Docalog: Multi-document Dialogue System using Transformer-based Span Retrieval

Sayed Hesam Alavian; Ali Satvaty; Sadra Sabouri; Ehsaneddin Asgari; Hossein Sameti

Docalog: Multi-document Dialogue System using Transformer-based Span Retrieval

Sayed Hesam Alavian, Ali Satvaty, Sadra Sabouri, Ehsaneddin Asgari^*, Hossein Sameti^*

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Information-seeking dialogue systems, including knowledge identification and response generation, aim to respond to users with fluent, coherent, and informative answers based on users' needs. This paper discusses our proposed approach, Docalog, for the DialDoc-22 (Multi-Doc2Dial) shared task. Docalog identifies the most relevant knowledge in the associated document, in a multi-document setting. Docalog, is a three-stage pipeline consisting of (1) a document retriever model (DR. TEIT), (2) an answer span prediction model, and (3) an ultimate span picker deciding on the most likely answer span, out of all predicted spans. In the test phase of MultiDoc2Dial 2022, Docalog achieved f1-scores of 36.07% and 28.44% and SacreBLEU scores of 23.70% and 20.52%, respectively on the MDD-SEEN and MDD-UNSEEN folds.

Original language	English
Title of host publication	DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop
Editors	Song Feng, Hui Wan, Caixia Yuan, Han Yu
Publisher	Association for Computational Linguistics (ACL)
Pages	142-147
Number of pages	6
ISBN (Electronic)	9781955917339
Publication status	Published - 2022
Externally published	Yes
Event	2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, DialDoc 2022 - Dublin, Ireland Duration: 26 May 2022 → …

Publication series

Name	DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop

Conference

Conference	2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, DialDoc 2022
Country/Territory	Ireland
City	Dublin
Period	26/05/22 → …

Cite this

Alavian, S. H., Satvaty, A., Sabouri, S., Asgari, E., & Sameti, H. (2022). Docalog: Multi-document Dialogue System using Transformer-based Span Retrieval. In S. Feng, H. Wan, C. Yuan, & H. Yu (Eds.), DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop (pp. 142-147). (DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop). Association for Computational Linguistics (ACL).

Alavian, Sayed Hesam ; Satvaty, Ali ; Sabouri, Sadra et al. / Docalog : Multi-document Dialogue System using Transformer-based Span Retrieval. DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop. editor / Song Feng ; Hui Wan ; Caixia Yuan ; Han Yu. Association for Computational Linguistics (ACL), 2022. pp. 142-147 (DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop).

@inproceedings{670bb96f0c2240b2b4dcd6cf8ec26b1d,

title = "Docalog: Multi-document Dialogue System using Transformer-based Span Retrieval",

abstract = "Information-seeking dialogue systems, including knowledge identification and response generation, aim to respond to users with fluent, coherent, and informative answers based on users' needs. This paper discusses our proposed approach, Docalog, for the DialDoc-22 (Multi-Doc2Dial) shared task. Docalog identifies the most relevant knowledge in the associated document, in a multi-document setting. Docalog, is a three-stage pipeline consisting of (1) a document retriever model (DR. TEIT), (2) an answer span prediction model, and (3) an ultimate span picker deciding on the most likely answer span, out of all predicted spans. In the test phase of MultiDoc2Dial 2022, Docalog achieved f1-scores of 36.07% and 28.44% and SacreBLEU scores of 23.70% and 20.52%, respectively on the MDD-SEEN and MDD-UNSEEN folds.",

author = "Alavian, {Sayed Hesam} and Ali Satvaty and Sadra Sabouri and Ehsaneddin Asgari and Hossein Sameti",

note = "Publisher Copyright: {\textcopyright} 2022 Association for Computational Linguistics; 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, DialDoc 2022 ; Conference date: 26-05-2022",

year = "2022",

language = "English",

series = "DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop",

publisher = "Association for Computational Linguistics (ACL)",

pages = "142--147",

editor = "Song Feng and Hui Wan and Caixia Yuan and Han Yu",

booktitle = "DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop",

address = "United States",

}

Alavian, SH, Satvaty, A, Sabouri, S, Asgari, E & Sameti, H 2022, Docalog: Multi-document Dialogue System using Transformer-based Span Retrieval. in S Feng, H Wan, C Yuan & H Yu (eds), DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop. DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop, Association for Computational Linguistics (ACL), pp. 142-147, 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, DialDoc 2022, Dublin, Ireland, 26/05/22.

Docalog: Multi-document Dialogue System using Transformer-based Span Retrieval. / Alavian, Sayed Hesam; Satvaty, Ali; Sabouri, Sadra et al.
DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop. ed. / Song Feng; Hui Wan; Caixia Yuan; Han Yu. Association for Computational Linguistics (ACL), 2022. p. 142-147 (DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Docalog

T2 - 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, DialDoc 2022

AU - Alavian, Sayed Hesam

AU - Satvaty, Ali

AU - Sabouri, Sadra

AU - Asgari, Ehsaneddin

AU - Sameti, Hossein

PY - 2022

Y1 - 2022

N2 - Information-seeking dialogue systems, including knowledge identification and response generation, aim to respond to users with fluent, coherent, and informative answers based on users' needs. This paper discusses our proposed approach, Docalog, for the DialDoc-22 (Multi-Doc2Dial) shared task. Docalog identifies the most relevant knowledge in the associated document, in a multi-document setting. Docalog, is a three-stage pipeline consisting of (1) a document retriever model (DR. TEIT), (2) an answer span prediction model, and (3) an ultimate span picker deciding on the most likely answer span, out of all predicted spans. In the test phase of MultiDoc2Dial 2022, Docalog achieved f1-scores of 36.07% and 28.44% and SacreBLEU scores of 23.70% and 20.52%, respectively on the MDD-SEEN and MDD-UNSEEN folds.

AB - Information-seeking dialogue systems, including knowledge identification and response generation, aim to respond to users with fluent, coherent, and informative answers based on users' needs. This paper discusses our proposed approach, Docalog, for the DialDoc-22 (Multi-Doc2Dial) shared task. Docalog identifies the most relevant knowledge in the associated document, in a multi-document setting. Docalog, is a three-stage pipeline consisting of (1) a document retriever model (DR. TEIT), (2) an answer span prediction model, and (3) an ultimate span picker deciding on the most likely answer span, out of all predicted spans. In the test phase of MultiDoc2Dial 2022, Docalog achieved f1-scores of 36.07% and 28.44% and SacreBLEU scores of 23.70% and 20.52%, respectively on the MDD-SEEN and MDD-UNSEEN folds.

UR - http://www.scopus.com/inward/record.url?scp=85137684965&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:85137684965

T3 - DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop

SP - 142

EP - 147

BT - DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop

A2 - Feng, Song

A2 - Wan, Hui

A2 - Yuan, Caixia

A2 - Yu, Han

PB - Association for Computational Linguistics (ACL)

Y2 - 26 May 2022

ER -

Alavian SH, Satvaty A, Sabouri S, Asgari E, Sameti H. Docalog: Multi-document Dialogue System using Transformer-based Span Retrieval. In Feng S, Wan H, Yuan C, Yu H, editors, DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop. Association for Computational Linguistics (ACL). 2022. p. 142-147. (DialDoc 2022 - Proceedings of the 2nd DialDoc Workshop on Document-Grounded Dialogue and Conversational Question Answering, Proceedings of the Workshop).

Docalog: Multi-document Dialogue System using Transformer-based Span Retrieval

Abstract

Publication series

Conference

Other files and links

Fingerprint

Cite this