CN-121998774-A - Intelligent claim settlement processing method and system based on large model-decision tree co-evolution

CN121998774ACN 121998774 ACN121998774 ACN 121998774ACN-121998774-A

Abstract

The application relates to the technical field of claim settlement processing and discloses an intelligent claim settlement processing method and system based on large model-decision tree co-evolution, wherein the method comprises the steps of analyzing historical claim settlement cases based on a large language model, extracting claim settlement information to generate a decision rule set and constructing a decision tree; acquiring a new claim case, extracting case feature vectors, traversing the case feature vectors from a root node of a decision tree to a terminal node, outputting a claim settlement decision result and evaluation confidence level, comparing the evaluation confidence level with a preset confidence level threshold value to determine whether a manual auditing result is obtained, comparing the claim settlement decision result with the manual auditing result, determining a new decision rule when the comparison result is different, and integrating the new decision rule into the decision tree in an increment mode. The application shortens the processing period of the claim settlement, reduces the proportion of manual participation and improves the overall processing efficiency of the claim settlement business.

Inventors

CHANG BO
WANG BUYI
MIAO WEI
SUN ZHAOMIN
MA JIE
XING XIN
ZHU JIE
SHEN YIHUI

Assignees

南京市智慧医疗投资运营服务有限公司

Dates

Publication Date: 20260508
Application Date: 20260112

Claims (10)

1. An intelligent claim settlement processing method based on large model-decision tree co-evolution is characterized by comprising the following steps: Analyzing historical claim settlement cases based on a large language model, extracting claim settlement information and generating a decision rule set; comparing the case feature vector with the condition threshold value of each decision node of the decision tree from the root node of the decision tree, traversing the case feature vector along a matching path to a final node, and outputting a claim settlement decision result; comparing the estimated confidence coefficient with a preset confidence coefficient threshold value, if the estimated confidence coefficient is larger than or equal to the confidence coefficient threshold value, passing through the new claim case, if the estimated confidence coefficient is smaller than the confidence coefficient threshold value, performing manual auditing on the new claim case, and determining a manual auditing result; And when the comparison result is different, carrying out multidimensional clustering analysis and semantic induction on the new claim settlement case based on the case feature vector and the decision path log, determining a new decision rule, and integrating the new decision rule increment into the decision tree.
2. The method for intelligent claim processing based on large model-decision tree co-evolution according to claim 1, wherein when analyzing historical claim cases based on a large language model, extracting claim information and generating a decision rule set, comprising: Based on the large language model, carrying out semantic understanding on historical claim case, extracting claim information fields, wherein the claim information fields comprise diagnosis codes, treatment modes, cost details and responsibility attribution; Inputting the claim information field and the structured prompt template into the large language model, analyzing a correlation mode between input features and decision results, and generating decision condition sentences and corresponding result judgment parameters, wherein the decision condition sentences comprise feature condition combinations and logic operators, and the result judgment parameters comprise claim settlement decision types and claim payment calculation parameters; Performing multi-round iterative optimization on the decision condition sentences, analyzing characteristic conditions, logic relations and threshold parameters in the decision condition sentences, converting the characteristic conditions, the logic relations and the threshold parameters into conditional expressions, constructing a priority system of decision rules based on the logic dependency relations among the decision condition sentences, and organizing the conditional expressions, result judgment parameters and the priority system into the decision rule set.
3. The intelligent claim processing method based on large model-decision tree co-evolution according to claim 2, wherein when constructing a decision tree according to the decision rule set, comprising: The method comprises the steps of sorting conditional expressions according to condition complexity and characteristic correlation, determining a node construction sequence of a decision tree, solving rule conflict according to a priority system, distributing node weights for each conditional expression, converting the sorted conditional expressions into decision tree nodes, mapping characteristic conditions, logic relations and threshold parameters in the conditional expressions into condition judgment codes, optimizing a branch structure of the decision tree according to information gain or a coefficient of a characteristic condition, organizing the decision tree nodes according to a hierarchical relation to form a tree structure with root nodes, internal decision nodes and termination nodes, adding result judgment parameters into the termination nodes, and constructing the decision tree.
4. The intelligent claim processing method based on large model-decision tree co-evolution according to claim 3, wherein when acquiring new claim cases and extracting case feature vectors, the method comprises: The method comprises the steps of obtaining claim application materials of new claim cases, wherein the claim application materials comprise medical notes, diagnosis certificates and bill of fees, carrying out optical character recognition processing on the claim application materials to generate structured text data, carrying out data cleaning on the structured text data to remove noise data and fill missing values, extracting diagnosis codes, treatment mode types, expense detail types, consultation mechanism grades, material integrity scores and OCR recognition confidence degrees from the structured text data based on preset feature extraction rules, and carrying out coding according to predefined vector dimensions to generate case feature vectors.
5. The method for intelligent claim processing based on large model-decision tree co-evolution according to claim 4, wherein determining the evaluation confidence based on the decision path length, the node matching accuracy, and the history verification accuracy comprises: Determining the length of a decision path according to the number of nodes in the decision path, carrying out weighted average on the matching degree of all the nodes in the decision path to determine the node matching precision, counting the correct decision proportion of the decision path in a historical claim case, determining the historical verification accuracy, carrying out comprehensive calculation on the length of the decision path, the node matching precision and the historical verification accuracy, and outputting the evaluation confidence.
6. The intelligent claim processing method based on large model-decision tree co-evolution according to claim 5, the method is characterized in that when comparing the claim settlement decision result with the manual auditing result, the method comprises the following steps: And respectively carrying out structural analysis on the claim settlement decision result and the manual auditing result, extracting comparison elements, wherein the comparison elements comprise decision types, pay amounts, responsibility ratios and claim rejected reasons, and carrying out consistency comparison on the claim settlement decision result and the manual auditing result based on the comparison elements and generating comparison results.
7. The intelligent claim processing method based on large model-decision tree co-evolution according to claim 6, wherein performing multidimensional clustering analysis and semantic generalization on the new claim cases based on the case feature vectors and decision path logs, determining new decision rules, and integrating the new decision rules into the decision tree in increment, comprises: Integrating the case feature vector of the new claim case with the error case feature vector in a historical error case database to generate a feature vector set, wherein the historical error case database stores feature vectors and decision path logs of claim cases which are judged to have substantial differences previously; And grouping the feature vector sets based on a density clustering algorithm, and grouping cases with similar feature patterns into the same cluster to form three error pattern clusters, namely a bill identification error cluster, a disease responsibility judgment fuzzy cluster and a cost calculation deviation cluster.
8. The intelligent claim processing method based on large model-decision tree co-evolution according to claim 7, wherein when performing multidimensional clustering analysis and semantic generalization on the new claim cases based on the case feature vectors and decision path logs, determining new decision rules and incrementally integrating the new decision rules into the decision tree, further comprising: Extracting common features and difference features of cases in each error mode cluster, inputting the common features and the difference features into the large language model, analyzing root causes generated by errors based on the large language model, generating error type description, wherein the error type description comprises error triggering conditions, typical feature expression, influence ranges and severity, and performing associated mapping on the error type description and uncovered nodes or error branches of a decision tree to construct an error mode knowledge base.
9. The intelligent claim processing method based on large model-decision tree co-evolution according to claim 8, wherein when performing multidimensional clustering analysis and semantic generalization on the new claim cases based on the case feature vectors and decision path logs, determining new decision rules and incrementally integrating the new decision rules into the decision tree, further comprising: Generating a new decision condition statement and a new result judgment parameter according to the error mode knowledge base and through the large language model, converting the new decision condition statement into a new condition expression, and integrating the new condition expression and the new result judgment parameter into a decision tree.
10. An intelligent claim processing system based on large model-decision tree co-evolution, for applying the intelligent claim processing method based on large model-decision tree co-evolution according to any one of claims 1 to 9, comprising: the construction unit is configured to analyze historical claim settlement cases based on the large language model, extract claim settlement information and generate a decision rule set; The system comprises an output unit, a case feature vector, a comparison unit and a comparison unit, wherein the output unit is configured to acquire a new case of claim settlement and extract a case feature vector, the case feature vector starts from a root node of the decision tree, compares the case feature vector with a condition threshold value of each decision node of the decision tree, traverses the case feature vector along a matching path to a final node, and outputs a result of the claim settlement decision; The judging unit is configured to compare the estimated confidence coefficient with a preset confidence coefficient threshold value, pass through the new claim case if the estimated confidence coefficient is larger than or equal to the confidence coefficient threshold value, manually audit the new claim case if the estimated confidence coefficient is smaller than the confidence coefficient threshold value, and determine a manual audit result; And when the comparison result is different, carrying out multidimensional clustering analysis and semantic induction on the new claim settlement case based on the case feature vector and the decision path log, determining a new decision rule, and integrating the new decision rule increment into the decision tree.

Description

Intelligent claim settlement processing method and system based on large model-decision tree co-evolution Technical Field The invention relates to the technical field of claim settlement processing, in particular to an intelligent claim settlement processing method and system based on large model-decision tree co-evolution. Background With the continuous improvement of informatization and intellectualization levels of the insurance industry, the claim settlement traffic of dangerous seeds such as medical health risks, accidental risks and the like presents a rapid growth trend. The claim processing is used as a key link in the insurance business chain, and directly relates to the operation efficiency, risk control capability and user service experience of an insurance company. However, the existing claim processing method still has a plurality of defects in practical application, and a more intelligent, efficient and sustainable evolution technical scheme is needed to be improved. At present, the claim settlement processing mainly adopts a mode of combining manual auditing with automatic auditing of a rule engine. On the one hand, manual auditing relies on experience judgment of claim settlement staff on medical policies, insurance clauses and actual cases, and has the problems of low processing efficiency, high labor cost, large subjective difference and difficulty in large scale although the flexibility is high, on the other hand, an automatic claim settlement system based on a rule engine usually relies on fixed rules configured manually in advance, has a certain effect on cases with higher structural degree and relatively stable scenes, but easily has the problems of insufficient rule coverage and increased misjudgment rate when facing actual claim settlement scenes with complicated medical behaviors, diversification and uneven material quality. In the prior art, machine learning and deep learning techniques are gradually applied to the field of intelligent claim settlement, for example, predicting the result of claim settlement by using a classification model or a neural network. Although the method improves the automation level to a certain extent, the method has the following defects that the model is of a black box structure, the decision process is lack of interpretability, the interpretation requirements of the insurance industry in compliance audit, responsibility identification and dispute processing are difficult to meet, the model is usually trained in an off-line mode by depending on large-scale labeling data, when the medical policy, insurance clauses or data distribution changes, the model updating cost is high, the adaptability is insufficient, the model is difficult to effectively integrate manual auditing experience, and structural summarization and continuous optimization of wrong decisions cannot be performed. Therefore, there is a need to design an intelligent claim processing method and system based on large model-decision tree co-evolution to solve the problems in the prior art. Disclosure of Invention In view of the above, the invention provides an intelligent claim settlement processing method and system based on large model-decision tree co-evolution, which aims to solve the problems that artificial auditing experience is difficult to fuse and structural summarization and continuous optimization of wrong decisions cannot be carried out. In one aspect, the invention provides an intelligent claim settlement processing method based on large model-decision tree co-evolution, which comprises the following steps: Analyzing historical claim settlement cases based on a large language model, extracting claim settlement information and generating a decision rule set; comparing the case feature vector with the condition threshold value of each decision node of the decision tree from the root node of the decision tree, traversing the case feature vector along a matching path to a final node, and outputting a claim settlement decision result; comparing the estimated confidence coefficient with a preset confidence coefficient threshold value, if the estimated confidence coefficient is larger than or equal to the confidence coefficient threshold value, passing through the new claim case, if the estimated confidence coefficient is smaller than the confidence coefficient threshold value, performing manual auditing on the new claim case, and determining a manual auditing result; And when the comparison result is different, carrying out multidimensional clustering analysis and semantic induction on the new claim settlement case based on the case feature vector and the decision path log, determining a new decision rule, and integrating the new decision rule increment into the decision tree. Further, analyzing historical claim cases based on the large language model, and when extracting claim information and generating a decision rule set, the method comprises the following steps: Based on the large