Enhancing Knowledge Retrieval for Knowledge-Grounded Dialogue with Topic Modeling

Anonymous

Abstract: Knowledge selection is one of the major challenges in building a knowledge-grounded dialogue system. A common method is to use a neural retriever with distributed approximate nearest-neighbor database to quickly find the relevant knowledge sentences. In this work, we propose an approach that utilizes topic modeling on the knowledge base to further improve retrieval accuracy. Experimental results on two datasets show that our model can increase retrieval and generation performance with the correct number of topics chosen. The results also indicate that selecting the right number of topics to segment the knowledge base should be data-dependent and a higher topic coherence of topic modeling does not necessarily lead to better knowledge retrieval performance.

Paper Type: short

Research Area: Information Retrieval and Text Mining