Look Who it Was Random video

Gagandeep Juneja - Recommendation System beyond traditional Collaborative filtering

1,497 views · Published 18 July 2015 · 43:02 · Indexed 20 September 2026

Channel: Hasgeek TV · 2015 · Science & Technology

Watch on YouTube

Though Collaborative filtering works quite well for companies like NetFlix but here in Snapdeal we are catering 12M huge product catalog and more than 100 categories which again comprised of 20-30 subcategories each. For us only Collaborative filtering doesn’t work well, because of the wide catalog and implicit feedback capturing instead of explicit and hence we developed a recommendation system which considers various other factors beyond collabarative Filtering.

In this session I would be discussing other factors (mentioned below) and their mathematical models that we have considered while building custom recommendation system for generating more personalized and relevant recommendations.

User Category Affinity (to some more granular level)
Content based product similarity
product which goes well with already bought products.
predicting the repurchase of already purchased products.
Suggesting trending products based on user’s affinity.
Capturing user’s feedback (implicit) to our served recommendations and use to improve relevancy.
Collaborative filtering (we have also used this but with some weight-age)
Finally I would be concluding session with technical challenges in building scalable recommendation system with massive datasets and serving these recommendations in realtime.

More from this channel