Non-parametric estimation of Multi-channel attribution | July 2014 - Present
The problem of interpreting the influence of various marketing channels to the user’s decision process is called marketing attribution. All of the present algorithmic models to estimate the attribution of marketing channels are parametric. In this work, we propose a novel non-parametric and a semi-parametric approach to solve the attribution problem. We build a simulation engine that captures important behavioral phenomenon known to affect a customer’s purchase decision; and compare the performance of five attribution approaches in their ability to closely approximate the known ground truth . Our proposed method works well when marketing channels have high levels of synergy. The work was accepted as a regular paper in WISE 2015. My first first authored paper. Yaaaaaay !
Evaluation framework for Multi-channel attribution | July 2014 - Present
All the algorithmic multi-channel attribution models aim to formalize an answer to the credit assignment problem, but a natural question to ask is which of these is better, or which of these gives a more correct answer. No academic literature has thus far explored this question. In this work, we extend the simulation engine proposed in the above work to evaluate and compare multiple attribution models. The simulation engine incorporates multiple traits customers and marketer exhibit in the real world. We present two metrics to calculate the accuracy of the attribution models. For comparison, we calculate the true attributions from the simulation engine and contrast it with the attributions of various models. This is one of my recent works and we are currently working on a research paper. Wish us all the best !
Behaviour of a customer in the next online session : A conversion funnel approach. | July 2014 - Present
A marketer would like to predict the behavior of a customer in the next online session. Such predictions would help the marketer understand the expectations of the customer and stratergise future targeting activities. In this work, while making predictions, we capture the customer’s stage in the conversion funnel. The stages were modeled as latent states of a non-homogeneous Hidden Markov Model. We provide a framework to predict the average number of page views and conversions of a customer in the next online session based on the referrer of the landing page. The proposed method is applied on two real world datasets. We observe significant improvements over the current prediction models by incorporating the stage of a customer in the conversion funnel. This is also one of my recent works and we are currently working on a research paper. The aim is to submit it to WWW 2015. Fingers crossed.
Visitor Classification Using Clickstream | May - June 2013.
Identifying and targeting visitors on e-commerce website with personalized content in real-time is important to marketers. In this work, we show that dynamic visitor attributes extracted from their click-stream provide much better predictive capabilities of visitor intent to make a purchase on a e-commerce website. We propose a mechanism for identifying similar visitor sessions on a website based on their click-streams. Novel techniques for extracting features from visitor clicks are employed. Large margin nearest neighbor (LMNN) algorithm is used to learn a similarity metric between any two sessions. Further the sessions are classified into purchasers and non-purchasers using k-nearest neighbor (kNN) classification. Experimental results showing significant improvements over baseline algorithms based on Hidden Markov Model(HMM), support vector machine (SVM) and random forest are presented on two large real-world data sets. This work was done as a part of my summer internship at Adobe. I was one of the primary contributors to the project. The work has been accepted at WISE 2014. My first research publication. Yaay !