International Journal of Pattern Recognition and Artificial Intelligence

As online information continues to grow at an exponential rate our ability to access this information effectively does not, and users are often frustrated by how difficult it is to locate the right information quickly and easily. So-called personalization technology is a potential solution to this information overload problem: by automatically learning about the needs and preferences of users, personalized information access solutions have the potential to offer users a more proactive and intelligent form of information access that is sensitive to their long-term preferences and current needs. In this paper, we document two case-studies of the use of personalization techniques to support information browsing and search. In addition, we consider the inevitable privacy issues that go hand-in-hand with profiling and personalization techniques and highlight the importance of striking the right balance between privacy and personalization when it comes to the development and deployment of practical systems.

Personalization is the ability to retrieve information content related to users' profile and facilitate their information-seeking activities. Several environments, such as the Web, take advantage of personalization techniques because of the large amount of available information. For this reason, there is a growing interest in providing automated personalization processes during the human-computer interaction.

In this paper we introduce a new approach for user modeling, which grounds in the Search of Associative Memory (SAM) theory. By means of implicit feedback techniques, the approach is able to unobtrusively recognize user needs and monitor the user working context in order to provide important information useful to personalize traditional search tools and implement recommender systems. Experimental results based on precision and recall measures indicate improvements in comparison with traditional user models.

We propose a technique for user modeling in Adaptive Hypermedia (AH) that is unobtrusive at both the level of observable behavior and that of cognition. Unobtrusive user modeling is complementary to transparent user modeling. Unobtrusive user modeling induces user models appropriate for Educational AH (EAH) based on metaphors characterized by a shared locus of control over learning, such as small group learning. Transparent user modeling, on the other hand, induces user models appropriate for EAH based on metaphors characterized by an independent locus of control, such as social navigation.

We extend an existing decomposition model of adaptation in AH to incorporate summative evaluation of unobtrusive user modeling. Summative evaluation examines the effects or outcomes of a system, while formative evaluation is used to improve a system under development. We also introduce a separate, two-stage model for formative evaluation of unobtrusive user modeling. We then present results from a field study for the first of the two developmental stages and describe the current field study of the second developmental stage. Finally, we compare unobtrusive and transparent user modeling, and explain the role of each.

This paper presents a user model server based on Web Services. User model servers are very important because they allow reusability of user modeling reasoning mechanisms which are typically very complex and difficult to construct from scratch. In this paper we show how the potential of interoperability, reusability and component sharing offered by the technology of Web Services have been exploited in the design of a user model server that performs decision making. The reasoning of the user modeling is based on a multi-criteria decision making theory and has been implemented as a Web Service to provide intelligent assistance to users over the Web. Reusability has been shown through the successful application of the user model server into two different applications: an e-mail and a file manager application.

Collaborative Filtering (CF) is currently one of the most popular and most widely used personalization techniques. It generates personalized predictions based on the assumption that users with similar tastes prefer similar items. One of the major drawbacks of the CF from the computational point of view is its limited scalability since the computational effort required by the CF grows linearly both with the number of available users and items. This work proposes a novel efficient variant of the CF employed over a multidimensional content-addressable space. The proposed approach heuristically decreases the computational effort required by the CF algorithm by limiting the search process only to potentially similar users. Experimental results demonstrate that the proposed heuristic approach is capable of generating predictions with high levels of accuracy, while significantly improving the performance in comparison with the traditional implementations of the CF.

This paper considers the information that can be captured about users and groups from a collaborative filtering dataset. The aims of the paper are to create a user model and to use this model to explain the performance of a collaborative filtering approach. A number of user and group features are defined and the performance of a collaborative filtering system in producing recommendations for users with different feature values is tested. Graph-based representations of the collaborative filtering space are presented and these are used to define some of the user and group features as well as being used in a recommendation task.

Recommender systems have already been engaging multiple criteria for the production of recommendations. Such systems, referred to as multicriteria recommenders, demonstrated early the potential of applying Multi-Criteria Decision Making (MCDM) methods to facilitate recommendation in numerous application domains. On the other hand, systematic implementation and testing of multicriteria recommender systems in the context of real-life applications still remains rather limited. Previous studies dealing with the evaluation of recommender systems have outlined the importance of carrying out careful testing and parameterization of a recommender system, before it is actually deployed in a real setting. In this paper, the experimental analysis of several design options for three proposed multiattribute utility collaborative filtering algorithms is presented for a particular application context (recommendation of e-markets to online customers), under conditions similar to the ones expected during actual operation. The results of this study indicate that the performance of recommendation algorithms depends on the characteristics of the application context, as these are reflected on the properties of evaluations' data set. Therefore, it is judged important to experimentally analyze various design choices for multicriteria recommender systems, before their actual deployment.

Recommender applications support decision-making processes by helping online customers to identify products more effectively. Recommendation problems have a long history as a successful application area of Artificial Intelligence (AI) and the interest in recommender applications has dramatically increased due to the demand for personalization technologies by large and successful e-Commerce environments. Knowledge-based recommender applications are especially useful for improving the accessibility of complex products such as financial services or computers. Such products demand a more profound knowledge from customers than simple products such as CDs or movies. In this paper we focus on a discussion of AI technologies needed for the development of knowledge-based recommender applications. In this context, we report experiences from commercial projects and present the results of a study which investigated key factors influencing the acceptance of knowledge-based recommender technologies by end-users.

Implicit recommender systems provide a valuable aid to customers browsing through library corpora. We present a method to realize such a recommender especially for, but not limited to, libraries. The method is cluster-based, scales well for large collections, and produces recommendations of good quality. The approach is based on using session histories of visitors of the library's online catalog in order to generate a hierarchy of nondisjunctive clusters. Depending on the user's needs, the clusters at different levels of the hierarchy can be employed as recommendations. Using the prototype of a user interface we show that, if, for instance, the user is willing to sacrifice some precision in order to gain a higher number of documents during a specific session, he or she can do so easily by adjusting the cluster level via a slider.

With the merge of digital television (DTV) and the exponential growth of broadcasting network, an overwhelmingly amount of information has been made available to a consumer's home. Therefore, how to provide consumers with the right amount of information becomes a challenging problem. In this paper, we propose an electronic programming guide (EPG) recommender based on natural language processing techniques, more specifically, text classification. This recommender has been implemented as a service on a home network that facilitates the personalized browsing and recommendation of TV programs on a portable remote device. Evaluations of our Maximum Entropy text classifier were performed on multiple categories of TV programs, and a near 80% retrieval rate is achieved using a small set of training data.

The generalized arrival of Digital TV will lead to a significant increase in the amount of channels and programs available to end users, making it difficult to find interesting programs among a myriad of irrelevant contents. Thus, in this field, automatic content recommenders should receive special attention in the following years to improve assistance to users. Current approaches of content recommenders have significant well-known deficiencies that hamper their wide acceptance. In this paper, a new approach for automatic content recommendation is presented that considerably reduces those deficiencies. This approach, based on the so-called Semantic Web technologies, has been implemented in the AVATAR tool, a hybrid content recommender that makes extensive use of well-known standards, such as TV-Anytime and OWL. Our proposal has been evaluated experimentally with real users, showing significant increases in the recommendation accuracy with respect to other existing approaches.

Text classification is the problem of classifying a set of documents into a pre-defined set of classes. A major problem with text classification problems is the high dimensionality of the feature space. Only a small subset of these words are feature words which can be used in determining a document's class, while the rest adds noise and can make the results unreliable and significantly increase computational time. A common approach in dealing with this problem is feature selection where the number of words in the feature space are significantly reduced.

In this paper we present the experiments of a comparative study of feature selection methods used for text classification. Ten feature selection methods were evaluated in this study including the new feature selection method, called the GU metric. The other feature selection methods evaluated in this study are: Chi-Squared (χ²) statistic, NGL coefficient, GSS coefficient, Mutual Information, Information Gain, Odds Ratio, Term Frequency, Fisher Criterion, BSS/WSS coefficient. The experimental evaluations show that the GU metric obtained the best F₁ and F₂ scores. The experiments were performed on the 20 Newsgroups data sets with the Naive Bayesian Probabilistic Classifier.