基于LLM的证券客户画像构建方法及系统
By quantifying abnormal scores and short-term memory windows of customer behavior, and combining subjective confidence and intervention vector components, the securities customer profile generated by LLM is dynamically adjusted, which solves the problem that existing technologies cannot respond to changes in short-term customer preferences in a timely manner, and improves the accuracy and matching degree of customer profiles.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING SECURITIES CO LTD
- Filing Date
- 2026-06-16
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, when building securities customer profiles based on LLM, it is impossible to respond in a timely manner to changes in short-term customer preferences. It is easily affected by short-term emotional fluctuations and noisy behaviors, causing the profile to deviate from the customer's true investment preferences. Furthermore, it is impossible to distinguish between the stability and suddenness of behavior.
By calculating customer time interval characteristics, short-term frequency characteristics, and pattern deviation characteristics, the abnormal scores of new behaviors are quantified, a short-term memory window is established, and the customer profile generated by LLM is corrected using subjective confidence and intervention vector components. A monitoring auxiliary line is constructed to dynamically adjust the labels.
It enables rapid response to changes in customers' short-term transaction patterns and emotions, reduces the impact of noisy behaviors on customer profiles, ensures the accuracy and timely updating of tags, and improves the matching degree of customer profiles.
Smart Images
Figure CN122415142A_ABST