CONFERENCE (INTERNATIONAL) TKB48 at TREC 2021 Conversational Assistance Track
Yubo Fang (University of Tsukuba), Hideo Joho (University of Tsukuba), Sumio Fujita
TEXT RETRIEVAL CONFERENCE 2021 (TREC 2021)
November 15, 2021
In this paper, we present our methods and submitted runs for the TREC Conversational Assistance Track of Y3. We incorporate dense retrieval methods into conversational task and leverages a bi-encoder stucture to encode user’s utterance together with the conversation context and each document of the corpus into dnese vector representation and compute their relevant score by computing dot product of the dense vector. Our submitted four runs show the effectiveness of dense retrieval methods compare to traditional sparse retrieval.
Paper : TKB48 at TREC 2021 Conversational Assistance Track (external link)