Jesús Bobadilla

A collaborative filtering approach to mitigate the new user cold start problem

The new user cold start issue represents a serious problem in recommender systems as it can lead to the loss of new users who decide to stop using the system due to the lack of accuracy in the recommendations received in that first stage in which they have not yet cast a significant number of votes with which to feed the recommender systems collaborative filtering core. For this reason it is particularly important to design new similarity metrics which provide greater precision in the results offered to users who have cast few votes. This paper presents a new similarity measure perfected using optimization based on neural learning, which exceeds the best results obtained with current metrics. The metric has been tested on the Netflix and Movielens databases, obtaining important improvements in the measures of accuracy, precision and recall when applied to new user cold start situations. The paper includes the mathematical formalization describing how to obtain the main quality measures of a recommender system using leave-one-out cross validation.

Collaborative filtering adapted to recommender systems of e-learning

In the context of e-learning recommender systems, we propose that the users with greater knowledge (for example, those who have obtained better results in various tests) have greater weight in the calculation of the recommendations than the users with less knowledge. To achieve this objective, we have designed some new equations in the nucleus of the memory-based collaborative filtering, in such a way that the existent equations are extended to collect and process the information relative to the scores obtained by each user in a variable number of level tests.

Improving collaborative filtering recommender system results and performance using genetic algorithms

This paper presents a metric to measure similarity between users, which is applicable in collaborative filtering processes carried out in recommender systems. The proposed metric is formulated via a simple linear combination of values and weights. Values are calculated for each pair of users between which the similarity is obtained, whilst weights are only calculated once, making use of a prior stage in which a genetic algorithm extracts weightings from the recommender system which depend on the specific nature of the data from each recommender system. The results obtained present significant improvements in prediction quality, recommendation quality and performance.

A collaborative filtering similarity measure based on singularities

Recommender systems play an important role in reducing the negative impact of information overload on those websites where users have the possibility of voting for their preferences on items. The most normal technique for dealing with the recommendation mechanism is to use collaborative filtering, in which it is essential to discover the most similar users to whom you desire to make recommendations. The hypothesis of this paper is that the results obtained by applying traditional similarities measures can be improved by taking contextual information, drawn from the entire body of users, and using it to calculate the singularity which exists, for each item, in the votes cast by each pair of users that you wish to compare. As such, the greater the measure of singularity result between the votes cast by two given users, the greater the impact this will have on the similarity. The results, tested on the Movielens, Netflix and FilmAffinity databases, corroborate the excellent behaviour of the singularity measure proposed.

Collaborative filtering based on significances

It seems reasonable to think that there may be some items and some users in a recommender system that could be highly significant in making recommendations. For instance, the recent and much-advertised Apple product may be regarded as more significant compared with an outdated MP3 device (which is still on sale). In this paper, we introduce a new method to improve the information used in collaborative filtering processes by weighting the ratings of the items according to their importance. We provide here a formalisation of the collaborative filtering process based on the concept of significance. In this way, the k-neighbours are calculated taking into account the ratings of the items, the significance of the items and the significance of each user for making recommendations to other users. This formalisation includes extensions of the concepts related to similarity measures and prediction/recommendation quality measures. We will show also the results obtained from a set of experiments using Movielens and Netflix. The results confirm the advantage of introducing the concept of significance in general recommender systems and especially in recommender systems in which it is easy to determine the relative importance of the items: for example, most widely sold products in e-commerce, most widely commented news items in web-news, most widely watched programs on TV, and the latest sports champions.

A non negative matrix factorization for collaborative filtering recommender systems based on a Bayesian probabilistic model

In this paper we present a novel technique for predicting the tastes of users in recommender systems based on collaborative filtering. Our technique is based on factorizing the rating matrix into two non negative matrices whose components lie within the range 0, 1 with an understandable probabilistic meaning. Thanks to this decomposition we can accurately predict the ratings of users, find out some groups of users with the same tastes, as well as justify and understand the recommendations our technique provides.

Improving collaborative filtering-based recommender systems results using Pareto dominance

Recommender systems are a type of solution to the information overload problem suffered by users of websites that allow the rating of certain items. The collaborative filtering recommender system is considered to be the most successful approach, as it makes its recommendations based on ratings provided by users who are similar to the active user. Nevertheless, the traditional collaborative filtering method can select insufficiently representative users as neighbours of the active user. This means that recommendations made a posteriori are not sufficiently precise. The method proposed in this paper uses Pareto dominance to perform a pre-filtering process eliminating less representative users from the k-neighbour selection process while retaining the most promising ones. The results from experiments performed on the Movielens and Netflix websites show significant improvements in all tested quality measures when the proposed method is applied.

Incorporating reliability measurements into the predictions of a recommender system

In this paper we introduce the idea of using a reliability measure associated to the predictions made by recommender systems based on collaborative filtering. This reliability measure is based on the usual notion that the more reliable a prediction, the less liable to be wrong. Here we will define a general reliability measure suitable for any arbitrary recommender system. We will also show a method for obtaining specific reliability measures specially fitting the needs of different specific recommender systems.

Recommending items to group of users using Matrix Factorization based Collaborative Filtering

Group recommender systems are becoming very popular in the social web owing to their ability to provide a set of recommendations to a group of users. Several group recommender systems have been proposed by extending traditional KNN based Collaborative Filtering. In this paper we explain how to perform group recommendations using Matrix Factorization (MF) based Collaborative Filtering (CF). We propose three original approaches to map the group of users to the latent factor space and compare the proposed methods in three different scenarios: when the group size is small, medium and large. We also compare the precision of the proposed methods with state-of-the-art group recommendation systems using KNN based Collaborative Filtering. We analyze group movie ratings on MovieLens and Netflix datasets. Our study demonstrates that the performance of group recommender systems varies depending on the size of the group, and MF based CF is the best option for group recommender systems.

Choice of metrics used in collaborative filtering and their impact on recommender systems

The capacity of recommender systems to make correct predictions is essentially determined by the quality and suitability of the collaborative filtering that implements them. The common memory-based metrics are Pearson correlation and cosine, however, their use is not always the most appropriate or sufficiently justified. In this paper, we analyze these two metrics together with the less common mean squared difference (MSD) to discover their advantages and drawbacks in very important aspects such as the impact when introducing different values of k-neighborhoods, minimization of the MAE error, capacity to carry out a sufficient number of predictions, percentage of correct and incorrect predictions and behavior when attempting to recommend the n-best items. The paper lists the results and practical conclusions that have been obtained after carrying out a comparative study of the metrics based on 135 experiments on the MovieLens database of 100,000 ratios.