CrEOS: Identifying Critical Events in Online Sessions

Abstract

Analyzing a consumer's experience in an online session is a fundamental task of a marketer to enrich, personalize and enhance their future interactions with a firm. Click (web) or tap (mobile) event streams efficiently capture a consumer's experience in an online session. In this work, we propose CrEOS contributing to the extant literature in the deployment of deep learning techniques to model online consumer behaviour. CrEOS models consumer conversion behaviour using a multi-variable LSTM, enriching the consumer event stream with a series of contextual features. In addition to the predictions, CrEOS identifies events that either hamper or help a conversion at three levels - consumer, consumer segment and transitions in marketing funnel. We validate CrEOS on click-stream data of a large US based e-commerce firm, compare it with single variable LSTM and discuss the insights derived from the critical event identification considering a series of marketing tasks.