andrewlin-umich / umich-events-classifier

A recommendation system for UMich events

UMich-Events-Classifier

A recommendation system for UMich events

The University of Michigan--Ann Arbor posts campus events on events.umich.edu. Every day, there are 80 events to choose from on average. Events are grouped into approximately 20 “event types” (e.g. Exhibition, Performance, Ceremony/Service, etc.) and tagged from among hundreds of subjects (e.g. Fitness, Books, Games, South Asia, etc.). Users can browse or keyword search to find events of interest; they can also filter results by event type or tag. However, our team feels the current system architecture does not effectively link users to event information. With a large number of events, event types, and tags to choose from, users undoubtedly retrieve events in which they are disinterested while missing many others of interest. Even when users isolate an event of interest, they must then manually compare the event date/time against their calendar. In combination, these factors defeat the purpose of the events.umich.edu page: to conveniently connect users with events for their enrichment or entertainment. Unlike formally teaching or learning in the university community, extracurricular event choice should be simple and fast.

To this end, our team has developed an event recommendation application for the University of Michigan--Ann Arbor. We have distilled event types and tags into 8 categories: Academics, Arts, Career, Workshops, Sports & Health, Community Service, Global Learning and Other. Using these categories, our team trained an algorithm to classify events based on their textual descriptions. We then applied the classifier to contemporary events. Finally, we developed a graphical user interface using Python’s TKinter package. Through the graphical user interface and by providing access to their Google calendar (we used the Google Calendar API to filter out events conflicting with the user’s schedule), users can select categories of interest and receive daily event recommendations compatible with their schedules.

Collaborated with Kate Barron, Deepak Krishnan, and Omkar Sunkersett.