Unifying Nearest Neighbors Collaborative Filtering

We study collaborative filtering for applications in which there exists for every user a set of items about which the user has given binary, positive-only feedback (one-class collaborative filtering). Take for example an on-line store that knows all past purchases of every customer. An important class of algorithms for one-class collaborative filtering are the nearest neighbors algorithms, typically divided into userbased and item-based algorithms. We introduce a reformulation that unifies user- and item-based nearest neighbors algorithms and use this reformulation to propose a novel algorithm that incorporates the best of both worlds and outperforms state-of-the-art algorithms. Additionally, we propose a method for naturally explaining the recommendations made by our algorithm and show that this method is also applicable to existing user-based nearest neighbors methods.

Introduction. Typically, the training data for collaborative filtering is represented by a matrix in which the rows represent the users and the columns represent the items. A value in this matrix can be unknown or can reflect the preference of the respective user for the respective item. Here, we consider the specific setting of binary, positive-only preference feedback. Hence, every value in the preference matrix is 1 or 0, with 1 representing a known preference and 0 representing the unknown. Pan et al. [10] call this setting one-class collaborative filtering (OCCF). Applications that correspond to this version of the collaborative filtering problem are likes on social networking sites, tags for photo’s, websites visited during a surfing session, articles bought by a customer etc. In this setting, identifying the best recommendations can be formalized as ranking all user-item-pairs (u, i) according to the probability that u prefers i. One class of algorithms for solving OCCF are the nearest neighbors algorithms. Sarwar et al.

Discussion / Conclusion. AND FUTURE WORK We proposed KUNN, a novel algorithm for one class collaborative filtering, a setting that covers many applications. KUNN originates from a reformulation that unifies userand item-based nearest neighbors algorithms. Thanks to this reformulation, it becomes clear that user- and itembased nearest neighbors algorithms discard important parts of the available information. KUNN improves upon these existing nearest neighbors algorithms by actually using more of the available information. Our experimental evaluation shows that KUNN not only outperforms existing nearest neighbors algorithms, but also state-of-the-art matrix factorization algorithms. Finally, we challenged the well accepted belief that itembased algorithms are superior for explaining the recommendations they produce. Thanks to our reformulation, we were able to show that also recommendations by KUNN and the traditional user-based algorithm come with a natural explanation. We see research on novel definitions of the functions L, N, G and S as the most important direction for future work.