CN-121981835-A - Insurance business interpretability processing method and apparatus

CN121981835ACN 121981835 ACN121981835 ACN 121981835ACN-121981835-A

Abstract

The embodiment of the application provides an insurance business interpretability processing method and device, wherein the method comprises the steps of obtaining rejection probability output by an insurance claim model, calling an interpretability tool of the insurance claim model to generate a key subgraph when the rejection probability is greater than or equal to a preset rejection threshold, wherein the key subgraph comprises side semantic information and side importance scores, searching term nodes matched with the side semantic information in a pre-built insurance knowledge graph to obtain a semantic matching result of the side and the terms, and determining a rejection text according to the semantic matching result. The application can automatically generate the insurance rejection reason in real time, reduce the manual workload, improve the claim settlement efficiency and meet the related supervision requirements.

Inventors

ZHANG SHUAI
ZHANG XINYUE
WANG ZHENYU

Assignees

北京中电普华信息技术有限公司

Dates

Publication Date: 20260505
Application Date: 20251202

Claims (10)

1. A method for processing the interpretability of insurance service, comprising: Obtaining rejection probability output by an insurance claim settlement model; when the rejection probability is greater than or equal to a preset rejection threshold, invoking an interpretability tool of an insurance claim model to generate a key subgraph, wherein the key subgraph comprises semantic information of edges and edge importance scores; Searching clause nodes matched with semantic information of each side in a pre-constructed insurance knowledge graph to obtain a semantic matching result of the side and the clause; and determining the rejection reason text according to the semantic matching result.
2. The method of claim 1, wherein retrieving term nodes matching semantic information of each edge in a pre-constructed insurance knowledge graph comprises: And matching semantic information of each side with each clause node of the insurance knowledge graph by adopting a maximum bipartite graph matching method.
3. The method of claim 2, wherein the objective function of the maximum bipartite graph matching method is: (1) Where E' is the edge set of the key subgraph, w e is the importance score of edge E, pi (E) is the clause node matching edge E, σ (E, pi (E)) is the semantic relevance of the semantic information of edge E to the matching clause node.
4. The method of claim 1, wherein said determining the repudiated text based on the semantic matching result comprises: inquiring a preset clause template list according to the matched clause nodes, and determining a clause template corresponding to the clause nodes; And generating the rejection reason text according to the clause module.
5. The method as recited in claim 4, further comprising: And constructing and updating the clause template list according to the insurance knowledge graph.
6. An insurance business interpretability processing device, comprising: The acquisition module is used for acquiring the rejection probability output by the insurance claim settlement model; The calling module is used for calling an interpretability tool of the insurance claim model to generate a key sub-graph when the rejection probability is greater than or equal to a preset rejection threshold, wherein the key sub-graph comprises semantic information of edges and edge importance scores; The matching module is used for searching the clause nodes matched with the semantic information of each side in the pre-constructed insurance knowledge graph to obtain the semantic matching result of the side and the clause; And the interpretation module is used for determining the rejection text according to the semantic matching result.
7. The apparatus of claim 6, wherein the device comprises a plurality of sensors, The matching module is used for matching the semantic information of each side with each clause node of the insurance knowledge graph by adopting a maximum bipartite graph matching method.
8. The apparatus of claim 7, wherein the objective function of the maximum bipartite graph matching method is: (1) Where E' is the edge set of the key subgraph, w e is the importance score of edge E, pi (E) is the clause node matching edge E, σ (E, pi (E)) is the semantic relevance of the semantic information of edge E to the matching clause node.
9. The apparatus of claim 6, wherein the device comprises a plurality of sensors, The interpretation module is used for inquiring a preset clause template list according to the matched clause nodes, determining a clause template corresponding to the clause nodes, and generating the rejection reason text according to the clause module.
10. The apparatus as recited in claim 9, further comprising: And the list maintenance module is used for constructing and updating the clause template list according to the insurance knowledge graph.

Description

Insurance business interpretability processing method and apparatus Technical Field The embodiment of the application relates to the technical field of intelligent insurance claims, in particular to an insurance business interpretability processing method and device. Background The existing insurance claim settlement system commonly introduces a graphic neural network to carry out fraud recognition, and obviously improves the rejection accuracy. However, the graphic neural network belongs to a black box model, only can provide a prediction result of whether to reject, cannot give a specific reject reason, and is difficult to meet the related supervision requirements. In order to realize the interpretability of the rejection reason, the rejection reason can be written only by manual postscript, has low efficiency and can not meet the real-time requirement on line. Disclosure of Invention Accordingly, an objective of the embodiments of the present application is to provide a method and a device for processing the interpretability of insurance service. Based on the above object, an embodiment of the present application provides an insurance business interpretability processing method, including: Obtaining rejection probability output by an insurance claim settlement model; when the rejection probability is greater than or equal to a preset rejection threshold, invoking an interpretability tool of an insurance claim model to generate a key subgraph, wherein the key subgraph comprises semantic information of edges and edge importance scores; Searching clause nodes matched with semantic information of each side in a pre-constructed insurance knowledge graph to obtain a semantic matching result of the side and the clause; and determining the rejection reason text according to the semantic matching result. Optionally, retrieving, in a pre-constructed insurance knowledge graph, a clause node matched with semantic information of each side, including: And matching semantic information of each side with each clause node of the insurance knowledge graph by adopting a maximum bipartite graph matching method. Optionally, the objective function of the maximum bipartite graph matching method is as follows: (1) Where E' is the edge set of the key subgraph, w e is the importance score of edge E, pi (E) is the clause node matching edge E, σ (E, pi (E)) is the semantic relevance of the semantic information of edge E to the matching clause node. Optionally, the determining the rejection text according to the semantic matching result includes: inquiring a preset clause template list according to the matched clause nodes, and determining a clause template corresponding to the clause nodes; And generating the rejection reason text according to the clause module. Optionally, the method further comprises: And constructing and updating the clause template list according to the insurance knowledge graph. The embodiment of the application also provides an insurance business interpretability processing device, which comprises: The acquisition module is used for acquiring the rejection probability output by the insurance claim settlement model; The calling module is used for calling an interpretability tool of the insurance claim model to generate a key sub-graph when the rejection probability is greater than or equal to a preset rejection threshold, wherein the key sub-graph comprises semantic information of edges and edge importance scores; The matching module is used for searching the clause nodes matched with the semantic information of each side in the pre-constructed insurance knowledge graph to obtain the semantic matching result of the side and the clause; And the interpretation module is used for determining the rejection text according to the semantic matching result. Optionally, the matching module is configured to match semantic information of each side with each term node of the insurance knowledge graph by using a maximum bipartite graph matching method. Optionally, the objective function of the maximum bipartite graph matching method is as follows: (1) Where E' is the edge set of the key subgraph, w e is the importance score of edge E, pi (E) is the clause node matching edge E, σ (E, pi (E)) is the semantic relevance of the semantic information of edge E to the matching clause node. Optionally, the interpretation module is configured to query a preset clause template list according to the matched clause nodes, determine a clause template corresponding to the clause nodes, and generate the repudiation text according to the clause module. Optionally, the apparatus further includes: And the list maintenance module is used for constructing and updating the clause template list according to the insurance knowledge graph. As can be seen from the above description, the method and the device for processing the interpretability of the insurance service provided by the embodiments of the present application obtain the rejection proba