Paris Search & Data Meetup – Modelling Implicit User Feedback for E-commerce Search

OSC’s Director of E-commerce René Kriegler, will be speaking (in person!) at the Paris Search & Data Meetup on implicit user feedback and how to use this to tune e-commerce search. The event also features talks by Lucian Precup, CTO of a new collaborative search engine and Jérémy Mésière from industrial supplies and office equipment supplier Manutan.

An approach to modelling implicit user feedback for optimizing e-commerce search

More than other domains, e-commerce search depends on implicit user feedback to optimise search result ranking. While search relevance is probably still a major criterion of search result quality in e-commerce, the user finally takes the buying decision based on criteria such as ‘an attractive price’ and ‘brand sympathy’ that are very hard to make explicit. On the other hand, this decision making can be observed implicitly in web tracking.

Unfortunately, e-commerce search cannot just use more generally known approaches to click modelling. Many of these models assume that the user would view the search results sequentially, top to bottom, while in e-commerce search we often use a grid layout which invites the user to go back and forth between results, especially given the heavy use of product imagery on the search result page. In addition, our model needs to consider contexts beyond the well-known position bias, such as the device type (mobile vs desktop) or the time of the day, together with the different types of signals (clicks, add-to-basket, checkout). Last but not least, many general click models ignore the fact that the number of observations can vary a lot between query-product pairs. Our greater uncertainty in the case of sparse events should be reflected in the model.

In this talk, René will introduce an approach to using implicit user feedback that is based on Bayesian hierarchical modelling. It will provide a solution for dealing with position bias, including for grid layouts, and for dealing with further contexts, such as device types. The model will cope with varying quantities of observations and it allows to incorporate different types of events, such as clicks and checkouts.

Get Notified about Upcoming Events

  • This field is for validation purposes and should be left unchanged.