Lightweight Multi-Role Recommendation System in TV live-streaming

Recommendation system has achieved great success in e-commerce business due to its strong modeling capacity for intent user preferences. Especially with more attention paid to live sale, recommendation shows obvious improvement in product CTR and users’ experience. For the TV terminal, the live sale is a mature business. However, since one TV terminal corresponds to several family members, TV live-streaming recommendation is challenging work. It needs to distinguish different family members and make targeted recommendations. Meanwhile, the recommendation system has the ability to quickly calculate and respond to user needs. This paper proposes a personalized TV recommendation system and a lightweight multi-role recommendation model for TV terminals, which can automatically identify multi-user preferences based on feedback from a single terminal. The accuracy of the recommendation system is verified by a public dataset.