Opinion Mining with Manifold Forests

Online reviews are becoming increasingly popular every day. They represent opinions and a wealth of information that can benefit organizations and individual consumers. However, studies on opinion mining have not focused much on classifying views according to the manifold to solve the problem of affinity between clusters of opinion, improving the accuracy, effectiveness, and generalizability of the modeling. In this paper, we have built an opinion mining framework with manifold forests to solve the influence of clusters of opinions based on the affinity between pairwise opinion points in each cluster and the relationship between different opinion clusters in large-scale data. In particular, we have focused on building a clustering trees ensemble and determining the affinity and distance of point pairs in feature space. Finally, the random forests are aggregated by ensemble methods such as stacking with a random forests classifier to identify opinion classification in reviews as either negative or positive. We used two datasets in the experiment to evaluate restaurants and hotels in two different scenarios, proving the effectiveness of the proposed model.