Combining Seq2Seq Models with Collaborative Filtering Techniques for Explainable Recommendation

Layer 6 builds state-of-the-art recommender systems for TDs online businesses. Collaborative Filtering (CF) is a common recommendation approach that widely adopted by many e-commerce platforms. Modern CF algorithms attempt to exploit latent features to represent users and items, which can lead to the lack of transparency of the recommender systems. In order to build a trustworthy recommender system, it is necessary to provide explanations associated with each recommendation so that users can understand why a specific item has been suggested. The proposed research project would explore the potential of combining sequence-to-sequence (seq2seq) natural language generation models with collaborative filtering techniques into a multi-task learning setting. The result would be a recommender system that could predict customers needs with a high degree of accuracy, while producing effective, personalized explanations. Such explainability would build trust between the recommender system and TDs customers, and accordingly drive sales and customer loyalty.

Faculty Supervisor:

Richard Zemel

Student:

Yichao Lu

Partner:

Layer 6 AI

Discipline:

Computer science

Sector:

Information and communications technologies

University:

Program: