Client: A Media company with 40mn+ MAUs
Objective:
Provide a personalized feed for users for better engagement
Team: Sr. Data Scientist, Data analyst, Backend Team, Frontend Team
The project kicked off with setting up a proper data pipeline. Since the data was placed at different places in hadoop clusters which was primarily the source db, a separate db was set up for analytics. 4 different models were utilized to create a recommendation engine based on users behavior, aggregate user data, trending stories and more : Personalized, Collaborative, Trending & Serendipity. Entity recognition was also done as the content was not tagged properly.
Project Execution
An A/B testing framework was set up and various models were exposed to 1%-5% of the users & based on the feedback models were finetuned.
Front end, backend and Data science teams worked collaboratively to ensure performance was high and page speed didn’t have significant impact.
Entire project was delivered in phases for faster gtm.
Impact: The overall CTRs on the feed went up by 225% compared to earlier feed