These are heady times for the German IR community. Norbert Fuhr’s recent winning of the Salton Prize not only rewarded a fantastic and long-term individual contribution to the field, but has also served to shed light on the whole IR scene in Germany – a fact underlined by this “Made in Germany” series.
There are several highly active IR research groups in Germany. These focus on diverse aspects of Information Retrieval, including linguistic and systems aspects, user modelling, human behaviour in IR and user interfaces. In this article I will describe some of the IR related research activities at the University of Regensburg, which incorporate many of these aspects.
As Ingo Frommholz explained in his opening article in this series, Regensburg played an important role in the development of the IR community in Germany. Recent appointments here have led to somewhat of an IR resurgence at our institution. Two research chairs at the university work on IR related topics. The Chair for Information Science led by Prof. Rainer Hammwöhner has several staff members with an IR related focus. The neighbouring Chair for Media Informatics, led by Prof. Christian Wolff has an HCI focus, but their research is also of interest to the IR community.
In this short article I will outline some of the research that has been recently performed and is on going at our University. The aim is to provide the reader with a flavour of what our strengths and interests are and, in the context of this series, how our work fits into the wider German IR scene. I will first describe three research projects at the Chair for Information Science and follow this by summarising some of the work from the Chair for Media Informatics.
Searching for Fun
Searching4Fun is one area of interest in Regensburg that has received particular attention in recent times, both within the research community and in the media. Much of the literature in Information Retrieval and Information Seeking assumes that search is performed in response to a task or some kind of work-based situation. Results from our studies suggest that this is often not the case. On the contrary, in many leisure situations, search is motivated by hedonistic reasons, such as to fight boredom, to come down from a hard day at work, to distract oneself from stressful or emotionally difficult situations etc.. This means that the way people behave to address these needs and what information they are happy with is different than in work-based situations. For IR researchers these differences (outlined in more detail in a book chapter co-authored with Max L. Wilson [Elsweiler et al., 2011]) lead to many open questions e.g. What do leisure queries look like? How should systems respond? How can systems be evaluated? Are our relevance-based metrics suitable or do we have to look for other metrics? Is it possible to simulate such needs and behaviour in controlled, repeatable experimental setups?
Currently our group, including doctoral students Hanna Knäusl (Chair for Information Science) and Richard Schaller (University of Erlangen) are looking to find answers to these kinds of questions by studying users in diverse leisure contexts including in the contexts of searching Wikipedia [Knäusl, 2012]; distributed leisure events [Schaller, 2012] and on Twitter.
In particular we are investigating what drives search in these situations, how people behave respond to such needs, what aspects (in terms of tool support and supplied results) lead to positive outcomes from the perspective of the user. The main goal of this work is to understand how best to build information systems to support users in leisure situations.
The Intelligent Browser
Led by Prof. Bernd Ludwig researchers in our group have been trying to use interaction signals in search scenarios to infer user contexts and needs. The idea is not only to use these to determine what information should be provided to the user, as is typical in IR implicit feedback research, rather to determine how the provided information should be presented. For instance, one could imagine that the Wikipedia article for a complex topic, such as quantum mechanics, might be displayed differently depending on the user’s level of knowledge of the topic. More elementary or advanced aspects may be left in or filtered out, and different language or terminology used to suit the user’s needs. Similarly, in a leisure-scenario, depending on a user’s preferences, it may be appropriate to show more multimedia content, which takes more time to process, but may be more pleasurable, than in a situation where the user’s goal is to quickly look up a fact or statistic.
The premise of our work is that these kinds of contexts and preferences can be inferred from the way a user interactions with the system, such as the queries submitted, the pages traversed and the way in which pages are consumed etc.
We have been running experiments in the context of Wikipedia search and Regensburg tourism and have utilised several modelling strategies to represent user interaction [Knäusl and Ludwig, 2013].
Current research is focusing on how these models can be used to make predictions regarding the context of the search (e.g. is the user an expert or novice?) and future user behaviour (e.g. can we guess what the user will do later in the search process?).
A further research focus in Regensburg is recommender systems. In particular we are interested in RecSys for lifestyle change. Poor nutritional habits are a major cause of ill health in today’s society and the evidence suggests that although many people want to make positive changes to their diet, many lack the nutritional understanding to achieve this. We (and our collaborator Morgan Harvey from the University of Lugano) believe that recommender systems can play a key supportive role here. If systems can provide recommendations that are both tasty (i.e. the user would want to eat,) and healthy (i.e. can be combined to form a balance diet with respect to the user’s lifestyle), this may go some way to supporting the user make desired behavioural changes. If recommendations were to be explained, it may even help increase the user’s level of nutritional knowledge.
Up until now our work has focused on understanding user preferences with respect to eating habits and how these and contextual factors influence the rating of recommendations of individual meals.
We collected a large set of food ratings in context and asked users to provide reasons for the ratings given [Harvey et al.,2012]. The data collected provided a basis to investigate key factors contributing to how recipes are rated and offered insight into how to best approach the recommendation problem. Based on analyses of the data, we have devised several recipe recommendation models that are able to leverage understanding of user’s likes and dislikes in terms of ingredients and combinations of ingredients as well as preferences with respect to the nutritional content of recipes. The performed experiments demonstrate that these models are able to outperform strong content-based and collaborative filtering baselines when predicting user ratings for recipes.
We have two demo systems in the context of this work. The first estimates the nutritional content of recipes by matching ingredients [Müller et al.,2012]. The second is a food portal Quizine.me, which allows users to add and browse recipes and receive personalised recommendations.
Our future aims are to work with nutritional experts to develop algorithms for combining recommendations into balanced meal plans based on the user’s sex, size, lifestyle and goals.
Research at the Chair for Media Informatics
The IR related research at the neighbouring Chair of Media Informatics mainly focuses on interfaces for and interactive aspects of Information Retrieval. Specific topics of interest include Social Networking, Tagging Systems and Personal Information Management.
One continuing project is attempting to build a digital bookshelf for an individual’s digital objects [Bazo, 2013]. The idea is to completely realise the well-known “desktop metaphor” by allowing the user interact with digital objects in a way close to how the user behaves in the physical world. Digital objects (e.g. files from dropbox and articles from Mendeley) are projected as items on a bookshelf onto the wall. The user can inspect and interact with items by manipulating the projected image and organise these as he or she pleases. The aim of the project is to understand how interacting with digital objects in a physical way influences user habits and plans are underway to extend the existing prototype to work with other interactive media e.g. touch surfaces.
Other PIM research at the Chair of Media Informatics has related to understanding how people use social media and tagging systems to share, manage and re-find their own resources or use such systems as search engines for items added by the whole user population. Studies performed have demonstrated that social media is used to support a wide range of behaviours and have seeded conceptual behavioual models [Heckner, 2009]. The models reflect that sharing and PIM are important motivational sources for social media use.
A current external doctoral project at the Chair for Media Informatics is concentrating on categorising items in personal streams of social networking content, e.g. from twitter and facebook. The idea is to combine an event based approach with text mining and information filtering in order to classify incoming information based on features of the stream. These features are fed into machine learning methods in order to classify the text streams. The experiments performed have shown that the separative capabilities of event based features can improve performance in real time search engines [Bauer, 2012].
This short article has summarised some of the diverse IR research interests in two research chairs at the University of Regensburg. Reflecting our activity in the Information Retrieval space, we are delighted to welcome the fifth Information Interaction in Context symposium (IIiX 2014) to the beautiful German city of Regensburg.
IIiX is a unique forum exploring the relationships between and within the contexts that affect information retrieval (IR) and information seeking, how these contexts impact information behavior, and how knowledge of information contexts and behaviors improves the design of interactive information systems.
IIiX aims to foster an integrated approach to information access by bringing together members of the diverse research communities in information seeking behavior (Behaviour Track), user interface design for IR systems (Interface Track), and IR system design (System Track). Thus, the remit of the conference perfectly matches our research interests at the University of Regensburg.
In 2014 we are putting particular emphasis on “building bridges” between the various research communities with specific activities being planned to encourage improved understanding of differing viewpoints and foster collaboration.
So, if you are performing IR research or research relating to interfaces, behaviour or systems for IR, please consider submitting. I hope to see you in Regensburg!
Bauer, A., & Wolff, C. (2012). Event based classification of Web 2.0 text streams. arXiv / ACM Computing Research Repository (CoRR). Retrieved from arXiv.org website: http://arxiv.org/pdf/1204.3362v1.pdf
Bazo, A.; Burghardt, M. & Wolff, C. (2013), Virtual Bookshelf, in ‘In Proc. International Symposium for Information Science’.
Elsweiler, D.; Wilson, M. L. & Kirkegaard Lunn, B.Spink, A. & Heinström, J., ed., (2011), New Directions in Information Behaviour, Emerald Publishing, chapter Understanding Casual-leisure Information Behaviour.
Harvey, M.; Ludwig, B. & Elsweiler, D. (2012), Learning user tastes: a first step to generating healthy meal plans?, in ‘First International Workshop on Recommendation Technologies for Lifestyle Change (LIFESTYLE 2012)’, pp. 18.
Heckner, M.; Heilemann, M. & Wolff, C. (2009), Personal information management vs. resource sharing: Towards a model of information behaviour in social tagging systems, in ‘Int’l AAAI conference on weblogs and social media (ICWSM)’.
Knäusl, H. (2012), ‘Searching Wikipedia: learning the why, the how, and the role played by emotion’, proceedings of EuroHCIR@ECIR2012.
Knäusl, H. & Ludwig, B. (2013), What Readers want to Experience – An Approach to Quantify Conversational Maxims with Preferences for Reading Behaviour, in ‘Proc. 5th International Conference on Agents and Artificial Intelligence (ICAART).
Müller, M.; Harvey, M.; Elsweiler, D. & Mika, S. (2012), Ingredient Matching to Determine the Nutritional Properties of Internet-Sourced Recipes, in ‘Proc. 6th International Conference on Pervasive Computing Technologies for Healthcare’.
Schaller, R.; Harvey, M. & Elsweiler, D. (2012), Entertainment on the go: finding things to do and see while visiting distributed events, in ‘Proceedings of the 4th Information Interaction in Context Symposium’, ACM, New York, NY, USA, pp. 90–99.