Evgeny Perevodchikov

About Evgeny Perevodchikov

Strong quantitative and analytical skills with more than seven years of experience in…

Experience

1. Applied Scientist

   Amazon

   Sep 2013 — Present

   NLP, CV, IR - Big data modeling, machine learning, attribution algorithms, measurement, targeting - As a scientist for the Amazon ad platform\'s targeting, i have developed a modeling infrastructure to build an effective audience segmentation strategy for our clients. A variety of predictive modeling techniques are used to capture patterns in customer behavior, and a number of audience lookalike models (model based segments) are developed for our advertisers to serve their various advertising strategies (sales, new customer acquisition, customer retention, etc.). I have built the pipeline in Scala on Spark- In addition, I have led the effective effort to scale up the model building process and exit beta for pixel based audience lookalikes, monitored online model performance and released the product for general use- I have develop a list of scientific requirements for the cross channel attribution algorithm and estimated the cross platform impact of cross channel attribution on ad-attributable units and sales. The requirements have been used to implement the new cross channel attribution algorithm. I have identified single channel attribution design flaws to be fixed in the cross channel attribution methodology and implemented statistical tests to evaluate correct implementation of the probabilistic attribution component for the new cross channel attribution logic- I have conducted a large scale analysis of incremental return on investment for a cross section of campaigns. The analysis increased the understanding of campaign performance and its various factors, which affect incremental sales. One of the findings is that the campaign performance can be improved by appropriately selecting products to be included in a campaign. The analysis has also laid the groundwork for building a tool to evaluate the future campaign performance before its launch and making recommendation how its performance can be optimized.