On behalf of UCSB Geography’s STKO lab members Yingjie Hu, Grant McKenzie, Jiue-An Yang, Song Gao, Amin Abdalla, and Professor Krzysztof Janowicz (Director), Yingjie was presented with 2nd prize for a presentation of “A Linked-Data-Driven Web Portal for Learning Analytics: Data Enrichment, Interactive Visualization, and Knowledge Discovery” in the 2014 LAK Data Challenge which was sponsored by the Association for Computing Machinery (ACM) International Conference on Learning Analytics and Knowledge (LAK), held in Indianapolis, IN, March 24-28.
The LAK Data Challenge presents machine-readable metadata about the publications, researchers, and conferences in the field of learning analytics. The mission of this challenge is to facilitate the understanding of the LAK discipline, such as the evolution of research topics, as well as the collaborative relations among researchers. The work presented by Geography’s Space and Time Knowledge Organization Lab (STKO Lab) is an online scientometrics workbench, called DEKDIV, which offers a number of powerful functional modules to help users explore the LAK dataset and gain more insights. “As a group in the geography department, our work features the functionalities that help discover spatiotemporal patterns from the locations of research institutes and conferences, as well as the geographic relations among them,” Yingjie explains. He goes on to point out that Semantic Web technologies and Natural Language Process (NLP) methods are also employed to extract meaningful information from the full text of each paper to examine the popular topics in this field. Specifically, this online workbench highlights the following functionalities:
- Discovering Geographic Patterns. The Collaborative Institute module (figure 1) represents the coauthorships among research institutes throughout the world and detects some collaboration patterns (e.g., domestic or international collaboration). The Conference Participant module (figure 2) displays the geographic distribution of conference participants throughout the world. The Reference Map module shows the locations of the first authors whose papers have been cited.
- Topic Modeling and Researcher Similarity. As the LAK dataset also provides full text for each paper, we employed the Latent Dirichlet Allocation (LDA) to identify the topics contained in the text. We then combine all the papers of each researcher and use LDA to identify their research interests. Consequently, researchers’ interests can be represented as a probability distribution of topics, and we derive their similarity using cosine distance metrics. The similarities of researchers are then displayed using Multidimensional Scaling (MDS), and a screenshot can be seen in figure 3.
- Recommending Reviewers and Finding Potential Collaborators. Based on the research interests of scholars in the field of LAK, our workbench also provides the functionality to recommend reviewers to a newly submitted paper (figure 4). The system can automatically extract the key concepts from the abstract of the paper and then find researchers who have similar publications, while avoiding interest conflicts (i.e., excluding researchers who have coauthored together before). The Potential Collaborator module (figure 5) can help find researchers who have similar research interests but who may have never collaborated before.
“While we are new to the field of learning analytics, our work received good feedback and comments from the LAK community,” Yingjie noted. “The challenge organization committee commended us for introducing a geographic perspective to the LAK dataset, which helped obtain novel insights.”
Editor’s note: Many thanks to Yingjie for contributing this material.