Deep learning-based method for generating intelligent dialogues in WeChat using digital pigeons

By building a tree structure, introducing three-dimensional timestamps and emoji emotional information, adopting a dynamic attention mechanism and user portrait features, and optimizing the Transformer model, the problems of contextual coherence and personalization in multi-round dialogue generation are solved, and the generated dialogues are more coherent and personalized.

CN120216641BActive Publication Date: 2025-10-03SICHUAN KUAIDATONG TECHNOLOGY CO LTD
2 Cites 0 Cited by

Patent Information

Application Number
CN202510286072.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-03-12
Publication Date
2025-10-03
Estimated Expiration
2045-03-12

AI Technical Summary

Technical Problem

Existing dialogue generation models have problems with contextual coherence and personalization in multi-round dialogues, making it difficult to maintain dialogue coherence and generate personalized answers based on user profiles.

Method used

By constructing a tree structure to organize multi-round conversation data, introducing three-dimensional timestamps and emoticon emotional information, using a dynamic attention mechanism and user portrait features, combining the Transformer model and Beam Search algorithm, the generation model is optimized to improve the coherence and personalization of conversations.

Benefits of technology

It effectively solves the context coherence problem of multi-round conversations, and the generated conversations are more personalized to meet the needs of different users. The accuracy and coherence of the generated model are further improved through the composite loss function.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

This application belongs to the field of dialogue generation technology, and relates to a deep learning-based digital pigeon WeChat intelligent dialogue generation method; obtains dialogue data and combines it with a public corpus, and performs semantic anomaly detection through regular expressions and BERT models; organizes dialogue data by constructing a tree structure and introducing three-dimensional timestamps; for the emotional information of emoticons, combines the text and visual features of emoticons through a hybrid encoder, adopts a dynamic attention mechanism, weights and adjusts the memory length according to the importance of historical sentences, introduces user portrait features, and then fuses user features with dialogue encoding. In the generation stage, a Transformer-based model and Beam Search algorithm are used, and a diversity penalty term is introduced in the generation process and the generation parameters are dynamically adjusted according to the user portrait, thereby ensuring the diversity and security of the generated dialogue, improving the accuracy and coherence of the dialogue generation model, and effectively solving the technical problems of context coherence and personalized generation.
Need to check novelty before this filing date? Find Prior Art