Fields of Interest

Natural Language Processing, Generative AI, Machine Learning

Research Content

Currently, I am working on detecting useless reviews using a multimodal approach. The framework consists of three main branches:

• Text Branch : A post-trained BERT model combined with LSTM to capture semantic and time burst features from review text.
• Image Branch : An improved architecture based on CLIP and Xception, optimized through Orthogonal Subspace Decomposition, to extract visual features from review-related images.
• Author Branch : BERT and the optimized image model are reused to model author behavior and writing style.

Features from all branches are then fused to predict the final uselessness score of a review.