MICES EU 2022

Registration

Welcome & Introduction

René Kriegler / Paul M. Bartusch
Organisers

Slides

Growing Your Search Team

Mikio Braun
Freelancer

As search systems grow, there is often a wish to build out specialized teams so that one can create a stable and resilient platform but also work on many topics in parallel. Given the technical complexity of search systems, this is no easy challenge. In this talk I'll draw from the team topology framework and my personal experience scaling a search team at Zalando from 3 to a couple dozen people to explain what to consider and give recommendations for how to grow your search team.

Mikio has worked as an ML researcher, team lead and individual contributor, and as a consultant on all aspects to put ML in practice. He previously worked at Zalando and GetYourGuide in recommendation, search, ML platforms, and general AI strategy.

Slides
Video (Youtube)

Break

Understanding Distance: Moving from Local Search to Nation-Wide Search

Xavier Sanchez Loro
Wallapop

Relevance is much more than textual matching and ranking based on documents characteristics. In commercial search engines, it also must include business logic and user context, not only user intention. Maximizing user satisfaction and revenue is never an easy task. During the last 2 years, Wallapop changed its search engine paradigm from pure geolocated searches aimed to support local sales to nation-wide searches supporting transactional and shipping services provided by the company.

This has been a long and chaotic journey. We had to deal with defects at every level (base ranking, per category and even per keyword). For this we had to know better our domain. One key factor to succeed has been understanding which role distance plays in our search ecosystem. This affects relevance itself but also satellite components like autocomplete and suggestions.

In this talk we will explain our journey moving Wallapop’s search ecosystem along with our business, sharing the challenges and learnings obtained from all the initiatives, experimentation and changes required to support this shift

Former computer science researcher turned search engineer almost a decade ago. Working at Wallapop's search team for 6 years. When having some spare time, Xavier likes to produce electronic and experimental music.

Slides
Video (Youtube)

Rank Factors: moving beyond basic relevance to focus on specific business objectives

Nate Day
OpenSource Connections

Search relevance has received a lot of attention, but ranking factors are perhaps a more important part of successful e-commerce search. A customer’s buying decision considers far more than just how well a product matches a query; it includes other features of the product like: price, availability, delivery-time, popularity and ratings. Successfully incorporating these ranking factors into your shop’s search, requires processing these raw signals into normalized versions.

This talk walks through the practical data science to create basic ranking factors that you can leverage in your e-commerce search.

Nate is a search consultant at OpenSource Connections, where he specialized in Elasticsearch and data insights for clients in the ecommerce space. Before search he was doing biostatistics and since transitioning into search has continued to focus on quantifying signals in complex systems. His favorite projects involve either delivering new understanding through analysis or operationalizing existing research via the search layer implementation. Outside of search Nate is an avid golfer, river boater, and dog lover. He has been practicing his German for a month leading up to this conference!

Slides
Video (Youtube)

Break

Beating the Status Quo Ranking on Shorthead Queries @ OTTO

Felix Rolf
OTTO

At OTTO, we are on the verge of going live permanently with Learning to Rank models. On this journey, a lot of attention goes into finding the right features, model types and training targets, also known as Judgements. Finding Judgements that optimize KPIs such as click through rate or conversion rate is crucial for the successful training and application of LTR-models. We have run a series of A/B tests to explore how various Judgements affect our KPIs on shorthead queries.

In this talk, we'd like to share our key findings and point out which Judgements worked best for OTTO Search.

Felix is product manager at OTTO Search and part of the ranking team. Having a background in the media and e-commerce industry, he is passionate about user experience, behavioral analytics and data science solutions.

Slides
Video (Youtube)

Add a Personalization touch

Aline Paponaud & Vincent Bosc
Adelean

Two different persons typing the same words in a search engine often expect different results. How can we satisfy both? That’s where Personalized Search comes into play!

In this session, we’ll look at the common techniques used for personalization, how personalization is related to recommandations and which difficulties are often met with its implementation. We’ll finish with a real solution applied to a big retailer using Elasticsearch.

Aline is CTO of Adelean, working with search and providing consulting services and expertise around Elasticsearch, Lucene and Solr. She brings her energy to leveraging search engines, as they become more and more essential in every domain.

Vincent has worked at Adelean as a search engineer for a couple of years. He also works as a consultant for Elasticsearch.

Slides
Video (Youtube)

Lunch

Panel session: Search maturity growth with an open source approach

Hosted by Paul M. Bartusch
productful.io

Video (Youtube)

Establishing search for Do It Yourself retail

Jörg Schönfisch
Bauhaus

As a part of a large cloud transformation, we created a new search team by the mid of 2021 in order to improve the customer experience by a customer-centric search engine. As a first step, we strived for three strategic goals: (1) establishing search management from a technical and organizational perspective, (2) improving customer insights by substantial KPIs and relevance metrics, as well as (3) enabling relevance engineering by supportive infrastructure and frameworks.

To achieve these goals efficiently, we established several tools from the Chorus tool-stack like SMUI, Querqy and Splainer. They are installed mainly on the basis of serverless technologies in order to reduce their maintenance effort to a minimum. Querqy is used as a query rewriting framework that facilitates the implementation of search management business rules, and additionally helps to ship the output of machine learning algorithms into the search.

The talk will discuss how Chorus tools can help to establish search from an organizational perspective, how frameworks on top of serverless technologies can accelerate relevance engineering and the specific characteristics of search behavior in the context of Do It Yourself retail.

Jörg is Tech Lead and Search Engineer at BAUHAUS, one of the largest DIY retailers in Europe. He is passionate about all things knowledge and information, especially in the space of semantic technologies. Having extensive experience in data and software engineering - working in companies big and small - he has set out to build a better search for DIY customers at BAUHAUS.

Building better onsite search: a team maturity perspective

Shaun Brazendale
Rubix Group

The quality and variety of solutions available for onsite search for e-commerce has increased significantly over the last few years: semantic search, personalisation, image-based search, retrieving results at speed from more than 100 million products - all this seems available to us, packaged as technological solutions that we would just need to integrate.

However, we usually have an existing search solution in place, maintained by a team that is responsible for delivering a good search experience and to increase sales via search.

The search team is at a certain maturity level - ranging from beginners and generalists to experts in information retrieval and search-related product development. In the light of the many technological opportunities, and given a company’s actual status quo, how can we find a good and appropriate strategy for improving onsite search to increase sales and other related business objectives?

In this talk, we will discuss the case of onsite search at Rubix Group - a multi-brand provider of industrial supplies in 22 countries. We shall lay out Rubix’ journey from an initial assessment using OpenSource Connection’s Search Maturity Model to implementing new onsite search merchandising capabilities based on the Chorus stack of open source search tools. Combined with improved capabilities to measure search quality and with adjustments to the search merchandising processes within the organisation, this process resulted in a better search experience for both the customer and the seller.

We think that this path could inspire other e-commerce companies to approach how to improve their onsite search from a maturity progression perspective.

As the Product Owner for on-site search at Rubix, Shaun heads up a search team responsible for relevancy and customer experience optimisation across 18 B2B ecommerce stores in Europe.

Shaun has worked in B2B ecommerce for 7 years, including projects in the UK, France and Turkey in industries such as energy, horticulture and industrial supply. Starting with a marketing background in B2B and finally progressing to customer experience and search Product Owner role at Rubix shaun has developed a core team of search engineers that curate a mix of proprietary and opensource applications. Shaun oversaw the implementation of the Chrous stack into Rubix’s search stack which has enabled teams across Europe to actively searchandize their stores like never before.

Shaun holds a BA (hons) in Business and Marketing from the University of Liverpool

Break

Similar Products Model and its applications

Aigul Gazimova
Joom

A Similar Products Model for internal search is crucial for e-commerce, especially for marketplaces helping the customers to find the best candidate for the least money while observing all sorts of goods.

Joom as an international group of e-commerce and fintech companies working with both B2B and B2C segments, with one core developers team building the internal search engine.

We will talk about:

• building the Similar Products Model for Joom Marketplace (B2C-product)
• implementing the Similar Products Model to JoomPro (B2B-product) where we lack user activity history
• how such a model optimized the recommendation system.

Aigul holds a Bachelor’s degree in applied mathematics and computer science. As a part of the search team at Joom, she improves internal search by developing machine learning algorithms. Previously, she was developing cache eviction algorithms for databases as an intern in a major cloud company. Math is the love of her life but she will do her best to share her best ideas not only with the developers but also with managers who do the most important work – making the users happy.

Slides
Video (Youtube)

Barcamp / self-organising sessions