As search solutions grow, so does the need for a more systematic approach to offline evaluation and optimization.
From only a few parameters to thousands or millions of combinations: What is the effect of a change? Does the tuning of search system parameters affect result quality? What is the optimal configuration?
In this talk Andreas Wagenmann will present Kolibri (https://github.com/awagen/kolibri), an open-source execution framework and UI, with focus on e-commerce search offline evaluations. Use it to explore and understand sorting, shape, the “what” and the “where” with minimal or no code.
Kolibri is source- and use-case agnostic, and works with any target system that returns a JSON response. It provides simple UI-supported no-code definition of requests, granularity of results and evaluation. This includes standard IR metrics and custom measures: customization is key. Exploration is encouraged through grouping and weighted aggregates on subsets.
Hewill discuss Kolibri’s design, practical use cases and outlook. The flexibility of the job definitions makes Kolibri a candidate for a general, scalable search toolbox, with potential use cases such as indexing, load testing and simulation.
Andreas Wagenmann is an Data Engineering Manager & Data Science Tech Lead at Empathy.co.
Get Notified about Upcoming Events