Experimental Comparison between Neural-Symbolic Question-Answering Methods

Hieu Hoang; Triet Nguyen; Nguyen Ho; Dung A. Tran; Van Long Ho; Hien D. Nguyen

doi:10.1109/KSE59128.2023.10299450

Experimental Comparison between Neural-Symbolic Question-Answering Methods

Hieu Hoang, Triet Nguyen, Nguyen Ho, Dung A. Tran, Van Long Ho, Hien D. Nguyen^*

^*Corresponding author for this work

Research output: Contribution to book/anthology/report/conference proceeding › Article in proceeding › Research › peer-review

Abstract

This paper presents an experimental comparison and analysis between two representative methods for neural-symbolic question-answering: the Complex Query Decomposition (CQD) method and the Graph Neural Network Question Embedding (GNN-QE) method. Starting with large and complex queries, CQD breaks down the large query into shorter and simpler sub-queries, thus decomposing the initial query into more manageable components. On the other hand, GNN-QE is a recent architecture for neural-symbolic question-answering that uses graph neural networks to encrypt question structures. GNN-QE portrays questions as graphs to capture the innate links between different question elements, allowing for more complex and comprehensive reasoning. This paper examined the main characteristics of CQD and GNN-QE methods and analyse their advantages and drawbacks towards question-answering problem through popular performance metrics, such as MRR and Hits@K. The results show how each method handles complex queries through the use of symbolic and neural representations, and how well it can produce precise and insightful responses.

Original language	English
Title of host publication	15th International Conference on Knowledge and Systems Engineering, KSE 2023 - Proceedings
Editors	Huynh Thi Thanh Binh, Van Thuc Hoang, Le Minh Nguyen, Sy Vinh Le, Thi Dao Vu, Duy Trung Pham
Publisher	IEEE (Institute of Electrical and Electronics Engineers)
Publication date	2023
Article number	10299450
ISBN (Electronic)	9798350329742
DOIs	https://doi.org/10.1109/KSE59128.2023.10299450
Publication status	Published - 2023
Event	15th International Conference on Knowledge and Systems Engineering, KSE 2023 - Virtual, Online, Viet Nam Duration: 18 Oct 2023 → 20 Oct 2023

Conference

Conference	15th International Conference on Knowledge and Systems Engineering, KSE 2023
Country/Territory	Viet Nam
City	Virtual, Online
Period	18/10/2023 → 20/10/2023
Sponsor	Intelligent Integration (INT2), Secure Metric Technology and Partner Key Factor, Vietnam National Cyber Security Technology Corporation (NCS), Vingroup Innovation Foundation (VINIF)

Series	International Conference on Knowledge and Systems Engineering (KSE)
ISSN	2694-4804

Bibliographical note

Publisher Copyright:
© 2023 IEEE.

Keywords

Complex Query Decomposition
GNN-QE
Graph Neural Networks
Neural Networks
Neural Symbolic
Question Answering
Symbolic Reasoning

Access to Document

10.1109/KSE59128.2023.10299450

AUB Link

Search for the material in Aalborg University Library's search engine

Cite this

Hoang, H., Nguyen, T., Ho, N., Tran, D. A., Ho, V. L., & Nguyen, H. D. (2023). Experimental Comparison between Neural-Symbolic Question-Answering Methods. In H. T. T. Binh, V. T. Hoang, L. M. Nguyen, S. V. Le, T. D. Vu, & D. T. Pham (Eds.), 15th International Conference on Knowledge and Systems Engineering, KSE 2023 - Proceedings Article 10299450 IEEE (Institute of Electrical and Electronics Engineers). https://doi.org/10.1109/KSE59128.2023.10299450

Hoang, Hieu ; Nguyen, Triet ; Ho, Nguyen et al. / Experimental Comparison between Neural-Symbolic Question-Answering Methods. 15th International Conference on Knowledge and Systems Engineering, KSE 2023 - Proceedings. editor / Huynh Thi Thanh Binh ; Van Thuc Hoang ; Le Minh Nguyen ; Sy Vinh Le ; Thi Dao Vu ; Duy Trung Pham. IEEE (Institute of Electrical and Electronics Engineers), 2023. (International Conference on Knowledge and Systems Engineering (KSE)).

@inproceedings{53a8ec5b926044848e06eb4a68ed0a75,

title = "Experimental Comparison between Neural-Symbolic Question-Answering Methods",

abstract = "This paper presents an experimental comparison and analysis between two representative methods for neural-symbolic question-answering: the Complex Query Decomposition (CQD) method and the Graph Neural Network Question Embedding (GNN-QE) method. Starting with large and complex queries, CQD breaks down the large query into shorter and simpler sub-queries, thus decomposing the initial query into more manageable components. On the other hand, GNN-QE is a recent architecture for neural-symbolic question-answering that uses graph neural networks to encrypt question structures. GNN-QE portrays questions as graphs to capture the innate links between different question elements, allowing for more complex and comprehensive reasoning. This paper examined the main characteristics of CQD and GNN-QE methods and analyse their advantages and drawbacks towards question-answering problem through popular performance metrics, such as MRR and Hits@K. The results show how each method handles complex queries through the use of symbolic and neural representations, and how well it can produce precise and insightful responses.",

keywords = "Complex Query Decomposition, GNN-QE, Graph Neural Networks, Neural Networks, Neural Symbolic, Question Answering, Symbolic Reasoning",

author = "Hieu Hoang and Triet Nguyen and Nguyen Ho and Tran, {Dung A.} and Ho, {Van Long} and Nguyen, {Hien D.}",

note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; 15th International Conference on Knowledge and Systems Engineering, KSE 2023 ; Conference date: 18-10-2023 Through 20-10-2023",

year = "2023",

doi = "10.1109/KSE59128.2023.10299450",

language = "English",

series = "International Conference on Knowledge and Systems Engineering (KSE)",

publisher = "IEEE (Institute of Electrical and Electronics Engineers)",

editor = "Binh, {Huynh Thi Thanh} and Hoang, {Van Thuc} and Nguyen, {Le Minh} and Le, {Sy Vinh} and Vu, {Thi Dao} and Pham, {Duy Trung}",

booktitle = "15th International Conference on Knowledge and Systems Engineering, KSE 2023 - Proceedings",

address = "United States",

}

Hoang, H, Nguyen, T, Ho, N, Tran, DA, Ho, VL & Nguyen, HD 2023, Experimental Comparison between Neural-Symbolic Question-Answering Methods. in HTT Binh, VT Hoang, LM Nguyen, SV Le, TD Vu & DT Pham (eds), 15th International Conference on Knowledge and Systems Engineering, KSE 2023 - Proceedings., 10299450, IEEE (Institute of Electrical and Electronics Engineers), International Conference on Knowledge and Systems Engineering (KSE), 15th International Conference on Knowledge and Systems Engineering, KSE 2023, Virtual, Online, Viet Nam, 18/10/2023. https://doi.org/10.1109/KSE59128.2023.10299450

Experimental Comparison between Neural-Symbolic Question-Answering Methods. / Hoang, Hieu; Nguyen, Triet; Ho, Nguyen et al.
15th International Conference on Knowledge and Systems Engineering, KSE 2023 - Proceedings. ed. / Huynh Thi Thanh Binh; Van Thuc Hoang; Le Minh Nguyen; Sy Vinh Le; Thi Dao Vu; Duy Trung Pham. IEEE (Institute of Electrical and Electronics Engineers), 2023. 10299450 (International Conference on Knowledge and Systems Engineering (KSE)).

Research output: Contribution to book/anthology/report/conference proceeding › Article in proceeding › Research › peer-review

TY - GEN

T1 - Experimental Comparison between Neural-Symbolic Question-Answering Methods

AU - Hoang, Hieu

AU - Nguyen, Triet

AU - Ho, Nguyen

AU - Tran, Dung A.

AU - Ho, Van Long

AU - Nguyen, Hien D.

PY - 2023

Y1 - 2023

N2 - This paper presents an experimental comparison and analysis between two representative methods for neural-symbolic question-answering: the Complex Query Decomposition (CQD) method and the Graph Neural Network Question Embedding (GNN-QE) method. Starting with large and complex queries, CQD breaks down the large query into shorter and simpler sub-queries, thus decomposing the initial query into more manageable components. On the other hand, GNN-QE is a recent architecture for neural-symbolic question-answering that uses graph neural networks to encrypt question structures. GNN-QE portrays questions as graphs to capture the innate links between different question elements, allowing for more complex and comprehensive reasoning. This paper examined the main characteristics of CQD and GNN-QE methods and analyse their advantages and drawbacks towards question-answering problem through popular performance metrics, such as MRR and Hits@K. The results show how each method handles complex queries through the use of symbolic and neural representations, and how well it can produce precise and insightful responses.

AB - This paper presents an experimental comparison and analysis between two representative methods for neural-symbolic question-answering: the Complex Query Decomposition (CQD) method and the Graph Neural Network Question Embedding (GNN-QE) method. Starting with large and complex queries, CQD breaks down the large query into shorter and simpler sub-queries, thus decomposing the initial query into more manageable components. On the other hand, GNN-QE is a recent architecture for neural-symbolic question-answering that uses graph neural networks to encrypt question structures. GNN-QE portrays questions as graphs to capture the innate links between different question elements, allowing for more complex and comprehensive reasoning. This paper examined the main characteristics of CQD and GNN-QE methods and analyse their advantages and drawbacks towards question-answering problem through popular performance metrics, such as MRR and Hits@K. The results show how each method handles complex queries through the use of symbolic and neural representations, and how well it can produce precise and insightful responses.

KW - Complex Query Decomposition

KW - GNN-QE

KW - Graph Neural Networks

KW - Neural Networks

KW - Neural Symbolic

KW - Question Answering

KW - Symbolic Reasoning

UR - http://www.scopus.com/inward/record.url?scp=85178510023&partnerID=8YFLogxK

U2 - 10.1109/KSE59128.2023.10299450

DO - 10.1109/KSE59128.2023.10299450

M3 - Article in proceeding

AN - SCOPUS:85178510023

T3 - International Conference on Knowledge and Systems Engineering (KSE)

BT - 15th International Conference on Knowledge and Systems Engineering, KSE 2023 - Proceedings

A2 - Binh, Huynh Thi Thanh

A2 - Hoang, Van Thuc

A2 - Nguyen, Le Minh

A2 - Le, Sy Vinh

A2 - Vu, Thi Dao

A2 - Pham, Duy Trung

PB - IEEE (Institute of Electrical and Electronics Engineers)

T2 - 15th International Conference on Knowledge and Systems Engineering, KSE 2023

Y2 - 18 October 2023 through 20 October 2023

ER -

Hoang H, Nguyen T, Ho N, Tran DA, Ho VL, Nguyen HD. Experimental Comparison between Neural-Symbolic Question-Answering Methods. In Binh HTT, Hoang VT, Nguyen LM, Le SV, Vu TD, Pham DT, editors, 15th International Conference on Knowledge and Systems Engineering, KSE 2023 - Proceedings. IEEE (Institute of Electrical and Electronics Engineers). 2023. 10299450. (International Conference on Knowledge and Systems Engineering (KSE)). doi: 10.1109/KSE59128.2023.10299450

Experimental Comparison between Neural-Symbolic Question-Answering Methods

Abstract

Conference

Bibliographical note

Keywords

Access to Document

AUB Link

Other files and links

Fingerprint

Cite this