A multi-aspect user-interest model based on sentiment analysis and uncertainty theory for recommender systems
- PDF / 1,561,371 Bytes
- 26 Pages / 439.37 x 666.142 pts Page_size
- 74 Downloads / 209 Views
A multi‑aspect user‑interest model based on sentiment analysis and uncertainty theory for recommender systems Lihua Sun1 · Junpeng Guo1 · Yanlin Zhu1
© Springer Science+Business Media, LLC, part of Springer Nature 2018
Abstract This work presents a new multi-aspect user-interest model for recommender systems to improve recommendation and prediction accuracy. We introduce the overall user satisfaction for a product to build a user-interest profile by computing the user-interest levels from multi-aspect reviews. A domain emotional dictionary is built to overcome the gap in quantity between negative and positive words for sentiment polarity analysis. A sentiment analysis model is designed to characterize the users’ sentiment polarity and strength based on uncertainty theory and the domain emotional dictionary. Accordingly, a new multi-aspect user-interest model is proposed by considering the sentiment analysis model with the user-interest profile. Then, our proposed model is applied to recommender systems and experimentally tested on five products of different categories from three e-commerce websites. Our model not only outperforms the traditional and state-of-the-art models on rating prediction tasks but also improves the recommendation accuracy in multiple domains. Keywords Recommender system · Sentiment analysis · Uncertainty theory · Product reviews · User interest
1 Introduction As the easiest and most convenient form of shopping, e-commerce has greatly developed in recent years. Surfing e-commerce websites for purchases has become one of the fastest growing activities, leading to a large amount of available information and ultimately to information overload [1]. The main solutions to the information overload problem are recommender systems. The main solutions to the information overload problem are recommender systems, which provide automated and personalized product suggestions to consumers [2–4]. Most current recommender techniques infer users’ opinions on the basis of user-provided * Junpeng Guo [email protected] 1
College of Management of Economics, Tianjin University, Tianjin 300072, China
13
Vol.:(0123456789)
L. Sun et al.
ratings and thus attempt to satisfy user preferences and demands [5–7]. However, these methods are limited by the well-known data sparsity problem, especially when a user has input limited historical data to the recommender system. The growing popularity of e-commerce websites has encouraged users to write reviews describing their experience with products. Typical reviews are composed of textual comments that explain why the user liked or disliked a product, thereby revealing their opinions about the product [8]. Reviews from marketing sales have proven to positively influence the decisions of new users [9]. Therefore, a large database of review information can not only improve the recommendation accuracy by alleviating the data sparsity problem but also improve the prediction accuracy [8, 10, 11]. Most reviews on e-commerce websites are written with a strong purpose to ex
Data Loading...