Cataldo Musto

Semantics-Aware Content-Based Recommender Systems

Content-based recommender systems (CBRSs) rely on item and user descriptions (content) to build item representations and user profiles that can be effectively exploited to suggest items similar to those a target user already liked in the past. Most content-based recommender systems use textual features to represent items and user profiles, hence they suffer from the classical problems of natural language ambiguity. This chapter presents a comprehensive survey of semantic representations of items and user profiles that attempt to overcome the main problems of the simpler approaches based on keywords. We propose a classification of semantic approaches into top-down and bottom-up. The former rely on the integration of external knowledge sources, such as ontologies, encyclopedic knowledge and data from the Linked Data cloud, while the latter rely on a lightweight semantic representation based on the hypothesis that the meaning of words depends on their use in large corpora of textual documents. The chapter shows how to make recommender systems aware of semantics to realize a new generation of content-based recommenders.

Enhanced vector space models for content-based recommender systems

The use of Vector Space Models (VSM) in the area of Information Retrieval is an established practice within the scientific community. The reason is twofold: first, its very clean and solid formalism allows us to represent objects in a vector space and to perform calculations on them. On the other hand, as proved by many contributions, its simplicity does not hurt the effectiveness of the model. Although Information Retrieval and Information Filtering undoubtedly represent two related research areas, the use of VSM in Information Filtering is much less analzyed. The goal of this work is to investigate the impact of vector space models in the Information Filtering area. Specifically, I will introduce two approaches: the first one, based on a technique called Random Indexing, reduces the impact of two classical VSM problems, this is to say its high dimensionality and the inability to manage the semantics of documents. The second extends the previous one by integrating a negation operator implemented in the Semantic Vectors1 open-source package. The results emerged from an experimental evaluation performed on a large dataset and the applicative scenarios opened by these approaches confirmed the effectiveness of the model and induced to investigate more these techniques.

Content-based and collaborative techniques for tag recommendation: an empirical evaluation

The rapid growth of the so-called Web 2.0 has changed the surfers’ behavior. A new democratic vision emerged, in which users can actively contribute to the evolution of the Web by producing new content or enriching the existing one with user generated metadata. In this context the use of tags, keywords freely chosen by users for describing and organizing resources, spread as a model for browsing and retrieving web contents. The success of that collaborative model is justified by two factors: firstly, information is organized in a way that closely reflects the users’ mental model; secondly, the absence of a controlled vocabulary reduces the users’ learning curve and allows the use of evolving vocabularies. Since tags are handled in a purely syntactical way, annotations provided by users generate a very sparse and noisy tag space that limits the effectiveness for complex tasks. Consequently, tag recommenders, with their ability of providing users with the most suitable tags for the resources to be annotated, recently emerged as a way of speeding up the process of tag convergence. The contribution of this work is a tag recommender system implementing both a collaborative and a content-based recommendation technique. The former exploits the user and community tagging behavior for producing recommendations, while the latter exploits some heuristics to extract tags directly from the textual content of resources. Results of experiments carried out on a dataset gathered from Bibsonomy show that hybrid recommendation strategies can outperform single ones and the way of combining them matters for obtaining more accurate results.

An investigation on the serendipity problem in recommender systems

We design a Knowledge Infusion (KI) process for providing systems with background knowledge.We design a KI-based recommendation algorithm for providing serendipitous recommendations.An in vitro evaluation shows the effectiveness of the proposed approach.We collected implicit emotional feedback on serendipitous recommendations.Results show that serendipity is moderately correlated with surprise and happiness. Recommender systems are filters which suggest items or information that might be interesting to users. These systems analyze the past behavior of a user, build her profile that stores information about her interests, and exploit that profile to find potentially interesting items. The main limitation of this approach is that it may provide accurate but likely obvious suggestions, since recommended items are similar to those the user already knows. In this paper we investigate this issue, known as overspecialization or serendipity problem, by proposing a strategy that fosters the suggestion of surprisingly interesting items the user might not have otherwise discovered.The proposed strategy enriches a graph-based recommendation algorithm with background knowledge that allows the system to deeply understand the items it deals with. The hypothesis is that the infused knowledge could help to discover hidden correlations among items that go beyond simple feature similarity and therefore promote non-obvious suggestions. Two evaluations are performed to validate this hypothesis: an in vitro experiment on a subset of the hetrec2011-movielens-2k dataset, and a preliminary user study. Those evaluations show that the proposed strategy actually promotes non-obvious suggestions, by narrowing the accuracy loss.

Semantics-aware Graph-based Recommender Systems Exploiting Linked Open Data

The ever increasing interest in semantic technologies and the availability of several open knowledge sources have fueled recent progress in the field of recommender systems. In this paper we feed recommender systems with features coming from the Linked Open Data (LOD) cloud - a huge amount of machine-readable knowledge encoded as RDF statements - with the aim of improving recommender systems effectiveness. In order to exploit the natural graph-based structure of RDF data, we study the impact of the knowledge coming from the LOD cloud on the overall performance of a graph-based recommendation algorithm. In more detail, we investigate whether the integration of LOD-based features improves the effectiveness of the algorithm and to what extent the choice of different feature selection techniques influences its performance in terms of accuracy and diversity. The experimental evaluation on two state of the art datasets shows a clear correlation between the feature selection technique and the ability of the algorithm to maximize a specific evaluation metric. Moreover, the graph-based algorithm leveraging LOD-based features is able to overcome several state of the art baselines, such as collaborative filtering and matrix factorization, thus confirming the effectiveness of the proposed approach.

Learning Word Embeddings from Wikipedia for Content-Based Recommender Systems

In this paper we present a preliminary investigation towards the adoption of Word Embedding techniques in a content-based recommendation scenario. Specifically, we compared the effectiveness of three widespread approaches as Latent Semantic Indexing, Random Indexing and Word2Vec in the task of learning a vector space representation of both items to be recommended as well as user profiles.

Learning Preference Models in Recommender Systems

As proved by the continuous growth of the number of web sites which embody recommender systems as a way of personalizing the experience of users with their content, recommender systems represent one of the most popular applications of principles and techniques coming from Information Filtering (IF). As IF techniques usually perform a progressive removal of nonrelevant content according to the information stored in a user profile, recommendation algorithms process information about user interests – acquired in an explicit (e.g., letting users express their opinion about items) or implicit (e.g., studying some behavioral features) way – and exploit these data to generate a list of recommended items. Although each type of filtering method has its own weaknesses and strengths, preference handling is one of the core issues in the design of every recommender system: since these systems aim to guide users in a personalized way to interesting or useful objects in a large space of possible options, it is important for them to accurately capture and model user preferences. The goal of this chapter is to provide a general overview of the approaches to learning preference models in the context of recommender systems. In the first part, we introduce general concepts and terminology of recommender systems, giving a brief analysis of advantages and drawbacks for each filtering approach. Then we will deal with the issue of learning preference models, show the most popular techniques for profile learning and preference elicitation, and analyze methods for feedback gathering in recommender systems.

Combining Distributional Semantics and Entity Linking for Context-Aware Content-Based Recommendation

The effectiveness of content-based recommendation strategies tremendously depends on the representation formalism adopted to model both items and user profiles. As a consequence, techniques for semantic content representation emerged thanks to their ability to filter out the noise and to face with the issues typical of keyword-based representations. This article presents Contextual eVSM (C-eVSM), a content-based context-aware recommendation framework that adopts a novel semantic representation based on distributional models and entity linking techniques. Our strategy is based on two insights: first, entity linking can identify the most relevant concepts mentioned in the text and can easily map them with structured information sources, easily triggering some inference and reasoning on user preferences, while distributional models can provide a lightweight semantics representation based on term co-occurrences that can bring out latent relationships between concepts by just analying their usage patterns in large corpora of data.

A folksonomy-based recommender system for personalized access to digital artworks

Museums have recognized the need for supporting visitors in fulfilling a personalized experience when visiting artwork collections, and they have started to adopt recommender systems as a way to meet this requirement. Content-based recommender systems analyze features of artworks previously rated by a visitor and build a visitor model or profile, in which preferences and interests are stored, based on those features. For example, the profile of a visitor might store the names of his or her favorite painters or painting techniques, extracted from short textual descriptions associated with artworks. The user profile is then matched against the attributes of new items in order to provide personalized suggestions. The Web 2.0 (r)evolution has changed the game for personalization from “elitist” Web 1.0, written by few and read by many, to Web content potentially generated by everyone (user-generated content - UGC). One of the forms of UGC that has drawn most attention from the research community is folksonomy, a taxonomy generated by users who collaboratively annotate and categorize resources of interests with freely chosen keywords called tags. In this work, we investigate the problem of deciding whether folksonomies might be a valuable source of information about user interests in the context of recommending digital artworks. We present FIRSt (Folksonomy-based Item Recommender syStem), a content-based recommender system which integrates UGC through social tagging in a classic content-based model, letting users express their preferences for items by entering a numerical rating as well as by annotating items with free tags. Experiments show that the accuracy of recommendations increases when tags are exploited in the recommendation process to enrich user profiles, provided that tags are not used as a surrogate for the item descriptions, but in conjunction with them. FIRSt has been developed within the CHAT project “Cultural Heritage fruition & e-learning applications of new Advanced (multimodal) Technologies””, and it is the core of a bouquet of Web services designed for personalized museum tours.

Random Indexing and Negative User Preferences for Enhancing Content-Based Recommender Systems

The vector space model (VSM) emerged for almost three decades as one of the most effective approaches in the area of Information Retrieval (IR), thanks to its good compromise between expressivity, effectiveness and simplicity. Although Information Retrieval and Information Filtering (IF) undoubtedly represent two related research areas, the use of VSM in Information Filtering is much less analyzed, especially for content-based recommender systems.