Day 123 of #NLP365: NLP Papers Summary -- Context-aware Embedding for Targeted Aspect-based…

Proposed context-aware embeddings to refine the embeddings of targets and aspects using highly correlative words. A lot of previous work uses context-independent vectors to construct targets and aspects embeddings, which lead to loss of semantic information and failed to capture the interconnection between the specific target, its aspect, and its context. This approach has led to SOTA results in targeted aspect-based sentiment analysis (TABSA). The goal of TABSA is that given an input sentence, we want to extract the sentiment of the aspect that belongs to a target. The refined target embeddings can be computed by multiplying the sentence word embeddings X with the sparse coefficient vector u'.