WPI - Computer Science Department, PhD Proposal Defense, Yichuan Li "Label Efficient Representation Learning for Text-Centric Tasks"
10:00 am to 11:00 am
WPI – Computer Science Department
Tuesday, November 21st, 2023
Time: 10:00 a.m. – 12:00 p.m.
Location: Campus Center, Morgan room
Zoom Link: https://wpi.zoom.us/j/99582893344
Prof. Kyumin Lee, WPI - Computer Science (Advisor)
Prof. Xiangnan Kong, WPI - Computer Science
Prof. Xiaozhong Liu, WPI - Data Science
Prof. Nima Kordzadeh – Business School (External Member
Prof. Kaize Ding, Northwestern (External Member)
This dissertation addresses a critical challenge in textual representation learning within the field of natural language processing (NLP): the heavy dependence on large volumes of labeled data. Traditional text-only approaches, which primarily rely on text as the primary data source and analytical medium, often face practical limitations in dynamic areas like fake news detection and social media analytics due to the resource-intensive nature of data acquisition and annotation.
To overcome this, the dissertation proposes a paradigm shift from text-only to text-centric methodologies, emphasizing the enrichment of text with auxiliary knowledge from various sources and modalities.
This innovative approach integrates domain-specific knowledge, human expertise, numerical data, and graphical information to augment and contextualize textual data. Such enrichment transforms the text analysis process, enhancing the depth and quality of text-centric models and reducing the reliance on extensive labeled datasets. The dissertation demonstrates how this enriched, multidimensional approach to text representation can lead to more robust, insightful, and efficient learning models in NLP, addressing some of the field's most pressing challenges.