Understanding the Massive Online Reviews: a Novel Representative Subset Extraction method

With the widespread use of e-commerce, explosive online reviews often get overwhelming to online consumers to read, leading to the depression of decision making in purchasing. To deal with the information overload problem caused by large-scale online reviews, this study focuses on the representative review extraction problem and formulates this problem as an optimization model with a submodularity property. Then, to analyze and draw topic information from the original review collection, the topic model of LDA is adopted in this study to semantically measure the similarity between each pair of reviews for the optimization model. Furthermore, a greedy extraction method named RR with a satisfactory error bound is proposed to extract a representative subset of reviews from the original review collection based on the optimization model. Experiments on real data and a user study are conducted in this study. The experimental results demonstrate that the proposed method is of high efficiency and scalability, and performs better than benchmark methods in terms of coverage, which proves that it can help online consumers better capture the main ideas of the whole set of original reviews in a limited time.