A project for the class SI 509: Networks: Theory and Applications, the Delicious Recommender System was a collaborative project with Junlan Liu. It uses a hitting time algorithm to rank the network of tags and bookmarks, which is visualised in an interactive interface programmed with Processing.
A primary concern for this recommender system was encouraging exploration. Rather than a system that prompts "channel-surfing" as StumbleUpon does, we aimed to visualise the tag and bookmark space as a network that resembles the venerable tag cloud to reflect tags that would be interesting to a user; clicking on a tag would reveal three of the most interesting bookmarks for a tag, where the size of each is proportional to the strength of the recommendation.
The result is an interactive network visualisation, where each edge between tag nodes signifies co-occurence in bookmarks.
FriendFeed is a social networking site that allows users to follow the activity of others across a spectrum of other social networking sites – essentially acting as an aggregator. In this group project, we analyse and visualise two months of data centred around events for follower activity, comments, "likes", and posts.
The result of this final project for the class SI 649: Information Visualization was a set of six visualisations that highlight these usage patterns over the two month period.
Final Project for the class SI 650: Information Retrieval, this project examines the network structure formed by hyperlinks. Search engines like Clusty apply clustering to their search results, presumably based on textual similarity. This project explores the potential for using graph-theoretic community detection algorithms for assigning Wikipedia articles to clusters based solely on the network structure.
Abstract: Wikipedia, seemingly more than other Web sites, has a rich link structure that seems ripe for analysis. The communities that might exist in Wikipedia may be evident in the network's structure. This paper investigates the network structure of Wikipedia, and explores the degree to which the community structures map to the tags that del.icio.us users annotate pages with. The extent to which Wikipedia's communities follow the cluster hypothesis is explored, along with potential ways to use community structures in information retrieval.
The final project for the class SI 760: Language and Information, this project focused on assessing the differences between academic divisions as seen in the Michigan Corpus of Upper-Level Student Papers (MICUSP). The analysis compared common bigrams used between fields as well as those merely beginning the same way (e.g., "language development" vs. "language acquistion" to see differences in what fields focus on), and investigated the use of grammatical/function words, which may reveal subconscious lexical biases rather than explicit choice.
Using R and the ggplot2 package to explore a collection of NSF Research Awards Abstracts from 1990 to 2003. This is a series of analyses involving Unix tools, Perl scripting, and Natural Language Processing to massage the data, and ggplot2 to visualise it.
The final project for the class EECS 545: Machine Learning. Social websites like Reddit thrive because of their active and vibrant communities. However, there is a contingency of users that may often be hesitant to contribute because they are not confident that their posts and comments will be helpful. Therefore, this project is an attempt to build the backend of system that would give the user an indicator of the potential degree of interestingness (using karma as a proxy) of his or her post before they submit it based on historical data in order to build confidence.
With drivers performing more information and communication tasks within their cars, it is important for automobile manufacturers to design in-dash systems with usability and safety in mind to mitigate the necessity to fumble with cell phones, radio, and navigation controls. This project involved the design and testing of a unified center stack that marries a large touch screen with traditional knobs for tactile feedback. The system was also designed to work equally well for users across a wide spectrum of technological familiarity.
The Ann Arbor District Library's IT Department sought to revise its help desk trouble ticket system, which had been largely unmodified since its creation in 1997. This was necessary as more modern demands meant that the system was not supporting the process as smoothly as it potentially could. As described in our final report to the AADL:
In order to collect information on the helpdesk trouble ticket system, how it works and the current strengths and weaknesses of the systems, we conducted interviews with IT department staff members as well as with non IT department staff members. We interviewed both groups so we could get a sense of how tickets are put into the system and how tickets in the system are resolved. Once we conducted interviews, we met as a group to discuss the interviews and make notes of the important information obtained. In these meetings, we also constructed models that showed how communication was accomplished during the helpdesk trouble ticket process and also how a helpdesk trouble ticket went through the system from the perspective of the interviewee. We also gathered artifacts such as forms or screenshots of programs that were used during the helpdesk trouble ticket process. Once we made our models based on each interview, we took the individual models and created consolidated models that combined the individual models of the same type. We then took the notes that we had made during the post interview meetings and grouped similar notes together in order to look at the helpdesk trouble ticket process from different perspectives. We utilized these notes and models to brainstorm recommendations. We [then] examined these recommendations to determine if they were feasible.
The Literature Resource Center is a comprehensive online literature database aimed primarily at high school students and undergraduates in need of credible literature references. In a semester-long group project, we evaluated Literature Resource Center in terms of usability and feature importance via interviews, surveys, comparative and heuristic evaluations, and user testing. Shown here is a sample of Literature Resource Center's interaction map—created with OmniGraffle Pro.