Advanced Multi-Graph Architectures for Natural Language Understanding

Xing, Bowen

Advanced Multi-Graph Architectures for Natural Language Understanding

Xing, Bowen

Permalink

Publication Type:: Thesis
Issue Date:: 2023

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (5.46 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Xing, Bowen
dc.date.accessioned	2024-04-08T23:32:58Z
dc.date.available	2024-04-08T23:32:58Z
dc.date.issued	2023
dc.identifier.uri	http://hdl.handle.net/10453/177604
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	Natural language processing (NLP) is a challenge while important subfield of artificial intelligence (AI). And natural language understanding (NLU) is a crucial component of NLP. In recent years, graph structures and graph neural networks are widely utilized in NLU. In single task scenarios, graph structures can represent the sentence or document in graphs (e.g. syntax graph of a sentence). And sometimes only one graph cannot represent sufficient information, therefore multiple graph structures are proposed to sufficiently represent the relations and dependencies. In multi-task scenarios, the interactions among the multiple tasks can be represented in a multi-task graph. And different kinds of graph neural networks have been proposed to achieve information aggregation on the graphs. In this thesis, the multi-graph structures for single tasks, the multi-task graph structures for multi-task learning, and the graph neural networks working on them are collectively referred to as multi-graph architectures. In this thesis, we design multi-graph architectures based on interconnected graphs to represent diverse linguistic dimensions and multi-task interactions, including syntactic, semantic, and contextual information. Each graph encapsulates specific linguistic features, fostering a more comprehensive understanding of language nuances. The fusion of these graphs enables the model to capture intricate relationships and dependencies among words, concepts, sentences and different tasks, contributing to a more robust and context-aware NLU system. This thesis explores the design principles, implementation details, and experimental results of multi-graph architectures applied to various NLU tasks, such as aspect sentiment classification, joint dialog sentiment classification and act recognition, and joint multiple intent detection and slot filling. Comparative analyses against state-of-the-art models demonstrate the efficacy of the proposed multi-graph architectures in handling ambiguous language constructs and improving overall NLU performance in both single-task and multi-task scenarios.	en_US.UTF-8
dc.format	Thesis (PhD)
dc.language.iso	en_US	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/177604/1/thesis.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2023 Bowen Xing
dc.rights	au.edu.uts.lib/cph
dc.title	Advanced Multi-Graph Architectures for Natural Language Understanding	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

Natural language processing (NLP) is a challenge while important subfield of artificial intelligence (AI). And natural language understanding (NLU) is a crucial component of NLP. In recent years, graph structures and graph neural networks are widely utilized in NLU. In single task scenarios, graph structures can represent the sentence or document in graphs (e.g. syntax graph of a sentence). And sometimes only one graph cannot represent sufficient information, therefore multiple graph structures are proposed to sufficiently represent the relations and dependencies. In multi-task scenarios, the interactions among the multiple tasks can be represented in a multi-task graph. And different kinds of graph neural networks have been proposed to achieve information aggregation on the graphs. In this thesis, the multi-graph structures for single tasks, the multi-task graph structures for multi-task learning, and the graph neural networks working on them are collectively referred to as multi-graph architectures. In this thesis, we design multi-graph architectures based on interconnected graphs to represent diverse linguistic dimensions and multi-task interactions, including syntactic, semantic, and contextual information. Each graph encapsulates specific linguistic features, fostering a more comprehensive understanding of language nuances. The fusion of these graphs enables the model to capture intricate relationships and dependencies among words, concepts, sentences and different tasks, contributing to a more robust and context-aware NLU system. This thesis explores the design principles, implementation details, and experimental results of multi-graph architectures applied to various NLU tasks, such as aspect sentiment classification, joint dialog sentiment classification and act recognition, and joint multiple intent detection and slot filling. Comparative analyses against state-of-the-art models demonstrate the efficacy of the proposed multi-graph architectures in handling ambiguous language constructs and improving overall NLU performance in both single-task and multi-task scenarios.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/177604