SWELL has supported and still supports several master students.
|Title||Persuading a knowledge worker using an adaptive coaching system. (tentative)|
|Name||Marloes van de Voort|
|Education||master Computer Science, Radboud University Nijmegen|
|Institute||Radboud University Nijmegen|
|Title||Manifestation of real world social events on Twitter|
Situations in which many people come together can result in very interesting and positive, lively social events. Unfortunately, large events with many people involved also include risks, ranging from small injuries to death. It would be useful if social media could be used to monitor these events, and to predict unwanted situations and casualties. To be able to know in advance how an event will develop based on information on social media, a good understanding of the relation between the development of real world events and their manifestation online is needed.
In this work, we study the relationship between real world events and their online manifestation. We create a model of both the real world event and its online manifestation. We compare these two models using data about five real world social events and Twitter data about them. We determine the correlation between the online and real world model. We find several weak to moderate correlations between online and real world characteristics of events. The intensity, readability and sentiment of the tweets are examples of variables in the online model that show a correlation with the real world and the weekends, school holidays and position of the moon are examples of real world variables which manifest themselves on Twitter.
|Education||Bachelor Information Science, Radboud University|
|Title||Associating words using MapReduce|
|Abstract||Patients suffering from aphasia often cannot remember which word to use, even though they are aware of context and meaning. Tools may be developed that allow patients to recover words related to a subject more easily. By automatically determining associations between words, a patient could be shown related words based on a word association graph. Such a graph could be created using the data from word co-occurence. However such a method of constructing an association graph requires large amounts of data. This data requires non-trivial processing and cannot be done on a single machine. In this thesis a method using MapReduce for distributing word association graph construction is proposed. By using an actor framework as a substitude for MapReduce and data from the social network Twitter, it is shown that association graph construction is feasable using MapReduce.|
|Name||Carles Salvador Soriano Perez|
|Institute||Universiteit Twente / Universitat Politecnica de Valencia|
|Title||Context Aware Systems in Well-Being: From User Needs to Software Architecture|
|Abstract||Mobile devices and sensors can be used in context-aware (CA) systems by gathering information from the sensors and the user's actions the system provides certain feedback. In this project we study the well-being domain defined in a context-aware system using Causal Loop Diagrams (CLD). When the user performs an activity the system collects information from mobile sensors and external agents, and with this information a feedback is provided to the user aiming to improve his well-being. Analysing the application Activity Coach helped identifying a transformation gap from the general well-being model to the application model. The solution proposed to prune the general model diagram following a bottom-up, a top-down or a combination of both approaches. For transforming the CLD application model we would use UML, but, how could the transformations be done? How much information does the application model need for correct developing? Would it be possible to produce a general model?. To answer the questions, this work proposed generating three different models, context structure, UML class and UML activity diagram, to represent context, implementation and behavior. After the study on previous works it was considered whether an existent CA framework could gave a better solution. If ConText Toolkit was used, the model obtained would be too strict for the development phase and its composition would mean a shorter battery life in the mobile devices. The evaluation of the solution approach showed that a choice should be made. Either using strict models, or open models for the transformations. For the developers, open models would require large assumptions and less information in the UML models, while strict models fewer assumptions and more information.|
|Education||Human Technology Interaction, TUE|
|Institute||Internship at Philips Research, 2013|
|Title||Collecting Heart Beats: Exploring the effects of progress feedback as a motivational intervention to increase physical activity|
|Abstract||Masters thesis was part of deliverable 5.4a; see the deliverables section in the results menu|
|Education||Bachelor Information Science, Radboud University|
|Title||Profiling knowledge workers using open online profiles|
This thesis explores the creation of user terminology models containing the interests, topics and expertise of a knowledge worker. These models are generated from existing, open profiles found on the web, specifically user profiles from Twitter, LinkedIn and ArnetMiner.
In order to correct for the sparseness of these profiles, information extracted from the network, related to the user, is used to enrich the profiles. The models are generated by using a frequency based scoring function, together with a background corpus.
The generated models are analyzed for overlap between networks and evaluated by their owners. Terms are rated for relevancy, specificity and whether the term belongs to the user’s professional or private profile.
Ultimately, the generated models proved to be of high average precision, but the overlap between models was quite low. Terms included in LinkedIn profiles were rated with the highest specificity, terms from Twitter models the lowest.
Academic profiles only contained professional terms, while terms from Twitter models were evenly distributed between private and professional. Including information from a user’s network didn’t show any improvement in the quality of the model.
|Education||Articial Intelligence at the University of Amsterdam|
|Title||Human activity classication based on arm and body movements|
|Name||Wouter van Teijlingen|
|Education||Utrecht University of Applied Sciences - Faculty of Natural Sciences & Technology - Institute for ICT|
|Title||Sensing activities - A signal processing pipeline for the use of a depth camera|
This thesis discusses a research project, with the aim to determine the accuracy and reliability of the Microsoft Kinect depth camera for sensing human activity. An experimental comparison between the Kinect and Xsens (i.e., an inertial motion capture system) was conducted, where Xsens served as the ground truth for measuring human-related activities. One reason for comparison is cost reduction: the Xsens equipment is priced around € 50,000 and the Kinect is € 100,0. The other reason is that the Kinect is non-obtrusive for humans. The Kinect is positioned somewhere in the room and it starts recording, while Xsens requires user’s to wear a complete body suit.
A program has been developed for recording human motions to capture video with the Kinect camera, baptized ONI Recorder. The first steps in mapping the Kinect and Xsens are realized by a software program called Kinalyze. See Figure 1 for the animations based on recordings of both Kinect and Xsens in parallel. The movements are synchronized, which means that they start and stop simultaneously. Using the Kinect as a partial substitute for measuring human activities is possible as illustrated in this research project.