Investigating Tree- and Graph-based Neural Networks for Natural Language Processing Applications

In the information explosion era, the ability to automatically extract knowledge and gain insights from diverse linguistic genres has become imperative. Comprehending intricate linguistic expressions constitutes an indispensable facet of artificial intelligence. Deep learning techniques have emerged as powerful tools for classification, relation extraction, semantic similarity measurement, and document summarization, offering the promise of revolutionizing our understanding of these crucial domains. In the dynamic landscape of Natural Language Processing (NLP), the integration of syntactic and semantic elements stands as a pivotal frontier. This investigation delves into incorporating both syntactic and semantic dimensions within NLP applications. By leveraging tree- and graph-based neural networks, this study pioneers a holistic approach that augments language understanding and processing capabilities. Through the fusion of structural and semantic-driven insights, this work tries to explore various NLP applications for two linguistic genres: scientific text, and psycho-linguistic texts. Scientific articles inherently embody a sophisticated framework of information representation, necessitating a depth of background knowledge for comprehension. This requisite background knowledge is gleaned through a meticulous examination of the citations interwoven within the ongoing paper. The objective of this endeavor is to scrutinize the citation linkage task, serving as an avenue for extracting the essential background information imperative for the meticulous analysis of scientific documents. Furthermore, for summarization, the citation network is leveraged to augment the performance of summarization models by furnishing additional contextual underpinnings. Different tree-structured neural networks are systematically explored to discern relations between various biomedical entities within scientific articles, thus contributing to the efficacy of relation extraction tasks. In the contemporary landscape dominated by the proliferation of social media, natural language processing emerges as a potent instrument for psychologists to delve into the analysis of individuals' personality traits. Conventional models, hampered by their incapacity to grapple with extended textual sequences exceeding their token intake limit, encounter limitations. This work propounds innovative solutions through the utilization of tree-structured neural networks and graph attention networks, facilitating the identification of personality traits from protracted written compositions.

Name thesis:degree_name

Ph D

Discipline thesis:degree_discipline

Computer Science

Grantor dc:publisher

The University of Western Ontario

Year dc:date.issued

2024

Author dc:creator

• Singha Roy, Sudipta

Advisor dc:contributor.advisor

• Mercer, Robert E.

• Tree Transformers and Graph Attention Network
• Syntactic Representation
• Semantic Similarity
• Relation Extraction
• Scientific Document Summarization
• Document Classification

Language dc:language.iso

en_ca

Handle dc:identifier.uri

https://hdl.handle.net/20.500.14721/37616

OAI identifier oai:identifier

oai:uwo.scholaris.ca:20.500.14721/37616