Recommender systems aim to forecast users’ rank, interests, and preferences in specific products and recommend them to a user for purchase. Collaborative filtering is the most popular approach, where the user’s past purchase behavior consists of the user’s feedback. One of the most challenging problems in collaborative filtering is handling users whose previous item purchase behavior is unknown, (e.g., new users) or products for which user interactions are not available, (e.g., new products). In this work, we address the cold-start problem in recommender systems based on frequent patterns which are highly frequent in one set of users, but less frequent or infrequent in other sets of users. Such discriminant frequent patterns can distinguish one target set of users from all other sets. The proposed methodology, first forms different clusters of old users and then discovers discriminant frequent patterns for each different such cluster of users and finally exploits the latter to hallucinate the purchase behavior of new users. We also present empirical results to demonstrate the efficiency and accuracy of the proposed methodology.
There is a growing interest in the offering of novel alternative choices to users of recommender systems. These recommendations should match the target query while at the same time they should be diverse with each other in order to provide useful alternatives to the user, i.e., novel recommendations. In this paper, the problem of extracting novel recommendations, under the similarity–diversity trade-off, is modeled as a facility location problem. The results from tests in the benchmark Travel Case Base were satisfactory when compared to well-known recommender techniques, in terms of both similarity and diversity. It is shown that the proposed method is flexible enough, since a parameter of the adopted facility location model constitutes a regulator for the trade-off between similarity and diversity. Also, our work can broaden the perspectives of the interaction and combination of different scientific fields in order to achieve the best possible results.
In this paper, we present a clustering heuristic for solving demand covering models where the objective is to determine locations for servers that optimally cover a given set of demand points. This heuristic is based on the concept of biclusters and processes the set of demand points as well as the set of potential servers and determines biclusters that result in smaller problems. Given a coverage matrix, a bicluster is defined as a sub-matrix spanned by both a subset of rows and a subset of columns, such that rows are the most similar to each other when compared over columns. The algorithm starts by using any biclustering algorithm in order to identify appropriate biclusters of the coverage matrix and then combines selected biclusters to define an aggregate solution to the original problem. The algorithm can be easily adapted to address a whole family of covering problems including set covering, maximal covering and backup covering problems. The proposed algorithm is tested in a series of widely known test datasets for various such problems. The main objective of this paper is to introduce the concept of biclustering as an efficient and effective approach to tackle covering problems and to stimulate further research in this area.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.