A context-aware recommendation approach based on feature selection
- PDF / 699,073 Bytes
- 11 Pages / 595.224 x 790.955 pts Page_size
- 91 Downloads / 221 Views
A context-aware recommendation approach based on feature selection Lei Chen1
· Meimei Xia1
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Contextual information can be used in recommender systems to make recommendation more efficient. Recent research has made progress in combining contextual information into representation models for recommendations. However, the existing approaches do not well address the problem of data sparsity, and they suffer from context redundancy. To deal with these problems, this paper proposes a context-aware recommendation approach based on embedded feature selection. It gets rid of context redundancy by generating a minimum subset of all contextual information and allocates the weight to each context appropriately. Experiments on the restaurant recommendation shows that the proposed approach has better performance. Keywords Recommender system · Context-aware · Feature selection · Restaurant recommendation
1 Introduction Much effort has been put in the area of recommender systems (RS) since it is introduced by Resnick et al [1]. Traditional RS can be classified into collaborative filtering, content-based systems, and hybrid methods, mostly using rating values on items to evaluate users’ opinions. For more accurate results, context-aware recommender systems (CARS) [2] extend traditional RS to better-understood users’ preferences. Context is defined as “any information that characterizes the circumstances of an entity” [3], such as time, location, and user activities. One may choose different restaurants when going on a quick business lunch instead of dining with friends. Therefore, incorporating contextual information into the recommendation progress will improve the performance of the prediction. Shin et al [6] proposed CAR-AUC that incorporates contexts by computing correlation, in which similarities between the current context and the context history of users are firstly computed and then are multiplied. Although the algorithm Meimei Xia
[email protected] Lei Chen [email protected] 1
Beijing Jiaotong Univeristy, Beijing, China
considers the context, which increases the accuracy of recommendation, the contexts it aggregated are multidimensional. That means as the contexts of users and items grows, this algorithm will become highly computational in calculating similarities. To avoid high dimensionality, recent research [33] presents a multidimensional approach named as DaVI, which can insert contextual information new useritem pairs. Thus, it can be applied to two-dimensional algorithms in order to improve the prediction precision. This is based on the assumption that different contextual information is independent. However, there exists contextual information with a high-level relationship and considering all of them will increase the redundancy. Furthermore, the weights of relevant contextual information should be investigated in order to reflect the contribution of each context. That is, certain context should be emphasized when they are p
Data Loading...
