Danielle H. Lee

Fighting Information Overflow with Personalized Comprehensive Information Access: A Proactive Job Recommender

Searching for jobs online is an information intensive activity, because thousands of jobs are posted on the Web daily and it takes a great deal of effort to find the right position. Job search sites require recommender systems to meet diversified information needs: Job seekers who have well-defined careers try to focus on relevant open positions while students who have general and evolving interests want to follow the dominant trends of the job market in order to plan their career path. In this paper, we introduce a comprehensive job recommender system. From the users perspective, four different kinds of recommendations are implemented. Users of this system can retrieve open jobs with different methods, ranging from exploring to searching.

An open integrated exploratorium for database courses

In this paper, we present an open architecture that combines different SQL learning tools in an integrated Exploratorium for database courses. The integrated Exploratorium provides a unique learning environment that allows database students to take complimentary advantages of multiple advanced learning tools.

Social tagging is no substitute for controlled indexing: A comparison of Medical Subject Headings and CiteULike tags assigned to 231,388 papers

Social tagging and controlled indexing both facilitate access to information resources. Given the increasing popularity of social tagging and the limitations of controlled indexing (primarily cost and scalability), it is reasonable to investigate to what degree social tagging could substitute for controlled indexing. In this study, we compared CiteULike tags to Medical Subject Headings (MeSH) terms for 231,388 citations indexed in MEDLINE. In addition to descriptive analyses of the data sets, we present a paper-by-paper analysis of tags and MeSH terms: the number of common annotations, Jaccard similarity, and coverage ratio. In the analysis, we apply three increasingly progressive levels of text processing, ranging from normalization to stemming, to reduce the impact of lexical differences. Annotations of our corpus consisted of over 76,968 distinct tags and 21,129 distinct MeSH terms. The top 20 tags/MeSH terms showed little direct overlap. On a paper-by-paper basis, the number of common annotations ranged from 0.29 to 0.5 and the Jaccard similarity from 2.12% to 3.3% using increased levels of text processing. At most, 77,834 citations (33.6%) shared at least one annotation. Our results show that CiteULike tags and MeSH terms are quite distinct lexically, reflecting different viewpoints/processes between social tagging and controlled indexing.

Re-assessing the Value of Adaptive Navigation Support in E-Learning Context

In a recent study, we discovered a new effect of adaptive navigation support in the context of E-learning: the ability to motivate students to work more with non-mandatory educational content. The results presented in this paper extend the limits of our earlier findings. We describe the implementation of adaptive navigation support for the SQL domain, and report the results of the classroom evaluation of our approach. Among other issues, we investigate whether the use in parallel of two different types of navigation support could change the nature or the magnitude of the previously observed effect. Our study confirms the motivational value of navigation support in the new domain. We observe the increase of this effect after adding the concept-based navigation layer to the existing topic-based adaptive navigation service. The results of the navigational pattern analysis allow us to determine the major source of this increase.

A comparison of meSH terms and CiteULike social tags as metadata for the same items

In this paper, we examine the degree of difference between two types of metadata for biomedical articles generated by different groups of people. The first type of metadata is social tags, which are assigned to articles by their readers using uncontrolled vocabulary. The second type is index terms, which are assigned by professionally trained indexers and domain experts using a controlled vocabulary. When the two kinds of metadata are assigned to the same item, we may expect that they overlap to a large extent and could substitute for one another. In this study, we compared social tags and index terms for a set of papers that appear both in CiteULike and MEDLINE, and assessed their differences. Due to the idiosyncratic nature of social tags, we preprocessed the tags through normalization, stop-word removal, stemming and spell-checking. Our results show that social tags and Medical Subject Heading (MeSH) index have little overlap and embody largely heterogeneous understanding of items.

Recommendations Based on Social Links

The goal of this chapter is to give an overview of recent works on the development of social link-based recommender systems and to offer insights on related issues, as well as future directions for research. Among several kinds of social recommendations, this chapter focuses on recommendations, which are based on users’ self-defined (i.e., explicit) social links and suggest items, rather than people of interest. The chapter starts by reviewing the needs for social link-based recommendations and studies that explain the viability of social networks as useful information sources. Following that, the core part of the chapter dissects and examines modern research on social link-based recommendations along several dimensions. It concludes with a discussion of several important issues and future directions for social link-based recommendation research.