The ECIR 2013 Industry Day will be held on Wednesday, March 27, 2013, during the regular conference program, and in parallel with the technical tracks. The Industry Day's goal is to present the state of the art in search and search-related areas, delivered as keynote talks by influential technical leaders from the Industry.
Industry Day chair:
Lessons from the Wild: How Context Can Shape Consumption in Content Recommendation Systems
Cranfield style evaluations have the advantage of reproducibility but ignore many factors that can shape user interaction. It is common in industry to augment static test collections with tests performed on live user traffic. Split or A/B testing can be used to help control compounding factors, but nevertheless they can have a large impact on online metrics. How users interact with the product also provide a valuable source of data for understanding a product and forming hypotheses. For example, users following an email link messaged "Top content, tailored for you from the people, industries, and companies you're following" may have a different expectation about the content recommendations than when following a link from the homepage of LinkedIn titled "LinkedIn Today recommends this content for you." In this talk, we share lessons and observations about patterns of user interaction with the content recommendations we provide in LinkedIn Today. We pay special attention to how members arrive at the product, the different contexts in which the product may be consumed (website, mobile, email), and what they do next.
Research Scientist, , Associate Professor at University of Maryland
Search and Discovery at Twitter
Twitter aims to be an information platform that connects users to what they care about, 140 characters at a time. Whether it's breaking new events around the world, the latest celebrity gossip, or the recent adventures of your closest friends, the search and discovery services aim to surface relevant and personalized content in real-time. In this talk, I will provide an overview of projects at Twitter in this space, discuss challenges that we face, and share some experiences we gained along the way. In particular, I'll focus on the real-time aspects of the problem, which has implications for both the design of search architectures and content-ranking algorithms.
Principal Researcher, Microsoft Research
This talk discusses the interplay of “Social” and “Search”. Social signals can be used to improve the relevance of algorithmic search results, and information retrieval techniques can be used to surface content from social networks. I will give an overview of the ways in which social content is used to improve the Bing search experience, and at a more abstract level will discuss how signals from a variety of social networks can be used to improve the user’s experience in different task scenarios. I will also touch on some problematic aspects of leveraging social information, such as demographics of the user base and trustworthiness of postings.
Principal Developement Manager, Bing Europe
Suggestion Technologies for Bing
Search engines have long provided clues to the topics people look up. In this talk, we discuss Bing suggestion technologies (autosuggest) for the use of showing the precise questions that are most frequently asked. These technologies anticipate what you are most likely to ask based on questions that other people have made. Simply type a question starting with a word like “is” or “was,” and search engines will start filling in the rest. Frequently asked questions include: “When will the world end?” “Is Neil Armstrong Muslim?” “What is the meaning of the life”. During the talk, we will demonstrate why autosuggest reflects the collective curiosities of its users and why this can help improve the quality of search and user experience.
Researcher, Telefonica Research
Mobile Search: a Force to be Reckoned with!
Recently the world has witnessed a revolution in terms of mobile web and mobile search usage. Mobile phones, once deemed as simple communications devices, now provide mobile users with access to a wealth of online content, anytime and anywhere. In 2012, the increasing presence of mobile devices caused desktop search to decline for the first time ever; a level of growth that simply cannot be ignored. In this talk, I’ll take a nostalgic look back at the simple beginnings of mobile search and discuss how, why and in what ways mobile search has evolved over the past 8years. I’ll highlight patterns of mobile search usage and show how they not only differ from desktop search, but they are continually evolving. And instead of taking a single, data-centric viewpoint of mobile search, I’ll also discuss user-centric studies, highlighting the unique needs, intents and motivations of mobile searchers. Finally, I’ll share some of my thoughts about where mobile search is heading, the challenges that lie ahead and discuss some of the factors that I think are important when it comes to enriching the future search experiences of mobile users.
Head of Ranking Department, Mail.Ru
Active Learning to Rank
Development of a system based on supervised machine learning includes three main steps: factors selection, building training set and appropriate ML algorithm application. The training set construction is the very problematic aspect, since usually it is not well controlled but may dramatically affect the resulting quality of ML model. In my talk I am going to introduce our active learning technique to manipulate the training set in context of learning to rank problem. Using simple and effective algorithm we can significantly reduce the training set size as well as improve the ranking quality.
Head of Web Ranking Team,
Aggregate and Conquer: Finding the Way in the Diverse World of User Intents
In our days a search engine result page is not represented by just ten blue links. It usually contains additional special result items, containing news, image or video results. These verticals are intended to satisfy specific user intents, but those needs are actually very diverse and search engines are often not able to provide, for example, a special vertical search engine for some rare user need like "buy a 3-wheel bicycle". In this talk I am going to describe the approach adopted by Yandex. Instead of fixing slots for vertical documents we adopt a more flexible technique based on a probabilistic user model. It allows us to deal with redundant and complimentary verticals in the situations when not all verticals are equally relevant and important for the query. Using our framework we can easily create a new vertical ranking using the existing web document collection. This ranking can then be easily incorporated into the aggregated search results page to maximize the chance of satisfying the user.
Chief Scientist, bitly
Opportunities in Web Search
Search is not a solved problem! We still lack robust products, algorithms, and tools for non-query driven web search. This talk explores progress in search of social data and realtime data, and looks at practical methods for query large volumes of social data in a realtime system.