POLYU at TREC 2020 Conversational Assistant Track: Query Reformulation with Heuristic Topic Phrases Discovery

Kaishuai Xu; Wenjie Li; Yongli Li

POLYU at TREC 2020 Conversational Assistant Track: Query Reformulation with Heuristic Topic Phrases Discovery

Kaishuai Xu
, Wenjie Li
, Yongli Li

Research output: Unpublished conference presentation (presented paper, abstract, poster) › Conference presentation (not published in journal/proceeding/book) › Academic research › peer-review

Abstract

This paper demonstrates a 2-stage conversational search architecture for the Conversational Assistant Track in TREC 2020, including the initial rule-based retrieval and BERT-based re-ranking. We propose a straightforward query reformulation method with topic phrases discovery and inheritance. The method can efficiently extract the key phrase as a topic and inherit phrases based on specific rules. Experimental results show that our method performs as well as top-2 teams in CAsT 2019 evaluation datasets (NDCG@3: 0.433) with a simpler query expansion and smaller BERT model.

Original language	English
Publication status	Published - 2020
Event	29th Text REtrieval Conference, TREC 2020 - Virtual, Online, United States Duration: 16 Nov 2020 → 20 Nov 2020

Conference

Conference	29th Text REtrieval Conference, TREC 2020
Country/Territory	United States
City	Virtual, Online
Period	16/11/20 → 20/11/20

Keywords

conversational search
information retrieval
query reformulation

ASJC Scopus subject areas

Language and Linguistics
Computer Science Applications
Linguistics and Language

Cite this

@conference{aafa612ce5284c3c8efca0a4cc538837,

title = "POLYU at TREC 2020 Conversational Assistant Track: Query Reformulation with Heuristic Topic Phrases Discovery",

abstract = "This paper demonstrates a 2-stage conversational search architecture for the Conversational Assistant Track in TREC 2020, including the initial rule-based retrieval and BERT-based re-ranking. We propose a straightforward query reformulation method with topic phrases discovery and inheritance. The method can efficiently extract the key phrase as a topic and inherit phrases based on specific rules. Experimental results show that our method performs as well as top-2 teams in CAsT 2019 evaluation datasets (NDCG@3: 0.433) with a simpler query expansion and smaller BERT model.",

keywords = "conversational search, information retrieval, query reformulation",

author = "Kaishuai Xu and Wenjie Li and Yongli Li",

note = "Publisher Copyright: {\textcopyright} 2020 Association for Computing Machinery.; 29th Text REtrieval Conference, TREC 2020 ; Conference date: 16-11-2020 Through 20-11-2020",

year = "2020",

language = "English",

}

POLYU at TREC 2020 Conversational Assistant Track: Query Reformulation with Heuristic Topic Phrases Discovery. / Xu, Kaishuai; Li, Wenjie; Li, Yongli.
2020. Paper presented at 29th Text REtrieval Conference, TREC 2020, Virtual, Online, United States.

Research output: Unpublished conference presentation (presented paper, abstract, poster) › Conference presentation (not published in journal/proceeding/book) › Academic research › peer-review

TY - CONF

T1 - POLYU at TREC 2020 Conversational Assistant Track: Query Reformulation with Heuristic Topic Phrases Discovery

AU - Xu, Kaishuai

AU - Li, Wenjie

AU - Li, Yongli

PY - 2020

Y1 - 2020

N2 - This paper demonstrates a 2-stage conversational search architecture for the Conversational Assistant Track in TREC 2020, including the initial rule-based retrieval and BERT-based re-ranking. We propose a straightforward query reformulation method with topic phrases discovery and inheritance. The method can efficiently extract the key phrase as a topic and inherit phrases based on specific rules. Experimental results show that our method performs as well as top-2 teams in CAsT 2019 evaluation datasets (NDCG@3: 0.433) with a simpler query expansion and smaller BERT model.

AB - This paper demonstrates a 2-stage conversational search architecture for the Conversational Assistant Track in TREC 2020, including the initial rule-based retrieval and BERT-based re-ranking. We propose a straightforward query reformulation method with topic phrases discovery and inheritance. The method can efficiently extract the key phrase as a topic and inherit phrases based on specific rules. Experimental results show that our method performs as well as top-2 teams in CAsT 2019 evaluation datasets (NDCG@3: 0.433) with a simpler query expansion and smaller BERT model.

KW - conversational search

KW - information retrieval

KW - query reformulation

UR - https://www.scopus.com/pages/publications/85180131223

M3 - Conference presentation (not published in journal/proceeding/book)

AN - SCOPUS:85180131223

T2 - 29th Text REtrieval Conference, TREC 2020

Y2 - 16 November 2020 through 20 November 2020

ER -

POLYU at TREC 2020 Conversational Assistant Track: Query Reformulation with Heuristic Topic Phrases Discovery

Abstract

Conference

Keywords

ASJC Scopus subject areas

Other files and links

Fingerprint

Cite this