Shaghayegh Sahebi

Recommender Systems: Sources of Knowledge and Evaluation Metrics

Recommender or Recommendation Systems (RS) aim to help users dealing with information overload: finding relevant items in a vast space of resources. Research on RS has been active since the development of the first recommender system in the early 1990s, Tapestry, and some articles and books that survey algorithms and application domains have been published recently. However, these surveys have not extensively covered the different types of information used in RS (sources of knowledge), and only a few of them have reviewed the different ways to assess the quality and performance of RS. In order to bridge this gap, in this chapter we present a classification of recommender systems, and then we focus on presenting the main sources of knowledge and evaluation metrics that have been described in the research literature.

Cross-Domain Collaborative Recommendation in a Cold-Start Context: The Impact of User Profile Size on the Quality of Recommendation

Most of the research studies on recommender systems are focused on single-domain recommendations. With the growth of multi-domain internet stores such as iTunes, Google Play, and Amazon.com, an opportunity to offer recommendations across different domains become more and more attractive. But there are few research studies on cross-domain recommender systems. In this paper, we study both the cold-start problem and the hypothesis that cross-domain recommendations provide more accuracy using a large volume of user data from a true multi-domain recommender service. Our results indicate that cross-domain collaborative filtering could significantly improve the quality of recommendation in cold start context and the auxiliary profile size plays an important role in it.

Modeling and analysis of Reo connectors using alloy

Reo is an exogenous coordination language based on a calculus of channel composition. Different formal models have been developed for this language. In this paper, we present a new approach to modeling and analysis of Reo connectors using Alloy which is a lightweight modeling language based on first-order relational logic. We provide a reusable library of Reo channels in Alloy that can be used to create a model of a Reo connector in Alloy. The model is simple and reflects the original structure of the connector. Furthermore, the model of a connector can be reused as a component for constructing more complex connectors. Using the Alloy Analyzer tool, properties expressed as predicates can be verified by automatically analyzing the execution traces of the Reo connector. We handle the context-sensitive behavior of channels as well as optional constraints on the interactions with environment. Our compositional model can be used as an alternative to other existing approaches, and is supported by a well known tool with a rich set of features such as counterexample generation.

It Takes Two to Tango: An Exploration of Domain Pairs for Cross-Domain Collaborative Filtering

As the heterogeneity of data sources are increasing on the web, and due to the sparsity of data in each of these data sources, cross-domain recommendation is becoming an emerging research topic in the recent years. Cross-domain collaborative filtering aims to transfer the user rating pattern from source (auxiliary) domains to a target domain for the purpose of alleviating the sparsity problem and providing better target recommendations. However, the studies so far have either focused on a limited number of domains that are assumed to be related to each other (such as books and movies), or a division of the same dataset (such as movies) into different domains based on an item characteristic (such as genre). In this paper, we study a broad set of domains and their characteristics to understand the factors that affect the success or failure of cross-domain collaborative filtering, the amount of improvement in cross-domain approaches, and the selection of best source domains for a specific target domain. We propose to use Canonical Correlation Analysis (CCA) as a significant major factor in finding the most promising source domains for a target domain, and suggest a cross-domain collaborative filtering based on CCA (CD-CCA) that proves to be successful in using the shared information between domains in the target recommendations.

An Enhanced Similarity Measure for Utilizing Site Structure in Web Personalization Systems

The need for recommendation systems to ease user navigations has become evident by growth of information on the Web. There exist many approaches of learning for Web usage-based recommendation systems. In hybrid recommendation systems, other knowledge resources, like content, semantics, and hyperlink structure of the Web site, have been utilized to enhance usage-based personalization systems. In this study, we introduce a new structure-based similarity measure for user sessions. We also apply two clustering algorithms on this similarity measure to compare it to cosine and another structure-based similarity measures. Our experiments exhibit that adding structure information, leveraging the proposed similarity measure, enhances the quality of recommendations in both methods.

Predicting Student Performance in Solving Parameterized Exercises

In this paper, we compare pioneer methods of educational data mining field with recommender systems techniques for predicting student performance. Additionally, we study the importance of including students’ attempt time sequences of parameterized exercises. The approaches we use are Bayesian Knowledge Tracing (BKT), Performance Factor Analysis (PFA), Bayesian Probabilistic Tensor Factorization (BPTF), and Bayesian Probabilistic Matrix Factorization (BPMF). The last two approaches are from the recommender system’s field. We approach the problem using question-level Knowledge Components (KCs) and test the methods using cross-validation. In this work, we focus on predicting students’ performance in parameterized exercises. Our experiments shows that advanced recommender system techniques are as accurate as the pioneer methods in predicting student performance. Also, our studies show the importance of considering time sequence of students’ attempts to achieve the desirable accuracy.

Recommending research colloquia: a study of several sources for user profiling

The study reported in this paper is an attempt to improve content-based recommendation in CoMeT, a social system for sharing information about research colloquia in Carnegie Mellon and University of Pittsburgh campuses. To improve the quality of recommendation in CoMeT, we explored three additional sources for building user profiles: tags used by users to annotate CoMeTs talks, partial content of CiteULike papers bookmarked by users, and tags used to annotate CiteULike papers. We also compare different tag integration models to study the impact of information fusion on recommendations outcome. The results demonstrate that information encapsulated in CiteULike bookmarks generally helps to improve several aspects of recommendation. The addition of tags by fusing them into keyword profiles helps to improve precision and novelty of recommendation, but may harm systems ability to recommend generally interesting talks. The effects of tags and bookmarks appeared to be stackable.

What Influences the Decision to Participate in Audience-bounded Online Communities?

Building online communities to support small, audience-bounded offline social structures such as neighborhoods or organizations can be challenging. Due to the small size of their potential audience, the contribution volume is likely to be insufficient to maintain a sustainable community-driven system. In our research, we are interested in studying how different characteristics of the offline structure of these communities can influence their online behavior. Particularly, we analyzed participation of researchers in a social system for conferences. Our analysis shows that centrality in the academic social structure is a significant predictor of the likelihood to accept an invitation to participate in the system. These results suggest that an understanding of the users’ offline context can increase the effectiveness of user engagement strategies in an online context.