Research On Review Summary Generation Based On Text Summarization

Many consumers resort to the reviews generated by previous consumers for decision making,while their time is limited to deal with too many reviews.Therefore,a review summary which contains all important features in user-generated reviews,is strongly expected.In this thesis,we study “how to generate a comprehensive review summary from a large number of user-generated reviews.” It can be implemented by text summarization,which mainly has two types of the extractive and abstractive approaches.The former is able to deal with both supervised and unsupervised scenarios,but may generate redundant and incoherent summaries,while the latter can avoid redundancy but it can only deal with supervised scenarios and short sequences.Moreover,both approaches usually neglect the sentiment information.To address the above issues,we propose a novel Review Summary Generation framework to integrate the supervised and unsupervised approaches,utilizing extractive and abstractive text summarization methods.The main research contents,contributions and innovations of this thesis are as follows:1)This thesis develops two comprehensive pre-process strategies to identify important sentences or reviews: Re-ranking model for sentences and Selecting model for review subsets.Re-ranking model is used to re-rank the sentences of the reviews by their semantic similarity and user's sentiment,which will be used as the input of the encoder-decoder summary generation model.Selecting model is used to select a subset of reviews covering as many aspects as possible,which can be used to generate both the summary and serve as the input for our encoder-decoder generation model.2)This thesis applies the encoder-decoder generation model to the review summary generation task for the supervised scenarios.This solution combines extractive and abstractive approaches.3)We propose an unsupervised method to deal with unsupervised scenarios that have no human written summaries as standard references.It is able to identify the aspects of each item(i.e.,a product)and generate the summaries that cover as many aspects as possible with as few sentences as possible.4)Experimental results in three real world data sets(Idebate,Rotten Tomatoes and JDPhoneReview)demonstrate that our work performs well in review summary generation.