Rich Recommendations

Item-to-item and user-to-item recommendations are prevalent on most ecommerce websites and digital content related mobile applications. At Kobo, we strive to constantly improve our recommendation system, which is based on co-purchase patterns on Kobos website or through Kobo eReaders and mobile apps. This internship is to explore improving the system along several dimensions: incorporating additional data sources, such as users ratings and reviews, and books that users have sideloaded onto devices (such as ePubs obtained from non-Kobo platforms); improving the core algorithm by introducing richer representations using deep learning; building recommendations for non-book, high-level concepts such as authors and series; building fine-grained recommendations based on subsets of the catalog to support personalization of item lists. All these explorations will be tested using A/B testing methodology, and expected results are determination of whether such ideas improve KPI metrics and full productization of the projects that lead to improvements in these KPIs.

Faculty Supervisor:

Brendan Frey

Student:

Hanyang Li

Partner:

Kobo Inc.

Discipline:

Computer science

Sector:

Media and communications

University:

University of Toronto