Mix-camp E-commerce Search
21st June 2023
You will find the slides and videos from the talks in the Programme section below.
MICES brings together participants from a variety of backgrounds, all sharing a common interest in e-commerce search. In order to stimulate and facilitate discussion, we will start with scheduled talks in the morning. This will be followed by self-organising sessions: participants are encouraged to initiate discussions about their topics of interest or to give an ad hoc presentation.
The workshop welcomes all topics related to e-commerce search. We have compiled a list of topics that will probably feature at the workshop. This list is not comprehensive. Participants are welcome to discuss further topics at the workshop.
The constant drive for improved economic efficiency in retail and eCommerce has led many businesses to prioritize exploitation over exploration, resulting in echo chambers and biased data sources that hinder optimal outcomes. However, it's important to note that achieving a balance between exploration and exploitation is crucial for long-term success in eCommerce. Businesses must continuously adapt to changing market conditions and reinvigorate their customer experience while maintaining a steady revenue stream. Employing non-critical data sources, creativity, and exploration can help companies shift the focus back to the customer experience, which is the primary asset and key differentiator of the retail industry, rather than merely the transaction.
Andreas Wagner is a serial entrepreneur who is passionate about teaching software to comprehend human behavior. With more than 15 years of extensive experience in the retail and eCommerce industries, he has held various positions in different companies. Currently, he is serving as the CTO and strategist at searchHub.io, where he continues to pursue his mission of improving software's ability to understand human needs.
Tom Gilke, Tobias Kässmann
The majority of user journeys in search starts with a click in a search box. How can you guide your users after this initial click? At otto.de we strive to show our users the best suggestions for their need, right after the first click. We work towards providing a tailored fit for each individual user, helping them to reach their goal by responding to their input in seemingly no time. How do we find the best suggestions? What architecture allows us to handle every single keystroke of each user in just a few milliseconds? Learn, how your team can jumpstart the customer experience, by focusing on the user beginning with the first click.
Tom started his journey as software developer in the search team at otto.de. After gaining experience in the search field, he started to work on the autosuggest, which is his current focus. Besides work, he enjoys learning new things about computer science, sharing his learnings with others, and gardening.
Tobias is Organizer of the Search Technology Meetup Hamburg. He has been in love with search engines for more than 10 years. In the past he worked for several companies in the context of search and was responsible for the whole search-infrastructure. By indexing, storing and delivering the best search results he took a deep dive into Solr, Lucene, Elasticsearch. As part of the otto.de autosuggest team he is now heavily in touch with Vespa.ai and DuckDB to deliver the best ML-Driven autosuggests to the user.
Abel Camacho Guardian, Juan Lapadula Plá, Joel Widmer
At Digitec Galaxus, we strive to enhance the search experience for shoppers by connecting them with the most relevant products. Product filters have been a crucial tool in achieving this goal. Filters help users quickly narrow down search results and focus on products that meet their specific needs. To make filter interaction more convenient, we have recently introduced suggestions for popular filter options.
This presentation demonstrates how we leverage behavioral and product data along with simple models to create effective suggestions for filter options. We also discuss experimental analyses and methods we use to identify potential issues and estimate business value. In particular, we focus on how our iterative approach to experimentation aims to mitigate risks and provide continuously better features to customers. Finally, we share the challenges and learnings along our journey.
Joel Widmer and Juan Lapadula Plá are Software Engineers, and Abel Camacho Guardian is Senior Analytics Engineer at Digitec Galaxus.
Most modern search engines heavily rely on behavioral or transactional feedback. However, it can be challenging to provide a high-quality search experience with new items that have not yet been purchased and therefore have no interaction data available. As behavioral data is one of the strongest ranking signals in e-commerce, popular items are getting even more purchases, and new ones have no chance to be seen. This situation leads to a poor experience for both users and vendors. We will explore multiple strategies to address the item cold start problem in search and share insights into which solutions worked best for Delivery Hero.
Evgeniia Trufanova works as Data Scientist at Delivery Hero specializing in search ranking. Previously worked on various machine learning projects for startups and major banks.
Arnab Dutta, Gleb Polushin
eCommerce search employs a plethora of modules to enhance the user experiences. At eBay search our constant effort has always been to deliver products catering to the above goal. One such user focussed module is automated spelling correction. In this talk, we will share our approach for iteratively solving the problem using naive methodologies and also advanced state of the art modeling paradigms. In the process we would also highlight the various data creation strategies one can employ to train such models. Our work lies at the intersection of enabling users towards a better ecommerce search experience and pushing the boundaries of AI in augmenting query reformulations in search. In particular, we explore and perform comparative analysis of several pretrained large language models, fine tune them on our task specific data and report its offline performance metrics. In particular, we choose a particularly complex learning task of spelling correction and employ variants of transformer based pre-trained LLMs like T5 and BART and combine its multilingual capabilities. We outline in this session the current spelling correction models, the need for pre-trained models and how this can be advanced further using state of the art prefix tuning style learning.
Arnab is a staff data scientist in the Search Science team at eBay. He focuses primarily in leveraging ML techniques towards natural language understanding. He has also worked extensively on knowledge graphs, intent understanding, context extraction to improve the search engine that powers eBay’s marketplace. He has extensive enterprise experience in model building, NoSQL databases, information extraction, probabilistic models. He has publications in ESWC, ISWC, WWW and more. Arnab received his PhD in Artificial Intelligence from the University of Mannheim in 2015.
Gleb is a Data Scientist in the Search Science team at eBay. He has experience of building NLP solutions in e-commerce companies, and now he focuses on applying ML solutions to improve search quality at eBay. Having a background in software engineering and data-journalism, he is passionate about NLP, data analytics and data science.
Marlety García Valdés
Innovation comes through creativity, commitment, investigation, courage, patience and iteration. The world is currently facing problems that go beyond us as individuals: hunger, war, climate change, political and economical uncertainty, the pandemic,... With all this weighting on our shoulders, us humans now value safety and people more than ever. Businesses not only must be competent but also ethical, trustworthy, reflecting societal values, meeting new arising societal needs and changing social interactions for the better. It is not about “me” anymore, but about “us”. That’s why, at Empathy, our vision is moving towards a society based on respectful relationships that put humans and their digital rights at the centre, championing ownership and action on our responsibility towards a more ethical digital world. Inspired by this hunger for change, brands take actions beyond their product and business, helping them gain customers’ trust by improving their performance and having a stronger societal impact, lining towards a digital design that contributes to making the world a better place. What are some of the tools in our hands, as designers, to achieve that?
Marlety is a Product Designer and Design Manager at Empathy.co and passionate about about people, design and their synergy. She is hungry for new experiences and challenges, committed to driving projects from ideas to execution, inspiration-seeker and a firm believer that self-achievement comes through the greatness of those around you.
Aparna Sundar (OpenSearch Project), René Kriegler (OpenSource Connections) and Ángel Maldonado (EMPATHY.CO). Hosted by Sebastian Russ (Tudock).
Search is one of the most important features of e-commerce platforms to support users in the process of making a buying decision. On the other hand, we know only very little about how search interacts with this decision making process, and, in practice, search teams are often missing out on insights and opportunities that they could generate from UX research.
This is why we would like to give greater visibility to UX in E-commerce Search and discuss about it in this panel. We think that UX deserves even greater attention in a time that is full of new opportunities coming from new AI-based technologies.
We come to this discussion from different backgrounds, providing different perspectives on UX and hoping to inspire the audience to (re-)think about UX in e-commerce search in their day-to-day work: Aparna approaches e-commerce search after having worked and published in consumer behavior research. René's perspective on UX in e-commerce search is based on his work on optimizing search result quality. Finally, Ángel approaches the user experience from the perspective of ethical commerce, and openness and trust in the relationship with the user.
Aparna Sundar has a PhD in Consumer Behaviour research and is UX Research Lead at OpenSearch. Ángel Maldonado is CEO at Empathy.co and an advocat of ethical commerce. René Kriegler works as Director E-commerce at OpenSource Connections. Sebastian Russ is Product Manager at Tudock.
René works as Director E-commerce at OpenSource Connections. As a search consultant he has supported clients in Germany and abroad for 15 years. Although he is interested in all aspects of search, key areas include search relevance consulting, e-commerce search and Apache Solr/Lucene. René maintains the Querqy open source library. He co-founded MICES in 2017 together with Paul Bartusch and Isabell Drost-From.
Sebastian works as Product Manager at Tudock focussing on e-commerce search. His passion for search is reflected in a wide range of search projects with both closed and open source solutions. Driven by couriosity and the challenging nature of search topics he believes that collaboration between technology, business and design brings up the best results. MICES is a great place to make that happen and Sebastian is happy to be able contribute to the search community he learned so much from.