Search

This paper addresses the problem of recommendation with the context of one-class classification. Specifically, we employ the alternative machine learning framework of Artificial Immune Systems (AIS) in order to develop user-specific preference models. Our approach is based on the fact that users experience a major difficulty in articulating their music preferences while at the same time they are extremely reluctant in providing counter-examples of their music habits. Therefore, developing coherent user models on the grounds of both positive (desirable) and negative (non-desirable) training samples is not a feasible task since the class of non-favorable data patterns is severely under-represented. Our recommendation approach alleviates the need to collect negative feedback form the user by building recommendation models that exclusively rely on the presence of a limited number positive data items. Such models are built, however, by trying to efficiently cover the complementary space of non-desirable patterns. This is achieved through the utilization of V-Detector, an AIS-based one-class classification algorithm, which operates by developing a set of variable-sized detectors for the subspace of non-preferable music items. V-Detector, despite being exclusively fed with instances from the positive class, focuses on delivering an accurate model of the negative space. Based on this complementarity, our recommendation algorithm is able to implicitly model individual user preferences that span arbitrary-shaped and fragmented regions of the complete space of patterns. The proposed recommendation approach was experimentally evaluated in terms of its efficiency to correctly identify the user-defined classes of positive and negative preference. The obtained results justify its superiority against traditional one-class classification approaches. [ABSTRACT FROM AUTHOR]

In this paper, we formulate the recommendation problem as a hybrid combination of one-class classification with collaborative filtering. Specifically, we decompose the recommendation problem into a two-level cascade scheme. In the first level, only desirable items are selected for each user from the large amount of all possible items, taking into account only a small portion of his/her available preferences. This is achieved via a one-class classification scheme trained only with positives examples, i.e. only with desirable items for which users have provided a rating value. In the second level, a collaborative filtering approach is applied to assign a rating degree to the items identified at the first level. The efficiency of our approach is analyzed theoretically in terms of best/worst case scenarios and respective lower/upper mean absolute error (MAE) bounds are computed. Moreover, our approach is experimentally tested against pure collaborative and cascade content-based approaches. The results show that our approach outperforms them in terms of MAE and, moreover, the experimental MAE is close to the theoretical lower bound corresponding to the best case scenario. The superiority of our approach is due to the existence of the one class classifier in the first level of the cascade. [ABSTRACT FROM AUTHOR]