An Approach for Automatic Aspect Extraction by Latent Dirichlet Allocation

Abstract

Now a days internet has taken over all form of communication and activities. One of the most affected area is e-commerce which survives through internet. One of the vital process of business survival is customer feedback. There are several form of platforms where the feedbacks can be posted. The main task is to accumulate all the reviews and summarize them in a conceivable manner. The approach here is to summarize the reviews in aspect based manner. This representation will help future consumers to make well informed decision. LDA is one of the popular methods to extract latent topics comprising a document. The present approach intends to use that characteristic to extract aspects from review corpora. The current scope of work is to extract and improve the quality of topics. After several corpora of product reviews were processed through this method, the results were examined through graphically plotting the topics and also examining the dominant keywords of the topics. Finally the accumulated results of the present method are compared with the results of previously implemented Word2Vec based model and human extracted aspects.