A Multi-Source Information Fusion Method Based on Source Credibility Assessment and Cross-Source Conflict Resolution
By unifying the spatiotemporal benchmark and semantic standardization processing, standard information units are generated and common reference judgment and conflict resolution are performed, which solves the problems of uneven source credibility and cross-source conflict in multi-source information fusion and realizes the reliability and consistency of structured state parsing.
Patent Information
- Application Number
- CN202611327094.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-30
- Publication Date
- 2026-11-17
AI Technical Summary
Under conditions of uneven credibility of multiple information sources, heterogeneous semantic descriptions, and cross-source conflicts, existing technologies struggle to achieve unified semantic standardization, source credibility assessment, and cross-source conflict resolution, resulting in insufficient consistency and reliability of the fusion results.
By performing spatiotemporal alignment and format normalization based on a unified spatiotemporal benchmark, standard information units are generated. Then, a pre-defined semantic standardization large language model is used to perform coreference judgment and conflict resolution, calculate cross-source consistency and comprehensive credibility, and output the fused structured state parsing results.
It enables the assessment of source credibility and the resolution of cross-source conflicts in multi-source information fusion, and generates structured state analysis results with fusion confidence and source tracing relationships, ensuring the reliability and auditability of the fusion results.
Smart Images

Figure CN123116151A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of data processing technology, and in particular to the field of multi-source information fusion technology, specifically to a multi-source information fusion method based on source credibility assessment and cross-source conflict resolution. Background Technology
[0002] In complex application environments, multi-source information typically originates from various sources, including sensing devices, communication monitoring equipment, manual reports, publicly available network information, and historical data. Information from different sources varies significantly in terms of acquisition time, spatial coordinates, descriptive granularity, and expression methods. Multiple, complementary, and even contradictory descriptions often emerge for the same target or event. Relying solely on manual analysts to verify each item is not only inefficient but also makes it difficult to ensure the consistency of the fusion results when the amount of information increases rapidly.
[0003] Currently, one approach to addressing the aforementioned issues involves first normalizing the multi-source information in terms of time and space, and then fusing the information from different sources based on preset rules, static weights, evidence theory, or rule engines to generate state information or risk assessment results. This approach can achieve basic fusion at the data, feature, or decision levels, but it is generally more suitable for structured observation data and lacks a unified, standardized entry point for unstructured textual information, image descriptions, and semantic information such as manual reports.
[0004] Another approach is to directly use large language models to summarize or answer questions from multiple sources. While this approach has strong natural language understanding capabilities, it often directly inputs multi-source information as a whole context into the model, lacking explicit modeling of source credibility, timeliness, and cross-source consistency, and also failing to resolve conflicting information beforehand.
[0005] It is evident that when multiple sources provide contradictory semantic descriptions of the same goal or event, the above approach suffers from at least the following shortcomings: First, the lack of a unified field system for raw information from different sources makes it difficult to conduct detailed comparisons at the semantic slot level; Second, the reliability, timeliness, and current cross-source consistency of sources are usually not uniformly quantified, making it difficult for the fusion process to effectively suppress low-quality sources; Third, the processing of conflict information often remains at the level of simple weighting, static rules, or manual interpretation, making it difficult to output traceable adjudication basis.
[0006] At the same time, directly calling a large language model can easily lead to low-credibility sources, outdated information, or unresolved contradictory descriptions being included in the output, resulting in a lack of source boundaries, fusion confidence, and verification paths in the results.
[0007] Therefore, the technical problem that this invention needs to solve is: how to obtain a structured state analysis result with fusion confidence and source tracing relationship under the conditions of uneven credibility of multiple information sources, heterogeneous semantic descriptions and cross-source conflicts, through unified semantic standardization, source credibility assessment and cross-source conflict resolution. Summary of the Invention
[0008] This disclosure provides a method, system, device, and storage medium for multi-source information fusion based on source credibility assessment and cross-source conflict resolution.
[0009] According to a first aspect of this disclosure, a multi-source information fusion method based on source credibility assessment and cross-source conflict resolution is provided. The method includes: Acquire multi-source information data to be integrated; Based on a unified spatiotemporal reference, the multi-source information data to be fused is spatiotemporally aligned and format normalized to obtain a set of aligned information fragments. The set of information fragments and the preset field system are input into a preset semantically standardized large language model to generate multiple standard information units; Common reference determination is performed on each standard information unit to form a preliminary slot compatibility matrix; Based on the preliminary slot compatibility matrix, the cross-source consistency and overall reliability of each standard information unit are calculated. Based on deterministic incompatibility rules, conflict relationships are established according to the calculated preliminary slot compatibility matrix, and field-level conflict resolution with confidence constraints is performed on the conflict relationships to output fused standard information units; the fused standard information units include at least fused field values, fused confidence, and source tracing sets; Based on the fused standard information units, an auditable structured state parsing result is generated.
[0010] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the multi-source information data to be fused includes at least two information data from distinguishable sources, and the sources belong to different information types, or belong to the same information type but have different source identifiers, acquisition links, or original record pointers.
[0011] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the standard information unit includes at least a structured representation of an entity or event, a relational representation, spatiotemporal elements, a source pointer, a semantic slot field, and a semantic extraction confidence level.
[0012] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein the step of determining common reference for each standard information unit to form a preliminary slot compatibility matrix includes: Based on a preset common goal or event, common reference is determined, and each standard information unit is assigned to the corresponding group to be merged; the preset common goal or event includes explicit goals, event identifiers, and goal or event types, degree of overlap in time range, spatial distance, name or attribute similarity; Within the same group to be merged, a preliminary slot compatibility matrix is formed according to data type, unit, and slot rules.
[0013] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein calculating the cross-source consistency and overall reliability of each standard information unit based on the preliminary slot compatibility matrix includes: Based on the preliminary slot compatibility matrix, calculate the cross-source consistency of each standard information unit; The overall credibility is calculated based on the cross-source consistency of each standard information unit, the historical reliability of its source, the timeliness score, and the semantic extraction confidence.
[0014] In addition to the aspects and any possible implementations described above, an implementation is further provided in which the deterministic incompatibility rule includes: Enumerated or Boolean slots are incompatible when their values are not equal after the vocabulary mapping is completed. The numerical slots first perform unit conversions, and are incompatible if the difference exceeds the absolute or relative tolerance configured for the task. Time slots are incompatible when intervals do not overlap or the endpoint difference exceeds the time tolerance. Spatial slots are incompatible if their confidence regions do not intersect or the distance exceeds the spatial tolerance after taking positional uncertainty into account. State or behavior slots are incompatible if they violate mutual exclusion rules in the domain ontology.
[0015] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein the conflict relationship includes a conflict slot, candidate field values participating in the conflict, a corresponding standard information unit and its source pointer; The field-level conflict resolution, which imposes credibility constraints on the conflict relationships, outputs fused standard information units, including: Calculate the original support of the candidate field values involved in the conflict in the conflict relationship; Calculate the normalized support based on the original support; Based on the preset normalized support threshold and the preset leading difference threshold, the fused standard information unit is output according to the normalized support.
[0016] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein the method further includes: The risk level is calculated based on the fusion confidence level, as well as preset capability, behavior, and spatiotemporal proximity scores.
[0017] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein the method further includes: The conflict relationship, the fused standard information unit, and the source tracing set are input into a preset optional conflict interpretation language model, and a conflict interpretation field is generated under the constraints of a preset interpretation output mode.
[0018] According to a second aspect of this disclosure, a multi-source information fusion system based on source credibility assessment and cross-source conflict resolution is provided. The system includes: The information access module is used to acquire multi-source information data to be integrated; The spatiotemporal alignment module is used to perform spatiotemporal alignment and format normalization on the multi-source information data to be fused based on a unified spatiotemporal benchmark, so as to obtain a set of aligned information fragments. The standard information unit generation module is used to input the set of information fragments and the preset field system into a preset semantic standardization large language model to generate multiple standard information units. The co-reference and preliminary compatibility calculation module is used to determine the co-reference of each standard information unit and form a preliminary slot compatibility matrix; The comprehensive credibility assessment module is used to calculate the cross-source consistency and comprehensive credibility of each standard information unit based on the preliminary slot compatibility matrix. The conflict relationship construction and resolution module is used to establish conflict relationships based on deterministic incompatibility rules and the calculated preliminary slot compatibility matrix, and to perform field-level conflict resolution on the conflict relationships with confidence constraints, and output the fused standard information unit; the fused standard information unit includes at least the fused field value, the fused confidence, and the source tracing set; The state resolution output module is used to generate auditable structured state resolution results based on the fused standard information units.
[0019] According to a third aspect of this disclosure, an electronic device is provided. The electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described above.
[0020] According to a fourth aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the method described above.
[0021] This application provides a multi-source information fusion method based on source credibility assessment and cross-source conflict resolution. It performs spatiotemporal alignment and format normalization on the multi-source information data to be fused based on a unified spatiotemporal benchmark to obtain an aligned set of information fragments. The set of information fragments and a preset field system are input into a preset semantic standardization large language model to generate multiple standard information units, achieving multi-source information access, spatiotemporal alignment, and standard information unit generation. Common reference judgment is performed on each standard information unit to form a preliminary slot compatibility matrix, and cross-source consistency and comprehensive credibility of each standard information unit are calculated accordingly. This achieves target or event common reference, preliminary slot compatibility matrix formation, and cross-source consistency and comprehensive credibility calculation of standard information units. Based on deterministic incompatibility rules, a conflict resolution mechanism is established according to the calculated preliminary slot compatibility matrix. This process identifies conflicting relationships and resolves field-level conflicts by imposing credibility constraints on these relationships. The output is a fused standard information unit. Based on this fused standard information unit, an auditable structured state resolution result is generated. This process enables deterministic conflict graph construction, source lineage group deduplication, field support and normalized support calculation, and structured state resolution output. Therefore, even with uneven credibility among multiple information sources, heterogeneous semantic descriptions, and cross-source conflicts, a structured state resolution result with fusion confidence and source tracing relationships can be obtained through unified semantic standardization, source credibility assessment, and cross-source conflict resolution. This allows the standard information unit to serve as a unified semantic processing carrier, embedding source-level reliability and unit-level quality factors into the multi-source information fusion process. The source pointer follows the standard information unit throughout the comparison, adjudication, and result tracing.
[0022] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description
[0023] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of this disclosure. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein: Figure 1 A flowchart is shown for a multi-source information fusion method based on source credibility assessment and cross-source conflict resolution according to an embodiment of the present disclosure; Figure 2 A schematic diagram illustrating standard information elements, source pointers, and conflict graph relationships according to embodiments of the present disclosure is shown. Figure 3 A flowchart illustrating source credibility assessment, conflict resolution, and status output according to an embodiment of this disclosure is shown; Figure 4A schematic diagram of the overall structure and data flow according to an embodiment of the present disclosure is shown; Figure 5 A block diagram of a multi-source information fusion system based on source credibility assessment and cross-source conflict resolution according to an embodiment of the present disclosure is shown. Figure 6 A block diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. Detailed Implementation
[0024] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.
[0025] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.
[0026] In this disclosure, under the conditions of uneven credibility of multiple information sources, heterogeneous semantic descriptions, and cross-source conflicts, a structured state analysis result with fusion confidence and source tracing relationship can be obtained through unified semantic standardization, source credibility assessment, and cross-source conflict resolution. Thus, standard information units are used as the unified semantic processing carrier for cross-source semantic fusion, and source-level reliability and unit-level quality factors are embedded into the multi-source information fusion process. The source pointer runs through comparison, adjudication, and result tracing along with the standard information units.
[0027] Figure 1 A flowchart of a multi-source information fusion method 100 based on source credibility assessment and cross-source conflict resolution according to an embodiment of the present disclosure is shown.
[0028] In box 110, obtain the multi-source information data to be merged.
[0029] In box 120, spatiotemporal alignment and format normalization are performed on the multi-source information data to be fused based on a unified spatiotemporal benchmark to obtain a set of aligned information fragments.
[0030] In box 130, the set of information fragments and the preset field system are input into the preset semantic standardization large language model to generate multiple standard information units.
[0031] In box 140, common reference is determined for each standard information unit to form a preliminary slot compatibility matrix.
[0032] In box 150, based on the preliminary slot compatibility matrix, the cross-source consistency and overall credibility of each standard information unit are calculated.
[0033] In box 160, based on deterministic incompatibility rules, conflict relationships are established according to the calculated preliminary slot compatibility matrix, and field-level conflict resolution with confidence constraints is performed on the conflict relationships to output the fused standard information unit; the fused standard information unit includes at least the fused field value, fused confidence, and source tracing set.
[0034] In box 170, an auditable structured state resolution result is generated based on the fused standard information units.
[0035] In box 110, input information, namely the multi-source information data to be fused. It is not limited to a fixed type combination. You can select information data that includes at least two distinguishable sources. These sources can belong to different information types, or they can belong to the same information type but have different source identifiers, acquisition links, or original record pointers.
[0036] In some embodiments, each piece of information can be associated with a source identifier, source category, collection time, location coordinates or location uncertainty, upstream source identifier, and processing link identifier to support source lineage identification and subsequent deduplication.
[0037] It should be noted that as long as the source pointer, collection time, spatiotemporal information and source lineage are retained, the multi-source information data to be fused can enter the subsequent processing chain.
[0038] In box 120, spatiotemporal alignment uses a unified time base and a unified coordinate system as a reference, maps timestamps from different sources to the same time axis, and converts different spatial descriptions into a unified coordinate expression; for information where there is an error range in location, the location uncertainty parameter is retained.
[0039] In box 130, the preset field system and preset semantic standardization large language model can be set according to the user's actual needs.
[0040] In some embodiments, each standard information unit includes at least a structured representation of an entity or event, a relational representation, a spatiotemporal element, a source pointer, a semantic slot field, and a semantic extraction confidence level.
[0041] In some embodiments, each standard information sheet may also include a target identifier, event type, relationship type, time range, spatial location, and source identifier.
[0042] In some embodiments, the semantic slot field is used to carry the specific values of the same target or event in terms of attribute values, behavioral descriptions, time descriptions, and location descriptions. The source pointer and semantic extraction confidence are retained together with the corresponding standard information unit and enter the subsequent coreference judgment, slot comparison, confidence calculation, and source tracing processes along with the standard information unit.
[0043] In some embodiments, standard information units are expressed using a preset field pattern. The fields may include unit identifier, target identifier or event identifier, target type or event type, time range, spatial location, location uncertainty, semantic slot name, candidate field value, unit of measurement, source pointer, semantic extraction confidence, and processing status.
[0044] It should be noted that for fields that do not appear in the original information, the corresponding fields will be marked as missing or pending verification, and will not be automatically filled in by the large language model.
[0045] As can be seen, through the above process, standard information units can be used as a unified processing carrier for cross-source semantic fusion, so that source pointers, spatiotemporal elements, semantic slots and semantic extraction confidence can enter into subsequent conflict resolution and source tracing along with the units.
[0046] In box 140, the above-mentioned common reference determination of each standard information unit forms a preliminary slot compatibility matrix, specifically including: Based on the preset common goals or events, common reference is determined, and each standard information unit is assigned to the corresponding group to be merged; the preset common goals or events include explicit goals, event identifiers, as well as the type of goals or events, the degree of overlap in time range, spatial distance (spatial proximity), and name or attribute similarity (attribute constraints). Within the same group to be merged, a preliminary slot compatibility matrix is formed according to data type, unit, and slot rules.
[0047] In some embodiments, preset common goals or events can be set according to the user's specific needs. For example, explicit goal and event identifiers can be set in preset common goals or events, that is, explicit identifier fields such as goal identifier and event identifier have priority comparison permissions.
[0048] Furthermore, when determining coreference for targets or events, explicit identifier fields are compared first; if explicit identifiers are missing, a combination of criteria is used based on target or event type, overlap in time range, spatial distance, and name or attribute similarity. The combined result must reach a preset coreference threshold θ. C When the corresponding standard information units are grouped into the same fusion group, if the preset common-point threshold θ is not reached... C Alternatively, when the data falls within an uncertain range, maintain them as different groups to be merged or mark them as pending verification to avoid incorrectly merging fields from different targets. A preset coreference threshold θ is used. CIt can be determined based on task configuration, historical samples, or expert rules.
[0049] In box 150, the above calculation of cross-source consistency and overall reliability of each standard information unit based on the preliminary slot compatibility matrix specifically includes: Based on the preliminary slot compatibility matrix, the cross-source consistency of each standard information unit is calculated; The overall credibility is calculated based on the cross-source consistency of each standard information unit, the historical reliability of its source, the timeliness score, and the semantic extraction confidence.
[0050] In some embodiments, after forming a preliminary slot compatibility matrix, the cross-source consistency of each standard information unit can be calculated based on this matrix. Subsequently, the source lineage of this unit is obtained by indexing the source pointer. Historical reliability of the source Combined with cross-source consistency Timeliness and semantic extraction confidence Calculate the overall credibility of the i-th standard information unit. Complete the overall credibility assessment. Only after calculation are conflict edges established for incompatible slots in the initial slot compatibility matrix according to deterministic incompatibility rules, thereby constructing conflict relationships and avoiding the reverse definition of cross-source consistency based on the already constructed conflict graph. This creates a cycle.
[0051] In some embodiments, for i standard information units, their comprehensive credibility score It can be represented as: in, Indicates the source of the i-th standard information unit. The reliability of its source history This represents the cross-source consistency score of the unit in the initial slot compatibility matrix. Indicates the timeliness score. This represents the confidence level of semantic extraction. , , and This represents the corresponding weight coefficient, and the sum of all weights is 1. Therefore, it is called the comprehensive credibility of standard information units, and is no longer called the simple source credibility.
[0052] In some embodiments, source history reliability According to the source The source level configuration, historical verification records, or expert evaluation results are initialized and updated when subsequent verification conclusions are obtained; new sources lacking historical records are given a preset neutral initial value and an insufficiently calibrated mark is retained.
[0053] In some embodiments, cross-source consistency Before a conflict relationship is established, the number of slots in the same fusion group that are comparable to and preliminarily compatible with the standard information unit is determined as a percentage of the total number of comparable slots; if there are no comparable units, the neutral value of the task configuration is used and the uncalibrated mark is retained.
[0054] In some embodiments, timeliness The time window is configured according to the information type and then segmented or exponentially decayed.
[0055] In some embodiments, semantic extraction confidence It combines the output of the extracted model, structural pattern verification, and field integrity verification.
[0056] It should be noted that the overall credibility of standard information units It can be represented by a single score, or the reliability of the source history can be preserved. Combined with cross-source consistency Timeliness and semantic extraction confidence Equal components; when performing field-level adjudication, a comprehensive credibility score is then formed according to the task configuration. .
[0057] like Figure 2 As shown, distinguishable source A and distinguishable source B are obtained respectively, both of which include source pointer, collection time, and original fragment. Distinguishing source A and distinguishable source B are standardized to obtain corresponding standard information unit A and standard information unit B, both of which include target, time, location, slot value, and extraction confidence. It is determined whether standard information unit A and standard information unit B point to the same target or event. If not, they are retained as different groups to be merged. If so, the semantic slot values with the same name are compared, and it is determined whether the field values are incompatible. If not, compatible fields are merged and the source is retained. If so, a slot conflict relationship is established, with the unit as a node and the conflicting slot as an edge. Subsequently, the original support and normalized support are calculated according to the source lineage group.
[0058] As can be seen, the source pointer enables each standard information unit to trace back to its original source. Semantic slot comparison is performed on standard information units determined to point to the same target or event; when incompatible field values appear in the same slot, a conflict relationship is constructed by treating the standard information unit as a node and the conflicting slot relationship as an edge. This conflict relationship is specifically used to record the incompatible relationships between candidate slot values for the same target or event, and is not used as a state knowledge graph describing general entity relationships.
[0059] In box 160, after the co-reference and preliminary compatibility matrix are formed, conflicts can be constructed based on deterministic incompatibility rules.
[0060] In some embodiments, conflict relationships can be represented by conflict sets or by conflict graphs with standard information units as nodes and incompatible semantic slots as edges.
[0061] In some embodiments, the aforementioned deterministic incompatibility rules specifically include: Enumerated or Boolean slots are incompatible when their values are not equal after the vocabulary mapping is completed. The numerical slots first perform unit conversions, and are incompatible if the difference exceeds the absolute or relative tolerance configured for the task. Time slots are incompatible when intervals do not overlap or the endpoint difference exceeds the time tolerance. Spatial slots are incompatible if their confidence regions do not intersect or the distance exceeds the spatial tolerance after taking positional uncertainty into account. State or behavior slots are incompatible if they violate mutual exclusion rules in the domain ontology.
[0062] In some embodiments, a conflict relationship is established only if two standard information units have been determined to point to the same target or event and slots with the same name meet the corresponding incompatibility conditions.
[0063] In some embodiments, the conflict relationship record includes at least the conflict slot, the candidate field values involved in the conflict, the corresponding standard information unit and its source pointer; it may also record the conflict type, discovery time and processing status. When new information arrives, the slot comparison is re-performed only for the fusion group that is co-referenced with the new standard information unit, and the corresponding conflict relationship and candidate field support are updated.
[0064] As can be seen, through the above process, semantic slot-level comparisons can be performed on standard information units pointing to the same target or event, and recorded in a conflict set or conflict graph specifically for slot incompatibility relationships.
[0065] In some embodiments, before calculating field support, source lineage groups can be identified based on the original record pointer, upstream source identifier, acquisition link, and processing chain hash to facilitate deduplication of source lineage groups. Multiple direct sources formed from the same upstream original information through forwarding, extraction, or format conversion are first aggregated within a group. The contribution within the group can be determined by taking the maximum overall confidence level, setting a contribution cap, or attenuating according to the correlation coefficient. Then, cross-group support calculations are performed with other independent lineage groups. The fusion result, i.e., the fused standard information unit, simultaneously preserves both the direct source and the upstream source lineage to prevent duplicate votes due to homologous duplication.
[0066] In some embodiments, the field-level conflict resolution that imposes credibility constraints on conflict relationships, as described above, outputs a fused standard information unit, specifically including: Calculate the original support of candidate field values involved in the conflict in the conflict relationship; Calculate normalized support based on the original support; Based on the preset normalized support threshold and the preset leading difference threshold, the fused standard information unit is output according to the normalized support.
[0067] In some embodiments, a preset normalized support threshold is established. and preset leading difference threshold It can be configured according to the user's actual needs.
[0068] In some embodiments, the decision on the candidate field value can be traced back to the standard information unit that participated in the support or conflict determination.
[0069] In some embodiments, the original support for the r-th candidate field value of the same semantic slot is... It can be represented as: in, Indicates support for candidate field values The source lineage set, Representation and support of candidate field values The source lineage of the conflict is a set of lineage groups; each lineage group contributes only once according to a predefined intra-group aggregation rule. and These represent the comprehensive credibility of representative standard information units in the corresponding groups, and η represents the conflict penalty coefficient. Original support. It is only used for sorting and subsequent normalization, and is not directly compared with the probability threshold.
[0070] In some embodiments, when the denominator is greater than zero, the normalized support of the r-th candidate field value in the same semantic slot is: in, Let be the original support of the r-th candidate field value, where r is the index of the current candidate field value, and l is the traversal index of all candidate field values in the same semantic slot. Indicates taking A value larger than 0 ensures that only non-negative support is considered during normalization. This represents the sum of the non-negative support scores for all candidate field values within the same semantic slot. This represents the normalized support value, ranging from [0,1], and is used for subsequent ranking or decision-making.
[0071] Only when the maximum normalized support Not lower than the normalized support threshold Furthermore, the difference between it and the second highest normalized support is not less than the leading difference threshold. Only when the denominator is zero or the maximum value is below the normalized support threshold will a single fused field value be output. Or the lead difference is lower than the lead difference threshold. If so, it remains in a state of pending verification or multiple hypotheses.
[0072] In other words, conflict resolution can directly determine the value of a single fusion field, or it can retain the top K candidate field values to form a multiple hypothesis result when the support is insufficient.
[0073] When the normalization condition is not met, or the maximum normalized support is lower than the normalized support threshold. Or, the difference between the maximum value and the second highest value is less than the leading difference threshold. Instead of forcing a single conclusion, this approach retains the top K candidate field values with the highest support, and saves the original support, normalized support, direct source, upstream source lineage, and conflict set for each candidate field value. This result represents a multi-hypothesis state that can be subsequently read, rather than an unconstrained natural language answer.
[0074] As can be seen, through the above process, slot conflict edges can be established according to the deterministic data type rules, and the original support and normalized support of candidate field values can be calculated after deduplication according to the source lineage group. The normalized support threshold and the leading difference threshold are used to jointly control the single value decision.
[0075] Meanwhile, when the maximum support does not reach the threshold, the top K candidate field values, along with their respective support, fusion confidence, and source tracing sets, are retained to form a multi-hypothesis state that can be further processed.
[0076] It should be noted that the above weights , , and Conflict penalty coefficient η, normalized support threshold Leading Difference Threshold Commonly pointed threshold θ C Both the candidate number K and the task configuration are saved according to the task configuration version.
[0077] In some embodiments, before or during conflict resolution, domain ontology rules, spatiotemporal consistency constraints, or a combination of both can be used to eliminate, downweight, or mark candidate field values that do not meet the constraints.
[0078] To further illustrate this point, a non-limiting numerical example is given below to illustrate the complete processing chain. This does not represent actual performance or recommended configuration.
[0079] Specifically, the "Operating Status" slots for the same fictional target have original records R1, R2, and R3: R1 comes from the direct source SA and points to the upstream record U-01; R2 comes from the direct source SB but also points to U-01, and both give candidate value A; R3 comes from the independent source SC and the upstream record U-02, and gives candidate value B.
[0080] After the three records complete spatiotemporal alignment, target co-reference judgment, and preliminary slot compatibility comparison, respectively, they are labeled with ( , , , The values (0.80, 0.75, 0.90, 0.85), (0.78, 0.74, 0.88, 0.80), and (0.60, 0.50, 0.80, 0.75) are entered into an equal-weighted calculation, resulting in... =0.825、 =0.800、 =0.6625.
[0081] R1 and R2 are grouped into the same source lineage group G1 because they share upstream U-01. The maximum value within the group is 0.825, and they cannot be accumulated repeatedly. R3 belongs to G2. The deterministic enumeration of slot rules determines that A and B are incompatible. When η=0.5, =0.825 - 0.5 × 0.6625 = 0.49375, =0.6625-0.5×0.825=0.25; After normalizing the positive support, Approximately 0.664, Approximately 0.336. If =0.60 and =0.20, then A satisfies both the threshold and the leading difference condition, output A and save the direct sources SA, SB and upstream U-01 in G1, while retaining the conflict relationship between G2, SC and U-02; if any condition is not satisfied, then A and B are retained as multiple hypotheses.
[0082] like Figure 3As shown, target coreference is performed based on aligned standard information units to form a preliminary compatibility matrix; cross-source consistency is calculated. and overall credibility A conflict graph is constructed based on incompatible slots; the original support is calculated after deduplication by the source phylogenetic group. and normalized support The maximum value is determined to be below the normalized support threshold. And the lead difference is lower than the lead difference threshold. If yes, determine the value of a single fusion field and write it into the fusion confidence and source set; otherwise, retain the top K candidate field values and retain the support and source traceability respectively. Subsequently, output the structured state parsing results, which include fusion state elements, confidence, source traceability and optional risk level.
[0083] As can be seen, the target common reference is first executed to form a preliminary slot compatibility matrix, based on which cross-source consistency is calculated. Overall credibility P of standard information unit i Then, a conflict graph is constructed based on the deterministic incompatibility rules; after deduplication of the source phylogenetic groups, the original support is calculated. and normalized support Only the normalized support threshold. Leading Difference Threshold Output a single value if all conditions are met; otherwise, retain multiple hypotheses and source tracing.
[0084] In box 170, the structured state parsing results include at least the fused state elements, fused confidence, and source information for direct and upstream sources; when there are semantic slots where single-value fusion has not been completed, the corresponding multiple hypothesis candidate results are also retained.
[0085] In some embodiments, the above method further includes: The risk level is calculated based on the fusion confidence level, as well as the preset capability, behavior, and spatiotemporal proximity scores.
[0086] In some embodiments, the risk level is an optional downstream field and may not be a necessary output of the main processing chain.
[0087] In some embodiments, for the nth target object or event, its risk level score is... , can be represented as: in, , and These represent the capability, behavior, and spatiotemporal proximity scores used by the optional risk modules, respectively. Indicates the fusion confidence level. , , and Risk weights are optional. They are only calculated if risk assessment is explicitly enabled in the task configuration and the corresponding input fields are complete. When not enabled, the main process still outputs fusion status elements, fusion confidence, and source tracing information.
[0088] In some embodiments, the above method further includes: Input the conflict relationship, the fused standard information unit, and the source tracing set into the preset optional conflict interpretation language model, and generate the conflict interpretation field under the constraints of the preset interpretation output mode.
[0089] In some embodiments, after the conflict resolution is completed, the conflict set, the fused standard information unit, and the source tracing set can be input into a preset optional conflict interpretation language model. Under the constraints of the preset interpretation output mode, a conflict interpretation field is generated to explain the main supporting source of the fused field value and the reason for the conflict resolution.
[0090] The default optional conflict interpretation large language model output uses a field whitelist or structural pattern for validation. If a field is missing, has an incorrect type, exceeds the allowed value range, or cannot be traced back to its input source, it will refuse to write to the standard information unit and will be re-extracted. If re-extraction still fails, the original information fragment and failure status are retained, and the process will proceed to rule extraction or manual verification. The above failure handling does not change the fused field values already formed by the structured trusted adjudication chain.
[0091] It should be noted that the aforementioned pre-defined semantic standardization large language model and pre-defined optional conflict interpretation large language model can be the same large language model or different large language models.
[0092] In this disclosure, the large language model can be used solely to generate standard information units, or it can be used to generate conflict explanation fields after resolving confidence constraint conflicts; regardless of the method used, the fusion field values, fusion confidence, and source tracing set are all stored by a structured processing chain.
[0093] It should be noted that the structured processing chain stores the fusion field, fusion confidence, source set, and unresolved candidate states, and restricts the large language model to the semantic standardization and optional conflict interpretation stages, but does not directly determine the final fusion field by the large language model.
[0094] According to the embodiments of this disclosure, the following technical effects are achieved: First, by using a large language model, heterogeneous raw information is transformed into standard information units, allowing raw information from different sources and with different forms of expression to enter a unified field system. This provides a consistent data foundation for coreference judgment, semantic slot comparison, and fusion processing. This effect is directly generated by the aforementioned spatiotemporal alignment, format normalization, and field constraints.
[0095] Second, by first forming the co-reference results and the preliminary slot compatibility matrix, and then calculating the cross-source consistency and overall reliability and establishing conflict relationships according to deterministic rules, the consistency score and conflict relationship are mutually defined; by limiting the duplicate contribution of homologous replication through aggregation within the source lineage group, the field-level decision reflects both source-level reliability and unit-level quality.
[0096] Third, the output retains the fusion confidence and source tracing set; when the support is insufficient, the corresponding semantic slot is kept in a multi-hypothesis state with candidate support and source tracing relationship, avoiding forcing the slot with insufficient evidence to converge to a single conclusion; in the optional implementation, a conflict explanation field can also be generated, thereby providing a verifiable path for subsequent processing and manual verification.
[0097] To further elaborate on the multi-source information fusion method based on source credibility assessment and cross-source conflict resolution, the following will explain the overall structure and data flow processing.
[0098] like Figure 4 As shown, the process involves acquiring multi-source raw information, including text, image descriptions, sensor records, and human reports; integrating the information with source pointers, acquisition time, location, and uncertainty; performing spatiotemporal alignment and format normalization; constrained invocation of a large language model (used only for semantic standardization) to generate standard information units, thereby structuring fields, source pointers, and extracting confidence levels; and establishing target co-reference, i.e., a preliminary slot compatibility matrix, and cross-source consistency. and overall credibility It calls a large language model (optional conflict interpretation) to construct a conflict graph and resolve fields, as well as generate structured state parsing results, including fused fields, confidence scores, source tracing, and multiple hypotheses.
[0099] It is evident that the large language model is limited to the generation of standard information units and optional conflict interpretation. The final fusion conclusion is still determined by the credibility of the source, conflict relationships, and preset constraints, thus avoiding the large language model from directly summarizing all the original information without boundaries.
[0100] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this disclosure is not limited to the described order of actions, because according to this disclosure, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this disclosure.
[0101] The above is an introduction to the method embodiments. The following describes the solution described in this disclosure further through device embodiments.
[0102] Figure 5 A block diagram of a multi-source information fusion system 500 based on source credibility assessment and cross-source conflict resolution according to an embodiment of the present disclosure is shown. Figure 5 As shown, the system 500 includes: Information access module 510 is used to acquire multi-source information data to be integrated; The spatiotemporal alignment module 520 is used to perform spatiotemporal alignment and format normalization processing on the multi-source information data to be fused based on a unified spatiotemporal benchmark, so as to obtain a set of aligned information fragments. The standard information unit generation module 530 is used to input the information fragment set and the preset field system into the preset semantic standardization large language model to generate multiple standard information units. The co-reference and preliminary compatibility calculation module 540 is used to determine the co-reference of each standard information unit and form a preliminary slot compatibility matrix; The comprehensive credibility assessment module 550 is used to calculate the cross-source consistency and comprehensive credibility of each standard information unit based on the preliminary slot compatibility matrix. The conflict relationship construction and resolution module 560 is used to establish conflict relationships based on deterministic incompatibility rules and the calculated preliminary slot compatibility matrix, and to resolve the conflict relationships at the field level with confidence constraints, and output the fused standard information unit; the fused standard information unit includes at least the fused field value, the fused confidence, and the source traceability set; The state resolution output module 570 is used to generate auditable structured state resolution results based on the fused standard information units.
[0103] The standard information unit generation module performs structured semantic extraction according to a preset field system; the coreference and preliminary compatibility calculation module first determines the groups to be merged and comparable slots, and then calculates cross-source consistency; the comprehensive credibility assessment module forms the comprehensive credibility. The conflict graph construction and resolution module builds edges based on deterministic incompatibility rules and performs field-level adjudication after deduplication of the source lineage group.
[0104] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0105] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0106] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.
[0107] Figure 6 A block diagram of an exemplary electronic device 600 capable of implementing embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.
[0108] Electronic device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in ROM 602 or a computer program loaded into RAM 603 from storage unit 608. RAM 603 may also store various programs and data required for the operation of electronic device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via bus 604. I / O interface 605 is also connected to bus 604.
[0109] Multiple components in electronic device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of displays, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows electronic device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0110] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as method 100. For example, in some embodiments, method 100 may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of method 100 described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to perform method 100 by any other suitable means (e.g., by means of firmware).
[0111] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.
[0112] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0113] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0114] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including voice input, speech input, or tactile input).
[0115] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.
[0116] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.
[0117] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.
[0118] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.
Claims
1. A multi-source information fusion method based on source credibility assessment and cross-source conflict resolution, characterized in that, include: Acquire multi-source information data to be integrated; Based on a unified spatiotemporal reference, the multi-source information data to be fused is spatiotemporally aligned and format normalized to obtain a set of aligned information fragments. The set of information fragments and the preset field system are input into a preset semantically standardized large language model to generate multiple standard information units; Common reference determination is performed on each standard information unit to form a preliminary slot compatibility matrix; Based on the preliminary slot compatibility matrix, the cross-source consistency and overall reliability of each standard information unit are calculated. Based on deterministic incompatibility rules, conflict relationships are established according to the calculated preliminary slot compatibility matrix, and field-level conflict resolution with confidence constraints is performed on the conflict relationships to output fused standard information units; the fused standard information units include at least fused field values, fused confidence, and source tracing sets; Based on the fused standard information units, an auditable structured state parsing result is generated.
2. The method according to claim 1, characterized in that, The multi-source information data to be fused includes at least two information data from distinguishable sources, and the sources belong to different information types, or belong to the same information type but have different source identifiers, acquisition links, or original record pointers.
3. The method according to claim 1, characterized in that, The standard information unit includes at least a structured representation of an entity or event, a relational representation, spatiotemporal elements, a source pointer, a semantic slot field, and a semantic extraction confidence level.
4. The method according to claim 1, characterized in that, The step of determining common reference for each standard information unit to form a preliminary slot compatibility matrix includes: Based on a preset common goal or event, common reference is determined, and each standard information unit is assigned to the corresponding group to be merged; the preset common goal or event includes explicit goals, event identifiers, and goal or event types, degree of overlap in time range, spatial distance, name or attribute similarity; Within the same group to be merged, a preliminary slot compatibility matrix is formed according to data type, unit, and slot rules.
5. The method according to claim 2, characterized in that, The calculation of cross-source consistency and overall reliability of each standard information unit based on the preliminary slot compatibility matrix includes: Based on the preliminary slot compatibility matrix, calculate the cross-source consistency of each standard information unit; The overall credibility is calculated based on the cross-source consistency of each standard information unit, the historical reliability of its source, the timeliness score, and the semantic extraction confidence.
6. The method according to claim 1, characterized in that, The deterministic incompatibility rules include: Enumerated or Boolean slots are incompatible when their values are not equal after the vocabulary mapping is completed. The numerical slots first perform unit conversions, and are incompatible if the difference exceeds the absolute or relative tolerance configured for the task. Time slots are incompatible when intervals do not overlap or the endpoint difference exceeds the time tolerance. Spatial slots are incompatible if their confidence regions do not intersect or the distance exceeds the spatial tolerance after taking positional uncertainty into account. State or behavior slots are incompatible if they violate mutual exclusion rules in the domain ontology.
7. The method according to claim 1, characterized in that, The conflict relationship includes conflict slots, candidate field values involved in the conflict, corresponding standard information units and their source pointers; The field-level conflict resolution that applies credibility constraints to the conflict relationship, and outputs the fused standard information unit, includes: Calculate the original support of the candidate field values involved in the conflict in the conflict relationship; Calculate the normalized support based on the original support; Based on the preset normalized support threshold and the preset leading difference threshold, the fused standard information unit is output according to the normalized support.
8. The method according to claim 1, characterized in that, The method further includes: The risk level is calculated based on the fusion confidence level, as well as preset capability, behavior, and spatiotemporal proximity scores.
9. The method according to any one of claims 1 to 8, characterized in that, The method further includes: The conflict relationship, the fused standard information unit, and the source tracing set are input into a preset optional conflict interpretation language model, and a conflict interpretation field is generated under the constraints of a preset interpretation output mode.
10. A multi-source information fusion system based on source credibility assessment and cross-source conflict resolution, characterized in that, include: The information access module is used to acquire multi-source information data to be integrated; The spatiotemporal alignment module is used to perform spatiotemporal alignment and format normalization on the multi-source information data to be fused based on a unified spatiotemporal benchmark, so as to obtain a set of aligned information fragments. The standard information unit generation module is used to input the set of information fragments and the preset field system into a preset semantic standardization large language model to generate multiple standard information units. The co-reference and preliminary compatibility calculation module is used to determine the co-reference of each standard information unit and form a preliminary slot compatibility matrix; The comprehensive credibility assessment module is used to calculate the cross-source consistency and comprehensive credibility of each standard information unit based on the preliminary slot compatibility matrix. The conflict relationship construction and resolution module is used to establish conflict relationships based on deterministic incompatibility rules and the calculated preliminary slot compatibility matrix, and to perform field-level conflict resolution on the conflict relationships with confidence constraints, and output the fused standard information unit; the fused standard information unit includes at least the fused field value, the fused confidence, and the source tracing set; The state resolution output module is used to generate auditable structured state resolution results based on the fused standard information units.
