Word Representation with Transferable Semantics
- Publication Type:
- Thesis
- Issue Date:
- 2021
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
This thesis is about semantic representation which is a core research problem in text-based machine learning such as natural language processing and information retrieval. The target of this thesis is to improve representation learning methods by utilising transferable semantics extracted from source domains. Specifically, this thesis aims to address four research questions: 1) how to reliably transfer semantics from a structural knowledge base to an unstructured representation space; 2) how to reliably transfer semantics from multiple source domains to a low-resource target domain; 3) how to achieve the reliable and low-cost cross-lingual transfer of semantics; and 4) how to adapt semantic representations for specific applications. To solve these questions, this thesis proposes a set of effective representation methods by exploring and modeling knowledge from 1) knowledge bases; 2) multiple pre-trained embeddings; 3) high-resource languages; and 4) task-related semantics. Comprehensive experiments and case studies have been conducted to evaluate and demonstrate the superior performance of the proposed method compared with baseline methods. To conclude, this thesis proposes a set of effective methods to improve semantic representation by exploring and modeling knowledge beyond raw text and places an emphasis on encoding task-specific features for real-world applications.
Please use this identifier to cite or link to this item: