Embedding Refinement Framework for Targeted Aspect-Based Sentiment Analysis

The state-of-the-art approaches to targeted aspect-based sentiment analysis (TABSA) are mostly built on deep neural networks with attention mechanisms. One problem is that embeddings of targets and aspects either pre-trained from large external corpora or randomly initialized. We argue affective commonsense knowledge words indicative could be used learn better target aspect embeddings. therefore propose an embedding refinement framework called RAEC ( xmlns:xlink="http://www.w3.org/1999/xlink">R efining xmlns:xlink="http://www.w3.org/1999/xlink">A ffective xmlns:xlink="http://www.w3.org/1999/xlink">E mbedding xmlns:xlink="http://www.w3.org/1999/xlink">C ontext), in which concepts extracted word relative location information incorporated derive context-affective Furthermore, a sparse coefficient vector exploited refining the separately. In this way, can capture highly relevant words. Experimental results two benchmark datasets show our easily integrated existing embedding-based TABSA models achieves compared relying other methods.