] (5) An incremental scheduling strategy constrained by both expected information gain and cost is adopted to improve the conflict resolution rate under the same retrieval budget; when the conflict cannot be resolved, the missing original file, time anchor or independent source can be clearly output to facilitate manual review.

[0059] (6) The execution diagram, evidence version, and conflict status changes can all be recorded, which can support result traceability, auditing, reproduction and model evaluation. Attached Figure Description

[0060] To more clearly illustrate the technical solution of this application, the accompanying drawings involved in the description of this invention will be briefly introduced below. It should be noted that the drawings only show some embodiments of the invention. For those skilled in the art, other related drawings can be derived from these drawings without creative effort.

[0061] Figure 1 This is a schematic diagram illustrating the operation steps of the conflict-driven incremental retrieval method of the present invention.

[0062] Figure 2 This is a schematic diagram of the overall architecture of the conflict-driven incremental retrieval system of the present invention.

[0063] Figure 3 This is a schematic diagram illustrating the process of generating fact fingerprints, lineages, and conflict graphs in an embodiment of the present invention.

[0064] Figure 4 This is a schematic diagram of the evidence gap vector generation and incremental retrieval action selection process in an embodiment of the present invention.

[0065] Figure 5 This is a schematic diagram of the conflict resolution, pending output, and evidence auditing process in an embodiment of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the protection scope of this application.

[0067] In this document, the term "comprising" and any variations thereof (such as "including," "including," etc.) are open-ended expressions and should be understood as "including but not limited to," meaning that the listed content is not exhaustive and may include other content not explicitly mentioned. The term "based on" should be understood as "at least partially based on," meaning that the basis or condition referred to may not be the only factor and may involve other relevant factors. The term "one embodiment" should be understood as "at least one embodiment," meaning that the described embodiment is not the only possible implementation, and other similar embodiments may exist.

[0068] In this application, the terms "a" and "a plurality of" are used to modify related elements or features, and their expression is illustrative rather than restrictive. Unless otherwise expressly stated in the context, "a" should be understood as "at least one," and "a plurality of" should be understood as "at least two." Those skilled in the art should reasonably interpret these terms based on the semantic and logical relationships of the context to ensure that they cover the possibility of "one or more."

[0069] This invention provides the following technical solution:

[0070] Firstly, the present invention provides a conflict-driven incremental retrieval method, such as... Figures 3-5 As shown, it includes the following steps:

[0071] 1. Receive the topic to be analyzed, the initial query, and the data source constraints; obtain the first round of search results, and parse each result into candidate fact units containing subject, predicate, object, time, location, value, unit, limiting conditions, source, and citation relationship.

[0072] 2. Perform entity alignment, time normalization, numerical unit conversion, event stage identification, and source lineage tracing on candidate fact units to generate fact fingerprints; the fact fingerprints include at least the fact core fingerprint, context fingerprint, and source lineage fingerprint.

[0073] 3. Construct a conflict graph based on fact fingerprints. Nodes in the conflict graph are candidate fact units, and edges include consistent edges, direct conflict edges, contextual conflict edges, undetermined edges, and homologous propagation edges; calculate the conflict type, conflict intensity, and resolvability score for each edge.

[0074] 4. Generate an evidence gap vector for each conflict-connected subgraph. The evidence gap vector shall include at least the following: authoritative source gap, independent source gap, time anchor gap, caliber definition gap, original document gap, geographical or object disambiguation gap, and version chain gap.

[0075] 5. Based on the conflict intensity, the impact of the conclusion, the evidence gap vector, the data source capability, the health of the data source, and the cost and timeliness requirements, evaluate the utility of the candidate search actions; select the search actions with the highest utility and marginal repetition below the threshold, and generate an incremental search execution graph.

[0076] 6. Call one or more data sources according to the incremental retrieval execution graph to obtain incremental evidence; perform deduplication, genealogy identification, fact extraction and confidence labeling on the incremental evidence, and write back the conflict graph.

[0077] 7. Recalculate the conflict state. When the conflict resolution conditions are met, output the conclusion and evidence spectrum; when the retrieval budget is reached or there are no positive utility retrieval actions, output the unresolved conflict, missing evidence type, and risk level.

[0078] The conflict resolution conditions include at least one of the following: authoritative original evidence that meets the independence threshold appears in the same conflict cluster; the conflicting parties are compatible with time windows, statistical methods or event phase rules; the source lineage of evidence from one side is identified as a reproduction or derivation of evidence from the other side; or the posterior probability of the conflict is lower than a preset threshold.

[0079] Preferably, the conflict intensity C in step 3 is determined by the following formula:

[0080]

[0081] In the formula, D_v represents the numerical or attribute difference, D_t represents the temporal incompatibility, D_c represents the context incompatibility, D_s represents the semantic relationship conflict, A_p represents the source lineage overlap, and w1 to w5 are configurable weights. By introducing source lineage overlap, the intensity of spurious conflicts between reposts from the same source can be reduced.

[0082] Preferably, the overall utility U(a) of the retrieval action a in step S5 is determined by the following formula:

[0083]

[0084] In the formula, EIG(a) is the expected information gain, AUTH(a) is the authority of the target source, IND(a) is the independence from existing evidence, FRESH(a) is the timeliness matching degree, COST(a) is the acquisition cost, RED(a) is the expected repetition degree with existing results, and α, β, γ, δ, λ and μ are weight parameters.

[0085] The incremental retrieval actions include at least: locating the original announcement, searching within the site of authoritative institutions, supplementing the time anchor, expanding the terminology and unit search, cross-language equivalence search, disambiguation of object aliases, tracing back the citation chain, rebuttal search, cross-data source search, and version difference search.

[0086] Secondly, the present invention provides a conflict-driven incremental retrieval system, such as... Figure 2 As shown, it includes:

[0087] 1. Initial retrieval module, used to perform the first round of multi-source retrieval and receive candidate evidence;

[0088] 2. Fact normalization module, used to extract fact units, calculate fact fingerprints, and normalize entities, time, units, and event stages;

[0089] 3. Source lineage module, used to identify original sources, reprinted sources, cited sources, and derived sources, and to calculate source independence;

[0090] 4. Conflict graph construction module, used to build and update conflict relationships between candidate facts;

[0091] 5. Evidence gap diagnosis module, used to generate evidence gap vectors for each conflict cluster;

[0092] 6. Incremental orchestration module, used to select retrieval actions, data sources, and execution order based on overall utility;

[0093] 7. Retrieval and execution module, used to obtain incremental evidence through search interface, database interface, web page collection interface or internal knowledge base interface;

[0094] 8. Conclusion generation module, used to output resolved conclusions, pending conflicts, evidence genealogy, retrieval process, and risk statement.

[0095] To more clearly illustrate the technical solution of this application, the following will provide further explanation through specific scenario embodiments.

[0096] The following example illustrates the invention using the scenario of "two different statements regarding the quantity of an export order for a certain piece of equipment: 60 and 80 aircraft," but the invention is not limited to this scenario.

[0097] Example 1: Initial Fact Construction

[0098] The user inputs the topic "Export order size for a certain equipment" and performs an initial query. The initial retrieval module retrieves first-round documents from news sites, company announcements, government procurement platforms, industry databases, and internal repositories. The fact normalization module extracts data in the form of:

[0099] F = <subject, predicate, object, time interval, numerical value, unit, statistical caliber, event stage, region, source>

[0100] The system parses the facts into units. For example, "Company A signed a purchase agreement for 80 aircraft in 2026" and "Country A plans to purchase 60 aircraft" are parsed separately. The system identifies different event stages for "signed" and "planned purchase," marks "2026" as a time interval, and marks "aircraft" as a quantity unit.

[0101] Subsequently, a three-layer fingerprint is generated: the core fact fingerprint Hc consists of the subject, predicate, object, and normalized value; the context fingerprint Hx consists of time, region, event stage, and statistical caliber; and the source lineage fingerprint Hp consists of the publishing entity, first publication time, URL, cited links, text similarity, and propagation path. This two-layer fact and lineage identifier allows for both determining whether content is identical and whether evidence is independent.

[0102] Example 2: Conflict Identification and False Conflict Elimination

[0103] The conflict graph construction module uses fact units as nodes. If the core objects of two nodes are the same, their predicates are comparable, and their numerical ranges do not intersect, a direct conflict edge is generated; if the numerical values ​​are different but the event stages are different, a contextual conflict edge is generated; if both articles cite the same original report, a common-source propagation edge is generated.

[0104] Regarding the discrepancy between "60 planned purchases" and "80 signed agreements," the system first identifies a numerical difference. However, due to the different stages of the events, the conflict type is labeled as "pending verification" rather than "direct contradiction." The evidence gap diagnosis module then generates the following: original contract gap, planning stage time anchor point gap, and quantity gap. This process prevents the system from generating erroneous alerts based solely on numerical differences.

[0105] Example 3: Incremental Search Action Generation and Arrangement

[0106] To address the aforementioned gaps, the system generates a set of candidate actions: A1 is to search for departmental announcements from the procuring country; A2 is to search for the company's investor relations page; A3 is to search for the contract number; A4 is to search for "equipment name + quantity + signature" in the local language; A5 is to search for revisions before and after the earliest publication date; and A6 is to search for third-party reposts.

[0107] The incremental orchestration module calculates the overall utility of each action. Since A1 and A2 are more likely to obtain the original documents, have higher independence, and lower costs, the system prioritizes executing A1 and A2; A6 has a high overlap with existing news source lines and is ranked lower due to duplication penalties. If A1 finds "first batch of 60 aircraft, total frame count of 80 aircraft," the system splits the original conflict into two non-conflicting facts: "first batch delivery volume" and "total frame agreement volume," and updates the conflict status to resolved.

[0108] The execution graph is not a fixed linear queue, but a directed acyclic graph or an iterable graph. Each node includes a retrieval statement, target data source, expected evidence gap to be filled, budget limit, timeout policy, and success criteria; each edge indicates that subsequent tasks are executed only after the preceding evidence is satisfied. For example, the subsequent node "retrieve original document by contract number" is triggered only when the contract number is found. This avoids meaningless broad searches.

[0109] Example 4: Counter-evidence retrieval and confidence update

[0110] For conflict clusters where the impact of the conclusion exceeds a threshold, the system automatically generates a counter-evidence query. This counter-evidence query does not simply negate keywords, but rather generates opposing numerical ranges, alternative event stages, cancellation / revision statuses, and adjacent time windows based on the current claim. For example, if the current evidence supports "80 aircraft have been signed," the system searches for combinations such as "80 canceled," "80 revised," "60 in the first batch," and "80 frames."

[0111] When new evidence is introduced, the system updates the posterior confidence level based on source authority, source independence, temporal proximity, originality, degree of structuring, and consistency with other evidence. For multiple reprints from the same original report, the source genealogy module counts them as only one chain of evidence to prevent unduly increasing the confidence level due to an increase in the number of media outlets.

[0112] Example 5: Stop, Output, and Audit

[0113] The system maintains a retrieval budget B, maximum retrieval depth L, target confidence T, and minimum marginal information gain G for each conflict cluster. The process stops when any of the following conditions occur:

[0114] (1) The posterior confidence level of a certain factual conclusion reaches T and there are at least two independent chains of evidence;

[0115] (2) Conflicts are compatible with time, caliber, or event phase rules;

[0116] (3) The expected information gain of all candidate actions is less than G;

[0117] (4) Budget B or depth L is exhausted.

[0118] In addition to the conclusions, the output includes the version of the facts, original and derived sources, the evolution of the conflict, the tasks performed, the reasons for not performing them, the remaining evidence gaps, and the risk level. This output can be presented in assessment reports, alarm systems, or audit interfaces.

[0119] The flowcharts and block diagrams in the accompanying drawings illustrate possible implementations of systems, methods, and computer program products according to various embodiments of this application, including architecture, functionality, and operation. In these figures, each block may represent a module, program segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should be noted that each block in the block diagrams and / or flowcharts, and combinations thereof, can be implemented using either a dedicated hardware-based system or a combination of dedicated hardware and computer instructions to achieve the specified function or operation.

[0120] Embodiments of this application also disclose an electronic device, including: a processor, a communication interface, a memory for storing a processor-executable computer program, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The processor executes the executable computer program to implement the steps of the conflict-driven incremental retrieval method described above.

[0121] It is understood that, in addition to memory and a processor, this electronic device may also include input devices (such as a keyboard), output devices (such as a display), and other communication modules. These input devices, output devices, and other communication modules all communicate with the processor through I / O interfaces (i.e., input / output interfaces).

[0122] The operations described in this application can be implemented by writing computer program code using one or more programming languages ​​or a combination thereof. The programming languages ​​include, but are not limited to, the following types:

[0123] Object-oriented programming languages, such as Java, Smalltalk, C++, etc.

[0124] Conventional procedural programming languages, such as "C" or similar programming languages.

[0125] The execution methods of program code include, but are not limited to:

[0126] It runs entirely on the user's computer;

[0127] Part of it executes on the user's computer, and part of it executes on a remote computer;

[0128] Execute as a standalone software package;

[0129] It is executed entirely on a remote computer or server.

[0130] In scenarios involving remote computers, the remote computer can connect to the user's computer via any type of network, including but not limited to local area networks (LANs) or wide area networks (WANs). Furthermore, the remote computer can also connect to external computers through an internet service provider, for example, by utilizing the internet for connection.

[0131] Furthermore, this application also discloses a computer-readable storage medium, wherein when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform the various steps of the conflict-driven incremental retrieval method disclosed in this application.

[0132] In the context of this application, a computer-readable storage medium refers to a tangible medium capable of storing computer program code and related data. Specific examples include, but are not limited to, the following:

[0133] (1) Portable computer disk: such as floppy disks and other removable magnetic storage media.

[0134] (2) Hard disk: including mechanical hard disks and solid-state hard disks and other fixed storage devices.

[0135] (3) Random Access Memory (RAM): A volatile storage medium used for temporary storage of data and program code.

[0136] (4) Read-only memory (ROM): a non-volatile storage medium used to store fixed programs and data.

[0137] (5) Erasable programmable read-only memory (EPROM) or flash memory: non-volatile storage media that supports multiple erasures and reprogrammings.

[0138] (6) Fiber optic storage devices: storage media based on fiber optic technology.

[0139] (7) Portable compact disc read-only memory (CD-ROM): a read-only medium that stores data in the form of an optical disc.

[0140] (8) Optical storage devices: such as DVDs, Blu-ray discs and other storage media based on optical principles.

[0141] (9) Magnetic storage devices: such as magnetic tapes, disks and other storage media based on magnetic principles.

[0142] (10) Any suitable combination of the above: for example, combining multiple storage media to meet different storage needs.

[0143] These computer-readable storage media can be used to store the program code and related data described in this application to support program execution and persistent data storage.

[0144] Specifically, according to embodiments of this application, the processes described in the flowcharts can be implemented as computer software programs. For example, embodiments of this application relate to a computer program product comprising a computer program carried on a non-transitory computer-readable medium. This computer program contains program code for executing the conflict-driven incremental retrieval method disclosed in this application. When this computer program is executed by a processing system, it can achieve the functions defined in the embodiments of this application.

[0145] While the foregoing discussion contains several specific implementation details, these details should not be construed as limiting the scope of this application. The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this application is not limited to technical solutions formed by specific combinations of the above-described technical features. Furthermore, this application should also cover other technical solutions formed by any combination of the above-described technical features or their equivalents without departing from the foregoing disclosed concept.

[0146] Those skilled in the art should also understand that modifications can be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features, without departing from the spirit and scope of the technical solutions of the embodiments of this application. These modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the core spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A conflict-driven incremental retrieval method, characterized in that, Includes the following steps: S1. Receive the topic to be analyzed, the initial query, and the data source constraints. Perform the first round of retrieval according to the initial query and the data source constraints, and parse each result in the first round of retrieval results into a candidate fact unit. Normalize the candidate fact units to generate a fact fingerprint. The fact fingerprint includes a fact core fingerprint, a context fingerprint, and a source lineage fingerprint. S2. Construct a conflict graph based on the fact fingerprint. The conflict graph uses normalized candidate fact units as nodes and the relationships between facts as edges. The conflict connected subgraphs in the conflict graph are determined as conflict clusters. The conflict type, conflict intensity and resolvability score of each edge in the conflict graph are also determined. S3. Generate an evidence gap vector for each conflict cluster. The evidence gap vector represents the dimension of the supplementary evidence required to resolve the corresponding conflict cluster. Evaluate candidate incremental retrieval actions based on the evidence gap vector and the utility of the retrieval action, select incremental retrieval actions and generate an incremental retrieval execution graph. Execute the incremental retrieval execution graph to obtain incremental evidence. After deduplication, genealogy identification, fact extraction and confidence labeling of the incremental evidence, write back to the conflict graph. S4. Based on the updated conflict status, when the conflict resolution conditions are met, output the resolved conclusion and evidence spectrum. When the retrieval budget is reached or no incremental retrieval action exists that satisfies the preset conditions, output the pending conflict, missing evidence type, and risk level.

2. The method of claim 1, wherein, In step S1, the candidate fact unit includes subject, predicate, object, time, location, value, unit, limiting condition, source and citation relationship; the normalization process includes entity alignment, time normalization, numerical unit conversion, event stage identification and source lineage tracing.

3. The method of claim 1, wherein, In step S1, the core fact fingerprint consists of a subject, predicate, object, and normalized numerical value; the context fingerprint consists of at least two of the following: time interval, region, statistical caliber, and event stage; the source lineage fingerprint is generated based on at least one of the following: publishing subject, first publication time, citation relationship, text similarity, and propagation path; the reprint or derivation relationship between candidate fact units is identified based on the source lineage fingerprint, and candidate fact units belonging to the same source lineage are weighted together in evidence counting and confidence updates to reduce the impact of homologous propagation on fact confidence.

4. The method of claim 1, wherein, In step S2, the edges in the conflict graph include consistent edges, direct conflict edges, context conflict edges, undetermined edges, and edges propagating from the same source; the conflict strength is determined by the following formula: Where C is the conflict intensity, D_v is the numerical or attribute difference, D_t is the temporal incompatibility, D_c is the context incompatibility, D_s is the semantic relationship conflict, A_p is the source lineage overlap, and w1 to w5 are weight coefficients.

5. The method of claim 1, wherein, In step S3, the evidence gap vector includes at least the authoritative source gap, independent source gap, time anchor gap, caliber definition gap, original document gap, geographical or object disambiguation gap, and version chain gap.

6. The method of claim 1, wherein, In step S3, the utility U(a) of the retrieval action is determined by the following formula: ; Wherein, EIG(a) is the expected information gain, AUTH(a) is the authority of the target source, IND(a) is the independence from existing evidence, FRESH(a) is the timeliness matching degree, COST(a) is the acquisition cost, RED(a) is the expected repetition degree with existing results, and α, β, γ, δ, λ and μ are weight parameters; the incremental retrieval action with the highest comprehensive utility and marginal repetition degree below the preset threshold is selected.

7. The method of claim 1, wherein, In step S3, the incremental retrieval action includes at least one of the following: original announcement location, authoritative institution site retrieval, time anchor supplementary inspection, terminology and unit expansion retrieval, cross-language equivalence retrieval, object alias disambiguation retrieval, reference chain reverse tracing, counter-evidence retrieval, data source cross-retrieval, and version difference retrieval; wherein, the counter-evidence retrieval query is constructed based on the current claim and includes at least one of the following: opposite numerical range, alternative event stage, cancellation or revision status, and adjacent time window; The incremental retrieval execution graph is a directed acyclic graph, whose nodes include retrieval statements, target data sources, evidence gaps to be filled, budget limits, timeout policies, and success criteria. Its edges indicate that the subsequent node is executed after the evidence of the corresponding preceding node satisfies the success criteria.

8. The method of claim 1, wherein, In step S4, the conflict resolution conditions include at least one of the following: authoritative original evidence that meets the independence threshold appears in the same conflict cluster; the conflicting parties can be compatible with time windows, statistical methods, or event phase rules; The lineage of evidence from one side of the conflict is identified as a reproduction or derivation of evidence from the other side; the posterior probability of the conflict is lower than a preset threshold. The conditions for stopping incremental retrieval include at least one of the following: the posterior confidence of the factual conclusion reaches the target confidence and there are at least two independent chains of evidence; the expected information gain of all candidate incremental retrieval actions is lower than the minimum marginal information gain; or the retrieval budget or maximum retrieval depth is exhausted.

9. A conflict-driven incremental retrieval system, characterized in that, The system runtime implements the conflict-driven incremental retrieval method as described in any one of claims 1 to 8, including: The initial retrieval module is used to obtain initial retrieval results and generate candidate fact units; The fact normalization module is used to normalize the candidate fact units and generate fact fingerprints, which include fact core fingerprints, context fingerprints, and source lineage fingerprints. The source genealogy module is used to identify original sources, reprinted sources, cited sources, and derived sources, and to calculate source independence; The conflict graph construction module is used to construct and update the conflict graph based on the fact fingerprint; The evidence gap diagnosis module is used to generate evidence gap vectors for conflict clusters in the conflict graph. The incremental orchestration module is used to select incremental retrieval actions and generate an incremental retrieval execution graph based on the evidence gap vector and the utility of the retrieval action. The retrieval execution module is used to execute the incremental retrieval execution graph to obtain incremental evidence, and to update the conflict graph using the incremental evidence; The conclusion generation module is used to output resolved conclusions or pending conflict results based on the updated conflict status.

10. The system according to claim 9, characterized in that, It also includes a data source adaptation layer for connecting to at least one of search engines, news databases, government and corporate announcements, internal knowledge bases, and web page collection interfaces; The source lineage module counts multiple pieces of evidence belonging to the same source lineage as a single chain of evidence. The conclusion generation module maintains a retrieval budget, maximum retrieval depth, target confidence level, and minimum marginal information gain for each conflict cluster. It terminates incremental retrieval based on the condition of stopping incremental retrieval. When the conflict is not resolved, it outputs the unresolved conflict, missing evidence type, and risk level for manual review.