Analysis of eCommerce clickstream data for intent prediction and recommendations

Coveo is sharing a rich anonymized dataset on eCommerce behavior for research purposes. The dataset includes more than 30M single product interactions by real shoppers – the clickstream data is enriched by (vectorized) catalog meta-data and by fine-grained search behavior, showing not only products clicked after a query is issued, but also products seen and not clicked (i.e. negative feedback). 

In continuity with previous research (from Coveo and the community) and with the 2021 SIGIR Data Challenge, we welcome research addressing two challenges :

1. a session-based recommendation task, where a model is asked to predict the next interactions between shoppers and products, based on the previous product interactions and search queries within a session;
2. a cart-abandonment task, where, given a session containing an add-to-cart event for a product X, a model is asked to predict whether the shopper will buy X or not in that session.

Participants are encouraged to read the following peer-reviewed articles : (https://arxiv.org/abs/1907.00400 and https://rdcu.be/b8oqN) to get a better understanding of the underlying theoretical and practical issues: class imbalance, conversion rate, etc. The SIGIR Data Challenge paper contains extensive literature review, in-depth details about the dataset, and discussions about the target use cases; the SIGIR Leaderboard can be consulted for quantitative benchmarks (test files to replicate the submission phase can be requested to Coveo); the SIGIR open source repository contains detailed explanation on the dataset, ready-to-use baseline models and links to successful implementations.