Know Center Research

Publikationen

Mutlu Belgin, Veas Eduardo Enrique, Trattner Christoph
2017
In today's digital age with an increasing number of websites, social/learning platforms, and different computer-mediated communication systems, finding valuable information is a challenging and tedious task, regardless from which discipline a person is. However, visualizations have shown to be effective in dealing with huge datasets: because they are grounded on visual cognition, people understand them and can naturally perform visual operations such as clustering, filtering and comparing quantities. But, creating appropriate visual representations of data is also challenging: it requires domain knowledge, understanding of the data, and knowledge about task and user preferences. To tackle this issue, we have developed a recommender system that generates visualizations based on (i) a set of visual cognition rules/guidelines, and (ii) filters a subset considering user preferences. A user places interests on several aspects of a visualization, the task or problem it helps to solve, the operations it permits, or the features of the dataset it represents. This paper concentrates on characterizing user preferences, in particular: i) the sources of information used to describe the visualizations, the content descriptors respectively, and ii) the methods to produce the most suitable recommendations thereby. We consider three sources corresponding to different aspects of interest: a title that describes the chart, a question that can be answered with the chart (and the answer), and a collection of tags describing features of the chart. We investigate user-provided input based on these sources collected with a crowd-sourced study. Firstly, information-theoretic measures are applied to each source to determine the efficiency of the input in describing user preferences and visualization contents (user and item models). Secondly, the practicability of each input is evaluated with content-based recommender system. The overall methodology and results contribute methods for design and analysis of visual recommender systems. The findings in this paper highlight the inputs which can (i) effectively encode the content of the visualizations and user's visual preferences/interest, and (ii) are more valuable for recommending personalized visualizations.
Trattner Christoph, Elsweiler David
2017
Food recommenders have the potential to positively inuence theeating habits of users. To achieve this, however, we need to understandhow healthy recommendations are and the factors whichinuence this. Focusing on two approaches from the literature(single item and daily meal plan recommendation) and utilizing alarge Internet sourced dataset from Allrecipes.com, we show howalgorithmic solutions relate to the healthiness of the underlyingrecipe collection. First, we analyze the healthiness of Allrecipes.comrecipes using nutritional standards from the World Health Organisationand the United Kingdom Food Standards Agency. Second,we investigate user interaction patterns and how these relate to thehealthiness of recipes. Third, we experiment with both recommendationapproaches. Our results indicate that overall the recipes inthe collection are quite unhealthy, but this varies across categorieson the website. Users in general tend to interact most often with theleast healthy recipes. Recommender algorithms tend to score popularitems highly and thus on average promote unhealthy items. Thiscan be tempered, however, with simple post-ltering approaches,which we show by experiment are better suited to some algorithmsthan others. Similarly, we show that the generation of meal planscan dramatically increase the number of healthy options open tousers. One of the main ndings is, nevertheless, that the utilityof both approaches is strongly restricted by the recipe collection.Based on our ndings we draw conclusions how researchers shouldattempt to make food recommendation systems promote healthynutrition.
Trattner Christoph, Kuśmierczyk Tomasz, Rokicki Markus, Herder Eelco
2016
Historically, there have always been differences in how men andwomen cook or eat. The reasons for this gender divide have mostlygone in Western culture, but still there is qualitative and anecdotalevidence that men prefer heftier food, that women take care of everydaycooking, and that men cook to impress. In this paper, weshow that these differences can also quantitatively be observed in alarge dataset of almost 200 thousand members of an online recipecommunity. Further, we show that, using a set of 88 features, thegender of the cooks can be predicted with fairly good accuracy of75%, with preference for particular dishes, the use of spices andthe use of kitchen utensils being the strongest predictors. Finally,we show the positive impact of our results on online food reciperecommender systems that take gender information into account.
Trattner Christoph, Elsweiler David, Howard Simon
2016
One government response to increasing incidence of lifestyle related illnesses, such as obesity, has been to encourage people to cook for themselves. The healthiness of home cooking will, nevertheless, depend on what people cook and how they cook it. In this article one common source of cooking inspiration - Internet-sourced recipes - is investigated in depth. The energy and macronutrient content of 5237 main meal recipes from the food website Allrecipes.com are compared with those of 100 main meal recipes from five bestselling cookery books from popular celebrity chefs and 100 ready meals from the three leading UK supermarkets. The comparison is made using nutritional guidelines published by the World Health Organisation and the UK Food Standards Agency. The main conclusions drawn from our analyses are that Internet recipes sourced from Allrecipes.com are less healthy than TV-chef recipes and ready meals from leading UK supermarkets. Only 6 out of 5237 Internet recipes fully complied with the WHO recommendations. Internet recipes were more likely to meet the WHO guidelines for protein than both other classes of meal (10.88% v 7% (TV), p<0.01; 10.86% v 9% (ready), p<0.01). However, the Internet recipes were less likely to meet the criteria for fat (14.28% v 24% (TV) v 37% (ready); p<0.01), saturated fat (25.05% v 33% (TV) v 34% (ready); p<0.01) and fibre (compared to ready meals 16.50% v 56%; p<0.01). More Internet recipes met the criteria for sodium density than ready meals (19.63% v 4%; p<0.01), but fewer than the TV-chef meals (19.32% v 36%; p<0.01). For sugar, no differences between Internet recipes and TV-chef recipes were observed (81.1% v 81% (TV); p=0.86), although Internet recipes were less likely to meet the sugar criteria than ready meals (81.1% v 83 % (ready); p<0.01). Repeating the analyses for each year of available data shows that the results are very stable over time.
Eberhard Lukas, Trattner Christoph
2016
Social information such as stated interests or geographic check-insin social networks has shown to be useful in many recommendertasks recently. Although many successful examples exist, not muchattention has been put on exploring the extent to which social im-pact is useful for the task of recommending sellers to buyers in vir-tual marketplaces. To contribute to this sparse field of research wecollected data of a marketplace and a social network in the virtualworld of Second Life and introduced several social features andsimilarity metrics that we used as input for a user-basedk-nearestneighbor collaborative filtering method. As our results reveal, mostof the types of social information and features which we used areuseful to tackle the problem we defined. Social information suchas joined groups or stated interests are more useful, while otherssuch as places users have been checking in, do not help much forrecommending sellers to buyers. Furthermore, we find that some ofthe features significantly vary in their predictive power over time,while others show more stable behaviors. This research is rele-vant for researchers interested in recommender systems and onlinemarketplace research as well as for engineers interested in featureengineering.
Trattner Christoph, Oberegger Alexander, Eberhard Lukas, Parra Denis, Marinho Leandro
2016
POI (point of interest) recommender systems for location-based social network services, such as Foursquare or Yelp,have gained tremendous popularity in the past few years.Much work has been dedicated into improving recommenda-tion services in such systems by integrating different featuresthat are assumed to have an impact on people’s preferencesfor POIs, such as time and geolocation. Yet, little atten-tion has been paid to the impact of weather on the users’final decision to visit a recommended POI. In this paper wecontribute to this area of research by presenting the firstresults of a study that aims to predict the POIs that userswill visit based on weather data. To this end, we extend thestate-of-the-art Rank-GeoFM POI recommender algorithmwith additional weather-related features, such as tempera-ture, cloud cover, humidity and precipitation intensity. Weshow that using weather data not only significantly increasesthe recommendation accuracy in comparison to the origi-nal algorithm, but also outperforms its time-based variant.Furthermore, we present the magnitude of impact of eachfeature on the recommendation quality, showing the need tostudy the weather context in more detail in the light of POIrecommendation systems.
Kusmierczyk Tomasz, Trattner Christoph, Nørvåg Kjetil
2016
Studying online food patterns has recently become an active fieldof research. While there are a growing body of studies that investi-gate how online food in consumed, little effort has been devoted yetto understand how online food recipes are being created. To con-tribute to this lack of knowledge in the area, we present in this paperthe results of a large-scale study that aims at understanding howhistorical, social and temporal factors impact on the online foodcreation process. Several experiments reveal the extent to whichvarious factors are useful in predicting future recipe production.
Dennerlein Sebastian, Rella Matthias, Tomberg Vladimir, Theiler Dieter, Treasure-Jones Tamsin, Kerr Micky, Ley Tobias, Al-Smadi Mohammad, Trattner Christoph
2015
Sensemaking at the workplace and in educational contexts has beenextensively studied for decades. Interestingly, making sense out of the own wealthof learning experiences at the workplace has been widely ignored. To tackle thisissue, we have implemented a novel sensemaking interface for healthcare professionalsto support learning at the workplace. The proposed prototype supportsremembering of informal experiences from episodic memory followed by sensemakingin semantic memory. Results from an initial study conducted as part ofan iterative co-design process reveal the prototype is being perceived as usefuland supportive for informal sensemaking by study participants from the healthcaredomain. Furthermore, we find first evidence that re-evaluation of collectedinformation is a potentially necessary process that needs further exploration tofully understand and support sensemaking of informal learning experiences.
Trattner Christoph, Parra Denis, Brusilovsky Peter, Marinho Leandro
2015
The use of contexts –side information associated to information tasks– has been one ofthe most important dimensions for the improvement of Information Retrieval tasks, helpingto clarify the information needs of the users which usually start from a few keywords in atext box. Particularly, the social context has been leveraged in search and personalizationsince the inception of the Social Web, but even today we find new scenarios of informationfiltering, search, recommendation and personalization where the use of social signals canproduce a steep improvement. In addition, the action of searching has become a social processon the Web, making traditional assumptions of relevance obsolete and requiring newparadigms for matching the most useful resources that solve information needs. This escenariohas motivated us for organizing the Social Personalization and Search (SPS) workshop,a forum aimed at sharing and discussing research that leverage social data for improvingclassic personalization models for information access and to revisiting search from individualphenomena to a collaborative process.
Cook John, Ley Tobias, Maier Ronald, Mor Yishay, Santos Patricia, Lex Elisabeth, Dennerlein Sebastian, Trattner Christoph, Holley Debbie
2015
In this paper we define the notion of the Hybrid Social Learning Network. We propose mechanisms for interlinking and enhancing both the practice of professional learning and theories on informal learning. Our approach shows how we employ empirical and design work and a participatory pattern workshop to move from (kernel) theories via Design Principles and prototypes to social machines articulating the notion of a HSLN. We illustrate this approach with the example of Help Seeking for healthcare professionals.
Trattner Christoph, Balby Marinho Leandro, Parra Denis
2015
Large scale virtual worlds such as massive multiplayer online gamesor 3D worlds gained tremendous popularity over the past few years.With the large and ever increasing amount of content available, virtualworld users face the information overload problem. To tacklethis issue, game-designers usually deploy recommendation serviceswith the aim of making the virtual world a more joyful environmentto be connected at. In this context, we present in this paper the resultsof a project that aims at understanding the mobility patternsof virtual world users in order to derive place recommenders forhelping them to explore content more efficiently. Our study focuson the virtual world SecondLife, one of the largest and mostprominent in recent years. Since SecondLife is comparable to realworldLocation-based Social Networks (LBSNs), i.e., users canboth check-in and share visited virtual places, a natural approach isto assume that place recommenders that are known to work well onreal-world LBSNs will also work well on SecondLife. We have putthis assumption to the test and found out that (i) while collaborativefiltering algorithms have compatible performances in both environments,(ii) existing place recommenders based on geographicmetadata are not useful in SecondLife.
Larrain Santiago, Parra Denis, Graells-Garrido Eduardo, Nørvåg Kjetil, Trattner Christoph
2015
In this paper, we present work-in-progress of a recently startedproject that aims at studying the effect of time in recommendersystems in the context of social tagging. Despite the existence ofprevious work in this area, no research has yet made an extensiveevaluation and comparison of time-aware recommendation methods.With this motivation, this paper presents results of a studywhere we focused on understanding (i) “when” to use the temporalinformation into traditional collaborative filtering (CF) algorithms,and (ii) “how” to weight the similarity between users and itemsby exploring the effect of different time-decay functions. As theresults of our extensive evaluation conducted over five social taggingsystems (Delicious, BibSonomy, CiteULike, MovieLens, andLast.fm) suggest, the step (when) in which time is incorporated inthe CF algorithm has substantial effect on accuracy, and the typeof time-decay function (how) plays a role on accuracy and coveragemostly under pre-filtering on user-based CF, while item-basedshows stronger stability over the experimental conditions.
Trattner Christoph, Parra Denis , Brusilovsky Peter, , Marinho Leandro
2015
Mutlu Belgin, Veas Eduardo Enrique, Trattner Christoph, Sabol Vedran
2015
isualizations have a distinctive advantage when dealing with the information overload problem: being grounded in basic visual cognition, many people understand visualizations. However, when it comes to creating them, it requires specific expertise of the domain and underlying data to determine the right representation. Although there are rules that help generate them, the results are too broad as these methods hardly account for varying user preferences. To tackle this issue, we propose a novel recommender system that suggests visualizations based on (i) a set of visual cognition rules and (ii) user preferences collected in Amazon-Mechanical Turk. The main contribution of this paper is the introduction and the evaluation of a novel approach called VizRec that is able suggest an optimal list of top-n visualizations for heterogeneous data sources in a personalized manner.
Mutlu Belgin, Veas Eduardo Enrique, Trattner Christoph, Sabol Vedran
2015
Identifying and using the information from distributed and heterogeneous information sources is a challenging task in many application fields. Even with services that offer welldefined structured content, such as digital libraries, it becomes increasingly difficult for a user to find the desired information. To cope with an overloaded information space, we propose a novel approach – VizRec– combining recommender systems (RS) and visualizations. VizRec suggests personalized visual representations for recommended data. One important aspect of our contribution and a prerequisite for VizRec are user preferences that build a personalization model. We present a crowd based evaluation and show how such a model of preferences can be elicited.
Kraker Peter, Trattner Christoph, Jack Kris, Lindstaedt Stefanie , Schlgl Christian
2013
At the beginning of a scientific study, it is usually quite hardto get an overview of a research field. We aim to addressthis problem of classic literature search using web data. Inthis extended abstract, we present work-in-progress on aninteractive visualization of research fields based on readershipstatistics from the social reference management systemMendeley. To that end, we use library co-occurrences as ameasure of subject similarity. In a first evaluation, we findthat the visualization covers current research areas withineducational technology but presents a view that is biasedby the characteristics of readers. With our presentation, wehope to elicit feedback from the Websci’13 audience on (1)the usefulness of the prototype, and (2) how to overcomethe aforementioned biases using collaborative constructiontechniques.