A team of researchers from the Georgia Institute of Technology has won the Workshop on Deep Learning for Recommender Systems’ Best Paper Award.
Presented during the 2016 ACM Recommender Systems (RecSys) conference, Georgia Tech students Hanjun Dai, Yichen Wang, and Rakshit Trivedi received the award for their paper “Recurrent Coevolutionary Feature Embedding Processes for Recommendation.”
The paper focuses on resolving the limitations of current online recommendation systems. Companies like Amazon and Netflix use these systems to suggest products and content based on consumers’ interests. However, current recommendation systems treat consumers’ information as static and are built from collective rather than individual interests.
The students, working with School of Computational Science and Engineering Assistant Professor Le Song, propose incorporating time-based and geographic dynamics into recommendation systems. These additions would allow the systems to be more proactive with suggestions and aware of related details, offering a consumer more tailored content rather than showing what others have purchased.
To achieve this, the researchers designed a novel framework that combines a recurrent neural network (RNN) over an evolving graph with a multidimensional point process (MPP) model. The RNN and MPP work in unison to mimic the human brain's ability to discern information and connect seemingly random points in time and space. These additions would make recommendation systems more robust and ultimately unlock potential uses in other disciplines.
“Our goal is to use recommendation systems beyond e-commerce, in such fields as artificial intelligence (AI), health care, and cybersecurity,” said Song. “For example, the approach we developed has the potential to help AI agents handle information better and perform more efficient chatbot-style duties. Another budding application is the potential to help create better drugs by suggesting specific chemical compounds that would work in conjunction with a person’s genetic code.”