Michael Ekstrand

Recommender Applications

Some of my recommender and information retrieval systems research has been application-focused. I have led two particular application-driven projects.

Context-Aware Help Search

While interning at Autodesk Research, I prototyped and studied a context-aware help search tool that would monitor the user's interactions with a software program and use the collected data to enhance searches for help resources on the web.

Our prototype and experiment were an interesting hack. We instrumented the Inkscape vector drawing program to report user activity to a context monitor process. The user then searched for help resources using a customized browser; this browser used JavaScript injection to enhance Google search results, both by incorporating context information into the result display and by mixing the results with the results of additional searches produced by query augmentation.

Michael Ekstrand, Wei Li, Tovi Grossman, Justin Matejka, and George Fitzmaurice. 2011. Searching for Software Learning Resources Using Application Context. In Proceedings of the 24th Annual ACM Symposium on User Interface Software and Technology (UIST '11). ACM, 195–204. DOI:10.1145/2047196.2047220. Acceptance rate: 25%.

Searching for Research Papers

My first recommender systems research project was a research paper recommender. The key idea of this project was to blend graph ranking algorithms (such as HITS, SALSA, and PageRank) with a collaborative or content-based filter; our goal was to make a recommender that was particularly good at producing reading lists for junior researchers.

We did a small user study indicating some modest improvement from the addition of the graph ranking. While our overall research method was sound and implemented the high-level research pipeline that I recommend, the user study we conducted in this should not be used as the starting point for new studies. Much better user study methods are now available.

Michael D. Ekstrand, Praveen Kannan, James A. Stemper, John T. Butler, Joseph A. Konstan, and John T. Riedl. 2010. Automatically Building Research Reading Lists. In Proceedings of the Fourth ACM Conference on Recommender Systems (RecSys '10). ACM, 159–166. DOI:10.1145/1864708.1864740. Acceptance rate: 19%.