Popularity bias and unfairness are problems caused by the lack of calibration in recommender systems. Works that intend to reduce effect popularity do not consider distribution item genres/categories users' profiles. Other studies aim calibrate system generate fair recommendations according profiles, but usually still biased towards popularity. We propose a approach based on preferences for different levels items their genres. The proposed works post-processing stage can be combined with...

Current approaches on collaborative filtering factorize user-item matrices in order to infer latent factors from ratings previously assigned by users. However, they all have deal with sparseness, whose workarounds are prone bias and/or overfitting. This paper proposes a recommender algorithm that is based factorized matrix composed of user preferences associated the movies' genres/categories. The advantage using such user-genre factorization model it requires less computational resources, as...

This paper proposes a recommender algorithm denominated "gSVD++" which exploits implicit feedback from users by considering not only the latent space of factors describing user and item, but also available metadata associated to content. Such descriptions are an important source construct profile containing relevant meaningful information about his/her preferences. The method is evaluated on MovieLens dataset, being compared against other approaches reported in literature. results show...

In this paper, we propose an approach based on sentiment analysis to describe items in a neighborhood-based collaborative filtering model. We use unstructured users' reviews produce vector-based representation that considers the overall of those towards specific features. and compare two different techniques obtain score such features from textual content, namely term-based aspect-based feature extraction. Finally, our proposal is compared against structured metadata under same...

In this paper, we propose a technique that uses multimodal interactions of users to generate more accurate list recommendations optimized for the user . Our approach is response actual scenario on Web which allows interact with content in different ways, and thus, information about his preferences can be obtained improve recommendation. The proposal consists an ensemble learning combines rankings generated by unimodal recommenders based particular interaction types. By using combination...

Recommender systems help users to deal with the information overload problem by producing personalized content according their interests. Beyond traditional recommender strategies, there is a growing effort incorporate users' reviews into recommendation process, since they provide rich set of regarding both items' features and preferences. This article proposes system that uses produce representations are based on overall sentiment toward features. We focus exploiting impact different...

This paper presents a polished open-source Python-based recommender framework named Case Recommender, which provides rich set of components from developers can construct and evaluate customized systems. It implements well-known state-of-the-art algorithms in rating prediction item recommendation scenarios. The main advantage the Recommender is possibility to integrate clustering ensemble with engines, easing development more accurate efficient approaches.

The research community has become increasingly aware of possible undesired effects algorithmic biases in recommender systems. One common bias such systems is to over-proportionally expose certain items users, which may ultimately result a system that considered unfair individual stakeholders. From technical perspective, calibration approaches are commonly adopted situations ensure the user's preferences better taken into account, thereby also leading more balanced exposure overall. Given...

Unlike traditional recommender systems, which make recommendations only by using the relation between users and items, a context-aware system makes incorporating available contextual information into recommendation process. One problem of approaches is that it required techniques to extract such additional in an automatic manner. In this paper, we propose use two text mining are applied textual data infer automatically: named entities recognition topic hierarchies. We evaluate proposed...

In this paper, we propose a technique to automatically describe items based on users' reviews in order be used by recommender systems. For that, extract items' features using robust term extraction method that applies transductive semi-supervised learning identify aspects represent the different subjects of reviews. Then, apply sentiment analysis sentence level indicate polarities, yielding consensus users regarding items. Our approach is evaluated collaborative filtering method, and...

This paper proposes a collaborative filtering approach that uses users' reviews to produce item descriptions represent consensus of users regarding items' features. While earlier works focused on using structured metadata items, recent approaches study how use user-provided text, such as reviews, better insights about the semantics in content. Some involved problems, noise, personal opinions and false information are reduced by an algorithm based sentiment analysis natural language...

Unlike the traditional recommender systems, that make recommendations only by using relation between user and item, a context-aware system makes incorporating available contextual information into recommendation process as explicit additional categories of data to improve process. In this paper, we propose use from topic hierarchies accuracy systems. Additionally, also two algorithms for item recommendation. These are extensions proposed in literature rating prediction. The empirical results...

In this paper, we propose an architecture which supports metadata extraction by exploring interaction mechanisms among users and content. The activity addressed in work is related to peer-level annotation, where any user acts as author, being able enrich the content making annotations, using, for instance, pen-based devices. Peer-level annotation makes comfortable when taking digital notes, they do every day life. This advantage over hierarchical authoring, a time-consuming task usually...

Nowadays, networks can be accessed by multiple devices with different characteristics. Some of these characteristics such as low processing power and memory capacity restrict the access to multimedia content. Researchers have then focused on making automatic system adaptations in order present content according devices' capabilities. Most works this area include description hardware software features using CC/PP specification. However, indiscriminate direct use makes it difficult generate...

In Recommender Systems, a large amount of labeled data must be available beforehand to obtain good predictions. However, are often limited and expensive obtain, since labeling typically requires human expertise, time, labor. This paper proposes framework, named CoRec, which is based on co-training approach that drives two recommenders agree with each other's predictions generate their own. We used three publicly datasets from movies, jokes books domains, as well well-known recommender...