Neural Knowledge Transfer in Low-Resource Sentiment Analysis

PhD Thesis Proposal Defence


Title: "Neural Knowledge Transfer in Low-Resource Sentiment Analysis"

by

Mr. Zheng LI


Abstract:

Sentiment analysis (SA), aiming to automatically characterize human beings'
opinions, stances, and attitudes from textual data, has been widely applicable
in a diverse range of domains or languages. Recent advancements in deep
learning enable breakthroughs in this typical NLP area. However, the huge
success highly depends on massive labeled data. This has spurred on the novel
transfer learning researches in more challenging low-resource sentiment
analysis (LRSA), which aims to improve predictive models in a target domain (or
language) with minimal supervision by leveraging knowledge from a source domain
(or language) that are sufficiently labeled.

In this proposal, we in-depth investigate domain adaptation in LRSA, of which
the key is to minimize the feature distribution shift between domains.
Specifically, we propose neural knowledge transfer in different perspectives of
LRSA, i.e., document-based sentiment analysis (DBSA), aspect-based sentiment
analysis (ABSA) and end-to-end ABSA (E2E-ABSA). Among them, we will research
three major aspects: (1) In DBSA, how to learn a deep neural model that can
explicitly and automatically identify both domain-invariant and domain-specific
information as transferable knowledge without any target domain supervision;
(2) how to transfer aspect-specific knowledge between domains by considering
the effect of aspects in ABSA; (3) how to jointly extract aspects and
aspect-oriented sentiments across domains in E2E-ABSA. Due to the diversity of
human languages, language adaptation remains to be another critical problem for
LRSA, where both the feature space and feature distribution are different
across languages. This inspires us to further research in cross-lingual
sentiment analysis (CLSA). Motivated by human beings' ability to learn new
skills rapidly with a few examples by leveraging knowledge from previous tasks
(experiences), we propose the final research objective: (4) how to enhance the
transfer effectiveness by leveraging previous cross-lingual transfer
experiences instead of transferring from scratch. We will use several public
and real-world datasets, to validate this research. The proposal will also
discuss some difficulties which have been tackled by related works, and then
point out some future research issues for extensive investigation.


Date:                   Tuesday, 17 March 2020

Time:                   3:00pm - 5:00pm

Zoom Meeting:           https://hkust.zoom.us/j/8075887760

Committee Members:      Prof. Qiang Yang (Supervisor)
                        Prof. Huamin Qu (Chairperson)
                        Dr. Kai Chen
                        Dr. Yangqiu Song


**** ALL are Welcome ****