The Perfect Movie: Recommendations to Suit the Mood

Computer Science Graduate Student Sudipta Kar Working on Movie Recommendation System

Picking the right movie is tough. Sometimes the mood is right for a funny movie, other times a serious one. The end result is often a Friday night of scrolling through endless options, trying to guess from limited descriptions what might hit the spot.

Sudipta Kar
Sudipta Kar, a Ph.D. student in computer science, uses natural language processing to make movie recommendations based on emotional impact.

Even after deciding, sometimes these movies offer an unwanted surprise – a comedy that turns dark or a thriller that lacks depth.

A More Detailed Recommendation System

Sudipta Kar, a University of Houston Ph.D. student in computer science, is working on a system that incorporates movie synopses with user reviews, in order to offer a more holistic recommendation based on a movie’s plot, as well as its emotional response.

“Often, the synopsis is not enough to get an idea of what is in a movie,” Kar said. “Sometimes the synopses might be more detailed than the reviews, while other times the reviews offer more information. We use a mechanism that accounts for the flow of information from the synopses and reviews.”

Recommendations Predictive of Emotional Response

In this recommendation system, movies are assigned a set of tags, which are predictive of a movie’s overall plot and emotional arc.

“These tags cover genres, feeling-related attributes, or events,” said Kar, whose research is advised by Thamar Solorio, associate professor of computer science in the College of Natural Sciences and Mathematics.

To do this, Kar uses natural language processing, which is a branch of artificial intelligence that helps computers understand, interpret, and manipulate human language. Natural language processing is what enables voice commands for smartphones, automated call services, and virtual assistants for customer service.

Natural language processing can also be used to automatically sort through reviews and synopses, coming up with words and phrases to describe a specific movie.

Trigger Words that Predict Story Attributes

“This model uses trigger words which explore different story attributes,” Kar said. “What are the attributes that make a story thought-provoking? What are the sequences that might make a movie sad?”

This leads to a recommendation system which makes predictions based on a movie’s depth and emotional impact.

Grab your popcorn and settle in – movie night is about to get a whole lot better.

- Rachel Fairbank, College of Natural Sciences and Mathematics