arrow_back Scaling the Largest Functional DataSet @Flipkart aka Catalog
Allocation and Forecasting in Guaranteed Delivery of Advertisements
Submitted by Aditya Ramana Rachakonda (@arrac) on Tuesday, 19 July 2016
Guaranteed delivery (GD) of advertisements helps brands book advertisement views of niche audience segments well in advance. To enable this, we need to create an intelligent system which allows for targeting of users, forecasting supply, optimally booking campaigns, allocating campaigns to users, pricing the guarantees and penalties correctly.
In this talk, we will discuss the following:
- Allocation: In GD, as advertisements don’t compete on bids, we allocate advertisements to user-views such that we deliver on the guarantees, while keeping the advertiser’s interests (reaching the right set of users) intact. This problem is modelled as constrained optimization over bipartite graph of advertisements and audience segments.
- Forecasting: We need to know the number of views in the future that we get from different audience-segments. We model views in an audience segment, as a time series and various external inputs as exogenous variables which affect the time series. We will briefly describe the various algorithms and processes that we follow to enable forecasting.
Aditya Rachakonda is a Data Scientist at Flipkart Ads. He is currently working on problems in advertisement optimization. Earlier, he was a Research Scientist at Big Data Labs in American Express. He has experience in machine learning, text mining and getting insights from short and noisy texts. Aditya is a PhD in Computer Science from IIIT Bangalore and his research interests include semantics in text and information retrieval.