Recommender Systems have become a crucial tool to serve personalized content and to promote online products and media, but also to recommend restaurants, events, news and dating profiles. The underlying algorithms have a significant impact on the quality of recommendations and have been the subject of many studies in the last two decades. In this thesis we focus on factorization models, a class of recommender system algorithms that learn user preferences based on a method called factorization. This method is a common approach in Collaborative Filtering (CF), the most successful and widely-used technique in recommender systems, where user preferences are learnt based on the preferences of similar users. We study factorization models from an algorithmic perspective to be able to extend their applications to a wider range of problems and to improve their effectiveness. The majority of the techniques that are proposed in this thesis are based on state-of-the-art factorization models known as Factorization Machines (FMs). ... Read more

In the CLEF NewsREEL 2017 challenge, we build a delegation model based on the contextual bandit algorithm. Our goal is to investigate whether a bandit approach combined with context extracted from the user side, from the item side and from user-item interaction can help choose the appropriate recommender from a recommender algorithm pool for the incoming recommendation requests. We took part in both tasks: NewsREEL Live and NewsREEL Replay. In the experiment, we test several bandit approaches with two types of context features. The result from NewsREEL Replay suggests that delegation model based on the contextual bandit algorithm can improve the click through rate (CTR). In NewsREEL Live, a similar delegation model is implemented. However, the delegation model from NewsREEL Live is trained by the data stream from NewsREEL Replay. This is due to the fact that the low volume of data received from the online scenario is not enough to support the training of the delegation model. For our future work, we will add more recommender algorithms to the recommender algorithm pool and explores other context features. ... Read more

This paper states the case for the principle of minimal necessary data: If two recommender algorithms achieve the same effectiveness, the better algorithm is the one that requires less user data. Applying this principle involves carrying out training data requirements analysis, which we argue should be adopted as best practice for the development and evaluation of recommender algorithms. We take the position that responsible recommendation is recommendation that serves the people whose data it uses. To minimize the imposition on users’ privacy, it is important that a recommender system does not collect or store more user information than it absolutely needs. Further, algorithms using minimal necessary data reduce training time and address the cold start problem. To illustrate the trade-off between training data volume and accuracy, we carry out a set of classic recommender system experiments. We conclude that consistently applying training data requirements analysis would represent a relatively small change in researchers’ current practices, but a large step towards more responsible recommender systems. ... Read more

Pairwise learning-to-rank algorithms have been shown to allow recommendersystems to leverage unary user feedback. We proposeMulti-feedback Bayesian Personalized Ranking (MF-BPR), a pairwisemethod that exploits different types of feedback with an extendedsampling method. The feedback types are drawn from different“channels”, in which users interact with items (e.g., clicks,likes, listens, follows, and purchases). We build on the insight thatdifferent kinds of feedback, e.g., a click versus a like, reflect differentlevels of commitment or preference. Our approach differs fromprevious work in that it exploits multiple sources of feedback simultaneouslyduring the training process. The novelty of MF-BPRis an extended sampling method that equates feedback sources with“levels” that reflect the expected contribution of the signal. Wedemonstrate the effectiveness of our approach with a series of experimentscarried out on three datasets containing multiple typesof feedback. Our experimental results demonstrate that with a rightsampling method, MF-BPR outperforms BPR in terms of accuracy.We find that the advantage of MF-BPR lies in its ability to leveragelevel information when sampling negative items. ... Read more

Crowdsourcing and Human computation have enabled industry and scientists to create innovative solutions by harnessing organised collective human effort. In human computation platforms, it is observed that workers spend large amount of time searching for appropriate tasks due to lack of effective task discovery mechanism. This loss of time translates to loss of incentives for worker and affects motivation to solve more tasks. Task recommendation in human computation can not only help mitigating this problem, but it can also result in high quality answers from motivated workers. While few works empirically proved the benefits of task recommendation in human computation platforms, we advocate for a better scientific understanding of how worker- and task-modelling can help to systematically achieve faster and higher-quality task executions. To this end, it is fundamental the availability of tools able to “open the box” of human computation, and offer direct control on worker and task properties, to be later used for recommendation. This paper presents BruteForce, a framework that simplifies experiments with commercial human computation platforms, while offering task recommendation features based on rich (and extensible) set of worker and task properties. We describe the characteristics of BruteForce, and we report on a set of preliminary experiments with three user profiling techniques namely featureindependent, feature-based and Composite. ... Read more