Research on Improving Context Coherence System of Multi-round Dialogue of General Electric Power Model by Reinforcement Learning

View at publisher Summarise with VIRAT
Authors: Tong Wu, Lei Su, Xiangyi Xu, Weiyun Mao, Liang Qiao, Yuan Hong
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

In order to solve the problem of insufficient context coherence in multi-round dialogue of general electric power model, this paper proposes a temporal attention reinforcement learning (TARL) algorithm and supporting system. The algorithm innovatively integrates the dialogue history and electric power domain knowledge to construct the state space, generates responses as actions, designs a multidimensional reward function including context consistency, topic continuity and electric power expertise, and enhances the capture of context temporal dependencies through the temporal attention mechanism. The experiment was conducted on a data set containing 100,000 multi-round dialogue data in the power sector. The simulation results showed that the context coherence score of the TARL algorithm reached 85.6 points, which was 18.4% higher than the Transformer-based dialogue model (72.3 points) and 24.9% higher than the RNN model with attention mechanism (68.5 points). The response quality score was 82.1 points, significantly higher than the comparison algorithm (70.5 points for the Transformer model and 67.8 points for the RNN model). The multi-round dialogue completion rate was 88.3%, which was better than the Transformer model (75.2%) and the RNN model (71.5%). The study confirmed that the TARL algorithm can effectively improve the context coherence and interaction quality of multi-round dialogues in the general large-scale power model, providing reliable technical support for human-computer interaction in the power sector.

Fields of Study & Subjects

Speech and dialogue systems Topic Modeling AI in Service Interactions

Publication details

Journal / source
IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
Pages
1455–1460
Publication date
28 January 2026
Language
EN