A multi-source default evidence processing method, device and equipment and a storage medium

Abstract

The application discloses a multi-source default evidence processing method, device and equipment and a storage medium. The method comprises the following steps: acquiring an evidence data set corresponding to a default event; marking an effectiveness level for each piece of evidence data according to a preset evidence effectiveness grading rule; performing multi-source cross verification on the marked evidence data, and adjusting the effectiveness level according to evidence combination; inputting all the evidence into a rule engine for unified analysis and judgment, wherein the unstructured evidence is first converted into structured elements through semantic element extraction; outputting machine review conclusions in two forms, i.e., direct identification and transfer to human review, wherein the transfer to human review is accompanied by identification suggestions for human review and adoption or another conclusion; recording review results and misjudgment feedback, and adjusting processing rule parameters through a backflow mechanism. The application realizes "machine review as the main part and human review as the auxiliary part", guarantees the identification accuracy, significantly reduces the artificial cost, and continuously optimizes through the backflow mechanism.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence and data processing technology, specifically to a method, apparatus, device, and storage medium for processing multi-source breach of contract evidence. Background Technology

[0002] In service breach determination scenarios, evidence of breach originates from various channels such as monitoring, patrols, work orders, and resident complaints. The evidence includes structured data (missing patrol signature records, work order timestamps) and unstructured data (photos, audio recordings, text descriptions). Existing processing methods either involve entirely manual review (high cost and poor timeliness) or direct reliance on a single data source (vulnerable to tampering or unilateral falsification and misleading). How to achieve automated triage while ensuring accuracy, and make the machine-reviewed conclusions interpretable and verifiable, is a pressing technical problem that needs to be solved. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method, device, equipment and storage medium for processing multi-source breach of contract evidence, which realizes "machine review as the main method and human review as the auxiliary method" by classifying the validity of evidence and cross-verifying multiple sources, extracting semantic elements and making unified judgments by rule engines, and continuously optimizing rule parameters by feeding back misjudgments.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for processing multi-source breach of contract evidence includes: acquiring a set of evidence data corresponding to a breach of contract event; labeling each piece of evidence data with a validity level according to a preset evidence validity grading rule; for the same fact to be proven based on a unique event identifier, performing multi-source cross-validation and adjusting the evidence validity level according to a preset evidence combination rule; inputting all evidence into a rule engine for unified analysis and judgment, wherein unstructured evidence is first transformed into structured elements through semantic element extraction; outputting machine review conclusions in two forms: direct determination and transfer to human review, with determination suggestions attached for human review or a separate conclusion when transferred to human review; recording the results of human review and feedback data on misjudgments, and feeding back to adjust rule parameters. Further solutions include: semantic element extraction. This invention also provides corresponding devices, electronic devices, and computer-readable storage media.

[0005] Technical effects: (1) Balance between accuracy and cost - after unstructured evidence is extracted from semantic elements and transformed into structured elements, it is uniformly judged by the rule engine, concentrating human resources on low confidence and insufficient recognition scenarios; (2) Anti-fraud - multi-source cross-verification and single-source downgrade acceptance mechanism suppress fraud from a single data source; (3) Human-machine collaboration can be closed-loop - the machine review conclusion has two forms (direct recognition / transfer to human review) and the transfer to human review is accompanied by recognition suggestions, forming a "machine review + human review" feedback closed loop with the misjudgment return mechanism; (4) Unified standards - the evidence validity grading rules are uniformly executed throughout the database.

[0006] Technical Issues, Technical Means, and Technical Effects: The technical issue is "multi-source heterogeneous evidence of breach of contract, high cost of full manual review, and high risk of misjudgment in full machine review"; the technical means are "gradual labeling of evidence validity + multi-source cross-verification + rule engine and semantic extraction combination based on evidence type"; the technical effects are "concentrating human resources on low-confidence scenarios, suppressing single-source forgery, providing two forms of machine review conclusions with attached identification suggestions for human review, which are interpretable and verifiable". Attached Figure Description

[0007] Figure 1 This is a flowchart of the multi-source breach of contract evidence processing method of the present invention; Figure 2 This is a schematic diagram illustrating the extraction of unstructured evidence semantic elements and the unified judgment of the rule engine in this invention. Detailed Implementation

[0008] Terminology Definitions: Structured evidence refers to evidence existing in the form of structured data that can be directly compared by the rule engine (such as missing signature records in patrol systems, work order acceptance timestamps, and gate logs); Unstructured evidence refers to evidence existing in unstructured forms such as photos, audio / video, and text; Semantic element extraction refers to the process of identifying and extracting elements from unstructured evidence and converting it into structured elements, which can employ artificial intelligence recognition and other methods; Machine review conclusion refers to the form of the conclusion automatically determined and output by the system, including direct determination and transfer to human review; Determination suggestion refers to the determination result of the rule engine as a reference opinion attached to human review when transferring to human review, and the human review may adopt it or give a separate conclusion; Unique event identifier refers to the breach event identifier generated by the merge key composed of the subject identifier, the violation type identifier, and the preset time window; Fact to be proven refers to the matter to be proven corresponding to the same breach fact based on the unique event identifier, used to determine whether multiple sources of evidence point to the same fact.

[0009] Example 1 (Method): As follows Figure 1 As shown.

[0010] S401: Evidence Acquisition. Gather evidence data from monitoring platforms, patrol systems, work order systems, and complaint channels, and link it to the unique event identifier (merge key) of the breach event.

[0011] S402: Validity rating. Ratings are as follows: Strong (system ledgers / fund flow records / surveillance video data / work order timestamps), Medium (timestamped photos / operation records), Weak (unilateral statements, requiring corroboration from at least two independent witnesses).

[0012] S403: Multi-source cross-validation. Multiple sources corroborate the same fact to be verified → upgrade; single source only → downgrade acceptance.

[0013] S404: Unified Judgment. All evidence is input into the rule engine for automatic judgment, where the engine performs a unified analysis of all evidence. Unstructured evidence is first transformed into structured elements through semantic element extraction (time / location / actor / object / type of breach, which can be achieved using artificial intelligence recognition methods) before being input into the rule engine. For example... Figure 2 As shown, unstructured evidence is transformed into structured elements through semantic element extraction and then enters the rule engine for judgment along with structured evidence.

[0014] S405: Conclusion Output. Two levels of conclusions are output according to the mapping rules: If the evidence level is strong and the rule engine's judgment is corroborated, the conclusion is directly output; if the rule engine has a judgment but the evidence level is not strong, the rule engine has no judgment, unstructured evidence cannot be fully structured after semantic element extraction, or the event is marked with a red-line level violation, the conclusion is transferred to human review; during the transfer to human review, the rule engine's judgment is attached as a recommendation for human review, and the human reviewer may adopt the recommendation or provide a separate conclusion; the conclusion transferred to human review enters the decentralization review process.

[0015] S406: Feedback Adjustment. Record the final conclusion given by manual review, the difference between the manual review conclusion and the rule engine's judgment result, and the situations in which the machine review conclusion is rejected or overturned by the parties; adjust the parameters of the evidence validity grading rules or the rule engine based on the above differences and misjudgment feedback data to improve the coverage and accuracy of machine review determination.

[0016] Example 2 (Apparatus): The apparatus described in claim 6.

[0017] Example 3 (Device and Medium): The processor executes a computer program in memory to implement the above method.

Claims

1. A method for processing evidence of multi-source default, characterized in that, include: Obtain the evidence data set corresponding to the breach of contract event, the evidence data set containing multiple pieces of evidence data from different data sources; label the validity level of each piece of evidence data according to a preset evidence validity grading rule: the evidence validity grading rule includes a mapping relationship table between source type code and validity level code; read the source type code of the evidence data, query the mapping relationship table to obtain the corresponding validity level code, and write it into the validity level field of the evidence data; Multi-source cross-validation is performed on evidence data after marking its validity level: For the same fact to be proven based on the unique identifier of the event, when the fact to be proven has multiple pieces of evidence data from different data sources, the validity level of the multiple pieces of evidence data is adjusted according to the preset evidence combination rules; The evidence in the evidence data set is input into the rule engine for automatic judgment, and the rule engine uniformly analyzes and judges all the evidence; wherein, unstructured evidence in the evidence data set is first transformed into structured elements through semantic element extraction before being input into the rule engine; The machine review conclusion is output, which includes two forms: direct determination and transfer to human review; wherein, when the adjusted evidence level is a strong validity level and the rule engine's judgment result is verified, direct determination is output; when the rule engine cannot output a judgment result, or the rule engine outputs a judgment result but the evidence level does not reach a strong validity level, or the unstructured evidence still cannot be fully structured after semantic element extraction, or the breach of contract event is marked as a red-line level violation type, transfer to human review is output; during transfer to human review, the rule engine's judgment result is attached as a determination suggestion to human review, and the human review may adopt the determination suggestion or give a separate conclusion. Record the results of manual review and the feedback data on misjudgments, and adjust the parameters of the evidence validity grading rules or the rule engine based on the feedback data on misjudgments.

2. The method according to claim 1, characterized in that, The rules for classifying the evidentiary validity include: evidence sourced from automatically generated ledger data, fund flow data, surveillance video data, or work order timestamp data is marked as having strong validity; evidence sourced from timestamped photos or operation records is marked as having medium validity; and evidence sourced from unilateral statements is marked as having weak validity, and evidence of weak validity must be corroborated by the statements of at least two independent witnesses.

3. The method according to claim 1, characterized in that, The rules for combining evidence include: if multiple pieces of strong evidence from different sources corroborate each other, the evidence remains at the strong level; if pieces of medium-strength evidence from different sources corroborate each other, the evidence is upgraded to the strong level; and if there is only evidence from a single source, the evidence is downgraded.

4. The method according to claim 1, characterized in that, The semantic element extraction includes time, location, subject of action, object of action, and type of breach of contract, and the extracted elements are associated with the merge key of the breach of contract event for storage.

5. The method according to claim 1, characterized in that, The unique identifier of the event is generated by a merge key composed of the subject identifier, the violation type identifier, and the preset time window.

6. A multi-source breach of contract evidence processing device, characterized in that, include: The evidence acquisition module is used to acquire a set of evidence data corresponding to the breach of contract event. The graded labeling module is used to label the validity level of each piece of evidence data according to preset evidence validity grading rules; The cross-validation module is used to adjust the validity level of evidence data from different data sources based on the evidence combination rules for the same fact to be proved based on the unique identifier of the event. The semantic extraction module is used to extract semantic elements from unstructured evidence to obtain structured elements; The rule determination module is used to input all evidence into the rule engine for automatic determination. The rule engine performs unified analysis and determination. Unstructured evidence is first transformed into structured elements through semantic element extraction before being input into the rule engine. The conclusion output module is used to output machine review conclusions in two forms: direct determination and transfer to human review. When transferring to human review, the judgment result of the rule engine is attached as a determination suggestion to the human reviewer. The feedback and adjustment module is used to record the feedback data of misjudgment and transfer the processing rule parameters back to adjust them.

7. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the method of any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the method described in any one of claims 1 to 5.

Images