System for predicting claims appeal risk

Abstract

This invention utilizes a claims appeal risk prediction system. It compares customer voiceprint information obtained from real-time voice call data with the customer's historical voiceprint baseline to assess their emotional state. Furthermore, it uses case type development trajectories, call waiting times, and call duration to predict the risk of a customer filing an appeal after the call ends. When a risk of appeal is identified, the system can prompt customer service representatives to use appropriate responses to mitigate the risk of being appealed. This reduces the likelihood of appeals and the manpower required to handle subsequent appeal cases.

Claims

1. A system for predicting claims appeal risk, comprising: Storage device, storing multiple modules and multiple historical cases; The system includes a processor coupled to the storage and configured to: acquire real-time voice call data of the customer and customer service representatives; execute a voice separation module to extract the customer's voice and the customer service representative's voice from the real-time voice call data; extract customer voiceprint information from the customer's voice and extract multiple customer historical cases from the historical cases based on the customer's voiceprint information; obtain the case type based on the real-time voice call data; and execute a risk prediction module to predict the claim appeal risk value based on at least one voiceprint feature of the customer's voice, at least one historical voiceprint feature from the customer's historical cases, and the case type.

2. The system for predicting claims appeal risk as described in claim 1, wherein the processor is further configured to: obtain a customer voiceprint baseline value based on the customer's historical cases; and determine that there is a claim risk in response to the difference between the customer voiceprint baseline value and the customer voiceprint information being greater than a threshold.

3. The system for predicting claims appeal risk as described in claim 2, wherein the customer voiceprint baseline value includes a speech rate baseline value and a volume baseline value, wherein the processor is further configured to: determine the existence of the appeal risk in response to the difference between the real-time speech rate in the customer voiceprint information and the speech rate baseline value being greater than a first speech rate threshold, or based on the percentage of the real-time speech rate relative to the speech rate baseline value being greater than a second speech rate threshold; and determine the existence of the appeal risk in response to the difference between the real-time volume in the customer voiceprint information and the volume baseline value being greater than a first volume threshold, or based on the percentage of the real-time volume relative to the volume baseline value being greater than a second volume threshold.

4. The system for predicting claims appeal risk as described in claim 1, wherein the processor is further configured to: obtain multiple relevant historical cases based on the case type; analyze the real-time voice call data and compare the relevant historical cases in real time to obtain multiple similar distances, and obtain the minimum similar distance among the similar distances; and in response to the minimum similar distance being less than a distance threshold, use the relevant historical case corresponding to the minimum similar distance as a trajectory reference for the real-time voice call data.

5. The system for predicting claims appeal risk as described in claim 4, wherein the processor is further configured to: determine that the real-time voice call data has an appeal risk in response to the relevant historical cases, including appeal records, corresponding to the minimum similarity distance.

6. The system for predicting claims appeal risk as described in claim 5, wherein the processor is further configured to: prompt the customer service representative to respond to the customer with suggested scripts to reduce the risk of the appeal.

7. The system for predicting claims appeal risk as described in claim 1, wherein the processor is further configured to: calculate the waiting time for connection in response to the customer making a call; calculate the duration of the call in response to the customer connecting with the customer service representative; increase the risk weighting value in response to the waiting time being greater than a connection time threshold; and increase the risk score in response to the duration of the call being greater than a call time threshold.

8. The system for predicting claims appeal risk as described in claim 7, wherein the processor is further configured to: calculate the claims appeal risk value using the following formula: P(t) = Model(F(t), P(t-1), C), where t is time, P(t) is the claims appeal risk value, P(t-1) is the claims appeal risk value at the previous moment, C is the transcript of the real-time voice call data, F(t) are multiple feature dimensions, Model is a calculation model, wherein the parameters of the calculation model include the risk weighting value and the risk score value, wherein the feature dimensions include the at least one voiceprint feature.

9. The system for predicting claims appeal risk as described in claim 8, wherein the processor is further configured to: obtain the feature dimensions based on the real-time voice call data, wherein the feature dimensions further include voice prosody features, semantic content features, two-way emotion features, and historical context features.

10. The system for predicting claims appeal risk as described in claim 8, wherein the processor is further configured to: tag the instant voice call data in response to the end of an instant call, and add the instant voice call data to the historical cases, wherein the instant voice call data includes a call recording, the transcript, the at least one voiceprint feature, and the claims appeal risk value at each time point during the duration of the call; and train the computational model based on the updated historical cases.

Images