Prediction of New User Preferences with Filtering Techniques
Collaborative filtering techniques are often used to predict the unknown preferences of a new user by applying rules derived from the known preferences of a group of users. In the literature users having high correlation with a large number of other users
- PDF / 251,227 Bytes
- 5 Pages / 439.363 x 666.131 pts Page_size
- 95 Downloads / 234 Views
Abstract. Collaborative filtering techniques are often used to predict the unknown preferences of a new user by applying rules derived from the known preferences of a group of users. In the literature users having high correlation with a large number of other users are referred to as ’white sheeps’, while those that express preferences which do not fall into any known to the system group are called ’grey sheeps’. Thus predictions for the latter type users are often inaccurate. To overcome this problem we propose application of residuated lattices. Keywords: Residuated lattices. Concept lattices. Evaluation.
1
Introduction
One of the fundamental ideas in recommender system [13] is to improve a system’s performance by incorporating users’ ratings. The main assumption is that if users A and B rate k items similarly, they share similar tastes, and hence will rate other items similarly. Approaches differ in how they define a ’rating,’ how they define k, and how they define ’similarly’, [7]. Collaborative filtering (CF) describes techniques that use the known preferences of a group of users to predict the unknown preferences of a new user; recommendations for the new user are based on these predictions, [11]. Collaborative filtering systems involve either implicit or explicit votes. Votings where users indicate their preferences based on a discrete numerical scale are referred to as explicit votings. The implicit ones are concerned with data collected in a way that counts users actions rather than users opinions. Similar to market basket analysis they focus on pattern recognition. Most users belong to the set of white sheep users. They are the ones who benefit most from recommender systems since their preferences are relatively easy to predict. Black sheep users are the ones with very peculiar tastes and as such working out recommendations for them is practically impossible. Since black sheep users create significant problems for various recommender types they are considered to be an acceptable failure of recommender systems. The term ’grey sheep users’ introduced in [2] is meant to denote users who cannot be placed in any available set since their items’ evaluations diverge significantly from the known ones, [5]. Such users can hardly obtain reliable predictions regardless how long the system has been running. James J. (Jong Hyuk) Park et al. (eds.), Future Information Technology, Lecture Notes in Electrical Engineering 309, DOI: 10.1007/978-3-642-55038-6_27, © Springer-Verlag Berlin Heidelberg 2014
175
176
S. Encheva
Grey sheep users are known to create problems for working out adequate predictions by interfering with the levels of similarities in a system. To overcome this problem we propose application of residuated lattices.
2
Preliminaries
A large number of recommender systems are using collaborative filtering, [1]. The main idea is to structure the process of information filtering. Thus new users are assigned to groups of existing users who share similar ratings of some item sets. One of the problems rel
Data Loading...
