Past work that improves document-level sentiment analysis has been limited to considering only the text of the current review . We investigate incorporating additional review text available at the time of sentiment prediction that may prove meaningful for guidingprediction . Experiment results on IMDB, Yelp 2013 and Yelp 2013 datasets show improvement to state-of-the-art of more than 2percentage points in the best case . We achieve this by explicitly storing representations of reviewswritten by the same user and about the same product and force the model tomemorize all reviews for one particular user and product . Additionally, we dropthe hierarchical architecture used in previous work to enable words in the text to directly attend to each other. We dropt

Author(s) : Chenyang Lyu, Jennifer Foster, Yvette Graham

Links : PDF - Abstract

Code :

Keywords : product - sentiment - text - user - document -

Leave a Reply

Your email address will not be published. Required fields are marked *