The European Journal on Artificial Intelligence

Massive taxonomies for product classification are currently gaining popularity among e-commerce systems for diverse domains. For instance, Amazon.com maintains an entire plethora of hand-crafted taxonomies classifying books, movies, apparel and various other types of consumer goods. We use such taxonomic background knowledge for the computation of personalized recommendations, exploiting relationships between super-concepts and sub-concepts during profile generation. Empirical analysis, both offline and online, demonstrates our proposal's superiority over existing approaches when user information is sparse and implicit ratings prevail. Besides addressing the sparsity issue, we use parts of our taxonomy-based recommender framework for balancing and diversifying personalized recommendation lists in order to reflect the user's complete spectrum of interests. Though being detrimental to average accuracy, we show that our method improves user satisfaction with recommendation lists, in particular for lists generated using the common item-based collaborative filtering algorithm. We evaluate our method using book recommendation data, including offline analysis on 361,349 ratings and an online study involving more than 2,100 subjects.

Collaborative filtering recommender systems are typically unable to generate adequate recommendations for newcomers. Empirical evidence suggests that the incorporation of a trust network among the users of a recommender system can significantly help to alleviate this problem. Hence, users are highly encouraged to connect to other users to expand the trust network, but choosing whom to connect to is often a difficult task. Given the impact this choice has on the delivered recommendations, it is critical to guide newcomers through this early stage connection process. In this paper, we identify several classes of key figures in the trust network, namely mavens, frequent raters and connectors. Furthermore, we introduce measures to assess the influence of these users on the amount and the quality of the recommendations delivered by a trust-enhanced collaborative filtering recommender system. Experiments on a dataset from Epinions.com support the claim that generated recommendations for new users are more beneficial if they connect to an identified key figure compared to a random user.

Recommendation techniques that analyze social trust networks attracted much attention in the last few years. They recommend such items that are appreciated by trusted friends. In this paper, we explore how to use trust information for generating personalized document recommendations such as for scientific papers or for webpages. The basic idea is to jointly analyze a trust network between readers who review the documents and the reference network between the documents. We develop trust-enhanced visibility measures for measuring the quality and the importance of documents and evaluate them in simulation studies.

Conversational recommenders can help users find their most preferred item among a large range of options, a task that we call preference-based search.

Motivated by studies in the field of behavioral decision theory, we take a user centric design perspective, focusing on the trade-off between decision accuracy and user effort. We consider example-critiquing, a methodology based on showing examples to the user and acquiring preferences in the form of critiques. In our approach critiques are volunteered in a mixed-initiative interaction. Some recommendations are suggestions specifically aimed at stimulating preference expression to acquire an accurate preference model.

We propose a method to adapt the suggestions according to observations of the user's behavior. We evaluate the decision accuracy of our approach with both simulations exploiting logs of previous users of the system (in order to see how adaptive suggestions improve the process of preference elicitation) and surveys with real users where we compare our approach of example critiquing with an interface based on question-answering.

Various kinds of recommendation services open to the general public have recently been integrated into the website of the University Library of Karlsruhe as a test bed for information providers and e-commerce alike. This contribution reports on the development of RecoDiver, a graph-based user interface for behavior-based recommender systems. A Java applet integrated into the library's online catalog dynamically displays recommended further documents in a clickable graph centered around the document of interest to the user. A local view of the complete graph of recommendations is presented in a radial tree layout based on a minimum spanning tree with animated graph transitions featuring interpolations by polar coordinates to avoid crisscrossings. Further graph search tools like a selectable histogram of years of publication are available as well. This article portrays the user interface as well as the distributed web service architecture behind it and features an evaluation by user surveys showing the preference of users compared to the common lists of recommended items.

With the advances in and popularity of mobile devices, mobile service providers have a direct channel for transferring information to their subscribers, i.e., short messaging service (SMS) and multimedia messaging service (MMS). Mobile service operators can recommend new content and information to users who opt in to receive such information directly through push messages at any time or place. However, as mobile push messages sent to users can cause interruptions, such as alarms, users who receive irrelevant push messages may become dissatisfied with their mobile Web service and even their service provider.

In this paper, we propose a mobile content recommender system for sending personalized mobile push messages with content that users are likely to find relevant. This system learns users' preferences from contents and keywords in their usage logs and recommends items that match these preferences or those of similar users. We analyzed (a) customer feedback on personalized content dissemination, and (b) the relationship between customer feedback and mobile Web usage of customers subscribing to a Korean mobile service provider. Push messages with personalized recommendations resulted in more positive feedback from customers, and the mobile Web usage of these customers increased.

Museums offer vast amounts of information, but a visitor's receptivity and time are typically limited, providing the visitor with the challenge of selecting the (subjectively) interesting exhibits to view within the available time. Mobile, electronic handheld guides offer the opportunity to improve a visitor's experience by recommending exhibits of interest, and adapting the delivered content. The first step in this personalisation process is the prediction of a visitor's activities and interests. In this paper we study non-intrusive, adaptive user modelling techniques that take into account the physical constraints of the exhibition layout. We present two collaborative models for predicting a visitor's next locations in a museum, and an ensemble model that combines the predictions of these models. The three models were trained and tested on a small dataset of museum visits. Our results are encouraging, with the ensemble model yielding the best performance overall.

We propose a novel hybrid recommendation model in which user preferences and item features are described in terms of semantic concepts defined in domain ontologies. The concept, item and user spaces are clustered in a coordinated way, and the resulting clusters are used to find similarities among individuals at multiple semantic layers. Such layers correspond to implicit Communities of Interest and enable enhanced recommendations.

Practice shows that game-playing programs using minimax search perform better when searching deeper. Mathematical analyses, however, showed the opposite. This paradox was termed minimax pathology. Our real-valued minimax model demonstrates that appropriate modeling of the heuristic error is enough to eliminate the pathology. We examine the conditions under which the pathology appears and explain the mechanism that makes minimax otherwise beneficial. The reasons for the pathology in single-agent search are also addressed.

This thesis presents a basic architecture named DLA (Distributed and Layered Architecture) to support navigation in unstructured dynamic environments for any autonomous mobile robot. DLA supports intuitive adaptation to physically different agents and simple expansion of their capacities via addition of new modules. DLA works by combining the responses of different deliberative and reactive algorithms through the interaction of freely distributed processes in an asynchronous way. This architecture provides transparency to the user through a high simplicity and portability.

In relational learning, one learns patterns from relational databases, which usually contain multiple tables that are interconnected via relations. Thus, an example for which a prediction is to be given may be related to a set of objects that are possibly relevant for that prediction. Relational classifiers differ with respect to how they handle these sets: some use properties of the set as a whole (using aggregation), some refer to properties of specific individuals, however, most classifiers do not combine both. This imposes an undesirable bias on these learners. This dissertation describes a learning approach that avoids this bias, using complex aggregates, i.e., aggregates that impose selection conditions on the set to aggregate on.