Article Preview
TopIntroduction
With the development of a “Knowledge Society,” we are obtaining increasingly more information through various flexible, fast and intelligent means. Users, however, are often overloaded with a large amount of continuous data in the decision-making process, especially on the Internet. In such a situation, users are confused about how to choose between alternatives that might best fit their preferences. For instance, micro-blog users who follow thousands of users will receive numerous interesting messages every moment (by refreshing the interface). These messages are from different fields of a social network. Users need to discriminate what they are most interested in and what should be neglected. Other examples that we face every day include choosing a movie to watch, searching for online news that is continuously generated, and selecting vacation destinations. Therefore, it is necessary to have a good model for storing and managing user preferences. Moreover, the model should enable all the incoming data to be rated (Montaner, López & De La Rosa, 2003), ranked, and filtered according to user interests.
A Recommender System (RS) (Porcel & Herrera-Viedma, 2010; Zhang & Lin, 2013; Basile, 2007; Marin & Valls, 2013) assists users in the decision-making process. A RS usually stores personal profiles associated to each user, in which their numerical preferences for an item are recorded. This information is important for recommending suitable alternatives to the target user since it is necessary to rank the available alternatives according to the user’s preference profile. Evaluating which alternative better fits the user preference profile is of equal importance to evaluating which alternative fits the real user.
It is usually assumed that the initial user preference profile (Konstan & Riedl, 2012; Marin & Isern, 2013; Marin & Isern, 2013) is incomplete and imprecise, and that is why it is necessary to update the latest user preference profile. In order to update the profile dynamically, the system must receive relevant feedback from interactions between the user and RS. There are two types of available feedback (Roy, 2005; Jawaheer & Szomszor, 2010): explicit feedback and implicit feedback.
Explicit feedback (Noppens & Luther, 2006; Zigoris & Zhang, 2006) is obtained when users are asked to evaluate items and indicate how relevant or interested they are regarding that item based on a numeric scale. Explicit feedback has limitations, however, and the most serious being the time necessary for users to provide feedback. In comparison, implicit feedback (Cheng & Jing, 2006; Marin & Isern, 2013) can be obtained from monitoring interactions between users and the RS and automatically handling the user preference profile. This paper analyzes preference information that is implicit in user actions to update a user preference profile.
Many alternatives in RS can be represented by numerical attributes (Marin & Isern, 2013; Alexis & Nikolaos, 2011). This paper focuses on updating the user preference profile with numerical attributes. The process of dynamic update of the profile involves two basic steps: the process of recommendation and the process of dynamic adaptation update. A group of possible alternatives are sorted and recommended to the target user by RS, and then valuable preference information can be collected from user selections. In (Marin & Isern, 2013), a ranking list of alternatives can be recommended by comparing the similarity score between a user preference profile and that of alternatives. However, an important factor has been neglected, as in a real situation, users usually pay more attention to one or two significant attributes when selecting the alternatives recommended by a RS. Therefore, different weights must be considered for each attribute in this process. In the process of dynamic adaptation update, the valuable feedback collected in the process of recommendation can be used to update the values of different weights. The values of weights can be used to update the user preference profile by using an efficient and dynamic update algorithm. The final goal of this paper is to update weights of different attributes, and to update the values stored in a user preference profile according to different weights, and then to make the user preference profile more similar to the ideal preference of the user after several interactions.