Collaborative Filtering With User Preference And User Opinion

With the emergence and development of Web 2.0, more and more people begin to express their opinions on products and services online. Users’ opinions usually consist of a free-text review and an overall rating, and can be very useful for merchants and other users. For merchants, they can find out the characteristics that users are very satisfied and the characteristics that users are not very satisfied from the reviews, and then make product improvement or sales plans accordingly. Merchants do this can help them obtain the maximum profits. For consumers, seeing others’ reviews on a product can help them make informed decisions. Such reviews and ratings also form the basis of recommender systems, which aim to predict, from the possibly millions of products available, the ones that users will like.Collaborative filtering (CF), a popular technique used in recommender systems, makes predictions about a user interests by collecting preferences information from many users, usually in the form of ratings of items. At present, there have been a lot of collaborative filtering algorithms. However, most existing CF methods rely only on the overall ratings the items have received. However, the overall ratings cannot provide us more detailed information. For example, a user giving a high rating to an item may indicate that the user loves the item as a whole. However, it is still possible that he dislikes some aspects. Also, users tend to place different emphases on different aspects when reaching the overall rating. The emphasis on aspects varies for different users and different items.In order to solve this new problem, in this paper, we propose a framework that incorporates user opinions and preferences on different aspects in order to predict the unknown ratings on items. This framework has three components, an aspect-based opinion mining component, an aspect weighting computing component, and a rating inference component. In opinion mining component, it exploits the opinion mining techniques to extract and summarize the opinions on multiple aspects from the reviews and generates ratings on the various aspects. In aspect weighting computing component, we use a tensor factorization approach to automatically infer the weights of different aspects in forming the overall rating. The rating inference component infers the overall ratings of items based on the aspect ratings and aspect weights.We evaluate our proposal on two datasets. We compare our model with some baseline methods, and the experiment results show that our model outperforms others. The major contributions of this paper can be summarized as follows:(?) We propose a new model that integrates user opinions and preferences on multiple aspects into collaborative filtering for the overall ratings prediction.(?) We employ a tensor factorization approach to capture the aspect weights, which alleviates the data sparsity problem and reduces the number of model parameters.(?) We propose a method to predict the unknown overall ratings through (again) tensor factorization, where the tensor is constituted by weighted aspect ratings and overall ratings. A major advantage of this method lies in its ability to capture the intrinsic interactions among the three dimensions:user, item, and aspect.(?) We conduct extensive experiments on some datasets to verify the effectiveness of our proposed approach.