Hidden topics modeling approach for review quality prediction and classification
Hoan Tran Quoc, Hideya Ochiai, Hiroshi Esaki · 2015
The automatic assessment of online review's quality is becoming important with the number of reviews increasing rapidly. In order to help determining review's quality, some online services provide a system where users can evaluate or feedback the helpfulness of review as crowdsourcing knowledge. This approach has shortcomings of sparse voted data and richer-get-richer problem in which favor reviews are voted frequently more than others. In this work, we use Latent Dirichlet Allocation (LDA) method to exploit hidden topics distribution information of all reviews and propose supervisor prediction model based on probabilistic meaning of the review's quality. We also propose a deep neural network to classify the review in quality and validate our proposals within some real reviews datasets. We demonstrate that using hidden topics distribution information could be helpful to improve the accuracy of review quality prediction and classification.