Undergraduates to Develop Search Personalization for Visual History Archive in Annual UCLA Program
Fron left: Kira Parker, Rachel A. Lewis, Eric Gao, Lucia Li, Ehsan Ebrahimzadeh
Four undergraduate students from around the world are hard at work developing new search capabilities for the Visual History Archive as part of the annual Research in Industrial Projects for Students (RIPS) program hosted by the UCLA Institute for Pure and Applied Mathematics (IPAM).
The RIPS program pairs small groups of students from around the world with sponsor organizations from the industry or public sectors. Over nine weeks during the summer, each group must offer solutions for an applied mathematics problem within their sponsor organization. At the end of the program, they present their findings to the RIPS faculty and their sponsor organization.
USC Shoah Foundation has been a sponsor since 2010, providing large data sets from the Visual History Archive so that its assigned group each year can develop potential improvements for the Visual History Archive user experience, typically centered on how users search through the archive of 55,000 testimonies and 64,000 keywords.
Past groups have developed algorithms for improving search relevance, finding ways to automatically link testimonies with relevant Wikipedia pages, and visualizing relationships between keywords, to name a few.
This year’s group is Rachel A. Lewis, Armstrong State University (project manager); Kira Parker, University of Utah; Lucia Li, Wellesley College, and Eric Gao, University of Cambridge. Their academic mentor is Ehsan Ebrahimzadeh, a PhD candidate in electrical engineering at UCLA.
The group will work on developing an algorithm to return suggestions or recommendations of similar testimonies whenever a user watches a testimony in the Visual History Archive, the way sites like Amazon and Netflix offer suggestions to users based on their preferences, viewing history, and behavior of similar users. The students plan to use a mix of memory-based and model-based collaborative filtering approaches to personalize search results.
“We take the user’s activity history and the keywords for each video as input, represent each user’s interest for a given keyword by a nonnegative score, use some similarity measurement to find the videos that best represent the user’s interest, and present these videos to the user,” the group explains in its work statement.
They measure “interest” in terms of the time a user spends watching a testimony; watching over 80 percent indicates high interest and less than 20 percent indicates low interest. Their algorithm will combine a user’s interest for certain keywords based on the videos they watch with videos that demonstrate similarity to the user’s interest and recommend those videos to the user.
USC Shoah Foundation Senior Software Architect Mills Chang is once again serving as the group’s sponsoring mentor. He said that Sam Gustman, Chief Technology Officer of USC Shoah Foundation, had the idea for this year’s RIPS group because the Institute has never analyzed user behavior before and also wanted to learn more about machine-learning techniques and artificial intelligence.
“That’s why we came up with this idea, to provide more personalized results than just our restricted search algorithm,” Chang said. “Usually the results [from RIPS] come out better than we expected. They usually come out with an idea that’s more than what we thought of so we’re open to seeing what they come back with.”
Gao said they settled on their methodology after researching programming languages and machine learning techniques and determining which would work best for USC Shoah Foundation, their client. Offering personalized recommendations has already proven helpful in most Internet users’ everyday lives, so they thought Visual History Archive users could also benefit from receiving suggestions based on their viewing history.
“When you go to YouTube right now you have a whole list of things you like to watch so that’s always helpful,” Lewis said. “Being able to do the same thing, maybe helping a researcher find something they need earlier, you can have a better way to find something that’s unknown to the user.”
Although the group didn’t have much background in the kinds of techniques they will be using for the project, they picked it up fast and have already gathered great preliminary results, according to Ebrahimzadeh. Their challenge will be to develop a system that offers users recommendations that are not only relevant, but also that they may have never even thought of on their own.
The group members agreed that working with USC Shoah Foundation for the RIPS program is a good introduction to the world of applied and industrial mathematics that they hope to one day be part of. This is their first experience working with a real-world organization to solve a problem using their math and computer science skills, and they find it rewarding to see their work have a positive impact outside academia.
“It’s interesting to do math stuff that ties directly to somebody else’s problem,” Parker said. “I’m not just doing research; hopefully someone else might find it useful.”