Marco Degemmis

A content-collaborative recommender that exploits WordNet-based user profiles for neighborhood formation

Collaborative and content-based filtering are the recommendation techniques most widely adopted to date. Traditional collaborative approaches compute a similarity value between the current user and each other user by taking into account their rating style, that is the set of ratings given on the same items. Based on the ratings of the most similar users, commonly referred to as neighbors, collaborative algorithms compute recommendations for the current user. The problem with this approach is that the similarity value is only computable if users have common rated items. The main contribution of this work is a possible solution to overcome this limitation. We propose a new content-collaborative hybrid recommender which computes similarities between users relying on their content-based profiles, in which user preferences are stored, instead of comparing their rating styles. In more detail, user profiles are clustered to discover current user neighbors. Content-based user profiles play a key role in the proposed hybrid recommender. Traditional keyword-based approaches to user profiling are unable to capture the semantics of user interests. A distinctive feature of our work is the integration of linguistic knowledge in the process of learning semantic user profiles representing user interests in a more effective way, compared to classical keyword-based profiles, due to a sense-based indexing. Semantic profiles are obtained by integrating machine learning algorithms for text categorization, namely a naïve Bayes approach and a relevance feedback method, with a word sense disambiguation strategy based exclusively on the lexical knowledge stored in the WordNet lexical database. Experiments carried out on a content-based extension of the EachMovie dataset show an improvement of the accuracy of sense-based profiles with respect to keyword-based ones, when coping with the task of classifying movies as interesting (or not) for the current user. An experimental session has been also performed in order to evaluate the proposed hybrid recommender system. The results highlight the improvement in the predictive accuracy of collaborative recommendations obtained by selecting like-minded users according to user profiles.

WordNet-based user profiles for neighborhood formation in hybrid recommender systems

Recommender systems help to reduce information overload and provide customized information access for targeted domains. Such systems take input from users and, based on their needs and preferences, provide personalized advices that help people to filter useful information. Collaborative filtering and content-based filtering are the most widely recommendation techniques adopted to date. The paper presents a new hybrid recommendation technique based on the combination of classic collaborative filtering and user profiles inferred using content-based methods.

Ontologically-Enriched unified user modeling for cross-system personalization

Personalization today has wide spread use on many Web sites. Systems and applications store preferences and information about users in order to provide personalized access. However, these systems store user profiles in proprietary formats. Although some of these systems store similar information about the user, exchange or reuse of information is not possible and information is duplicated. Additionally, since user profiles tend to be deeply buried inside such systems, users have little control over them. This paper proposes the use of a common ontology-based user context model as a basis for the exchange of user profiles between multiple systems and, thus, as a foundation for cross-system personalization.

Improving Social Filtering Techniques Through WordNet-Based User Profiles

Collaborative filtering algorithms predict the preferences of a user for an item by weighting the contributions of similarusers, called neighbors, for that item. Similarity between users is computed by comparing their rating styles, i.e. the set of ratings given on the sameitems. Unfortunately, similarity between users is computable only if they have common rated items. The main contribution of this paper is a (content-collaborative) hybrid recommender system which overcomes this limitation by computing similarity between users on the ground of their content-based profiles. Traditional keyword-based profiles are unable to capture the semanticsof user interests, due to the natural language ambiguity. A distinctive feature of the proposed technique is that a statistical model of the user interests is obtained by machine learning techniques integrated with linguistic knowledge contained in the WordNet lexical database. This model, called the semanticuser profile, is exploited by the hybrid recommender in the neighborhood formation process. The results of an experimental session in a movie recommendation scenario demonstrate the effectiveness of the proposed approach.

Learning semantic user profiles from text

This paper focuses on the problem of choosing a representation of documents that can be suitable to induce more advanced semantic user profiles, in which concepts are used instead of keywords to represent user interests. We propose a method which integrates a word sense disambiguation algorithm based on the WordNet IS-A hierarchy, with two machine learning techniques to induce semantic user profiles, namely a relevance feedback method and a probabilistic one. The document representation proposed, that we called Bag-Of-Synsets improves the classic Bag-Of-Words approach, as shown by an extensive experimental session.

WordNet-Based word sense disambiguation for learning user profiles

Nowadays, the amount of available information, especially on the Web and in Digital Libraries, is increasing over time. In this context, the role of user modeling and personalized information access is increasing. This paper focuses on the problem of choosing a representation of documents that can be suitable to induce concept-based user profiles as well as to support a content-based retrieval process. We propose a framework for content-based retrieval, which integrates a word sense disambiguation algorithm based on a semantic similarity measure between concepts (synsets) in the WordNet IS-A hierarchy, with a relevance feedback method to induce semantic user profiles. The document representation adopted in the framework, that we called Bag-Of-Synsets (BOS) extends and slightly improves the classic Bag-Of-Words (BOW) approach, as shown by an extensive experimental session.

Personalization for the web: learning user preferences from text

As more information becomes available electronically, tools for finding information of interest to users become increasingly important. Information preferences vary greatly across users, therefore, filtering systems must be highly personalized to serve the individual interests of the user. Our research deals with learning approaches to build user profiles that accurately capture user interests from content (documents) and that could be used for personalized information filtering. The learning mechanisms analyzed in this paper are relevance feedback and a naive Bayes method. Experiments conducted in the context of a content-based profiling system for movies show the pros and cons of each method.

A RDF-based framework for user profile creation and management

The semantic evolution of the Web has an heavy impact on traditional systems, as the ability to use a formal interoperable language simplifies information exchange between different systems. In order to foster information exchange and to easily connect new functionalities to semantic knowledge bases, in order to be able to use and reuse the valuable knowledge embedded in the existing systems, we designed a plugin-based framework, and used it to connect together different tools and systems developed in the LACAM laboratory. Our pilot project includes user profiling abilities coming from two components, namely Profile Extractor (PE) and Item Recommender (ITR), and storage capabilities implemented by a repository tool called RDFCore.

Learning Usage Patterns for Personalized Information Access in e-Commerce

The World Wide Web is a vast repository of information, much of which is valuable but very often hidden to the user. Currently, Web personalization is the most promising approach to remedy this problem, and Web usage mining, is considered a crucial component of any effective Web personalization system. Web usage mining techniques such as clustering and association rules, which rely on offline pattern discovery from user transactions, can be used to improve searching in the Web. We present the Profile Extractor, a personalization component based on machine learning techniques, which allows for the discovery of preferences and interests of users that have access to a Web site. More specifically, we present the module that exploits unsupervised learning techniques for the creation of communities of users and usage patterns applied to customers of an online bookshop. To support our work, we have performed several experiments and discussed the results.

A Framework for the Development of Personalized Agents

The amount of information available on the web, as well as the number of e-businesses and web shoppers is growing exponentially. Customers spend a lot of time to browse the net in order to find relevant product information. One way to overcome this problem is to use dialoguing agents that exploit the knowledge stored in user profiles in order to generate personal recommendations. This paper presents a general framework designed according to this idea in order to develop intelligent e-business applications.