logo-print

Chair Web Engineering (WE)

Information on Collaborations (with Industry)

MastersPortal.eu (http://www.mastersportal.eu/) is a website which aims to provide detailed information about all Masters programs in Europe. It has seen tremendous growth, both in number of visitors (over half a million visitors per month) and programmes listed (over 13 000) just in a few years. With this growth the company has encountered a multitude of challenges, both technical and commercial in their nature. One of the serious concerns was a very high bounce-rate of visitors coming from Google, the single biggest referrer for the MastersPortal.eu website. Thijs Putman in his MSc dissertation (guided by M. Pechenizkiy) developed and deployed a program recommender solution that has resulted in an increase in visitor retention of over 90%. Besides, an R&D infrastructure has been developed for a quick prototyping, deploying and testing of new techniques in a sound way. Both the recommender and R&D infrastructure are now deployed and used at the company.

Teezir B.V. (http://www.teezir.nl/) is a search solutions company that uses state-of-the-art, proprietary search technology solutions to improve various businesses. In indexing, search, content extraction and opinion mining on the web it is important to have a solution of extracting the main (relevant) content from web pages. Most of the previous approaches used heuristic rule sets to locate the main content. Such approaches usually not generic (do not work well with previously unseen structures and formatting) and require considerable efforts for designing implementing new and new heuristics. Samuel Louvan in his MSc dissertation (guided by M. Pechenizkiy) developed web content extraction module which uses a hybrid solution consisting of machine learning and heuristic approaches. The evaluation of developed module showed that it was a competitive content extraction method compared to the state of the art web content extraction methods and therefore it was decided to become a part of Teezir solution.

C-Content (http://www.c-content.nl) is a company providing content integration solutions and professional information retrieval (particularly legal documents) providing direct access to all information needed by a professional user in a specific segment through one single search system. Bas Gijzen in his MSc dissertation (guided by M. Pechenizkiy) developed a tailored automated document classification system for Rechtsorde, an online portal containing about 400,000 Dutch legal documents (at the moment of spring 2007). This system has been deployed and used successfully in the company.

Adversitement B.V. (http://www.adversitement.com/) is an e-marketing consultancy specialized in web analytics – understanding visitors behavior. Guido Budziak in his MSc dissertation (guided by M. Pechenizkiy) addressed the case of NUON (http://www.nuon.nl), a large energy providing company that has an extensive self-service portal on their web site and a large call center to answer telephone questions. Guido applied data mining solutions for getting more insights into the use of the self-service portal on its own and in relationship with the call center. Based on this case he developed a generic framework and reference architecture to derive segment definitions for web analytics suites based on customer cross channel usage behavior. Based on this successful combination of applied research and deployment, Guido decided to continue his work at Adversitement and combine it with pursuing PhD in WE group under supervision of M. Pechenizkiy.

Sligro B.V. (www.sligro.nl/) is a food service company. As many other companies in the food and beverages market it has to deal with short shelf-life products and uncertainty and fluctuations in consumer demands. Sales or demand prediction is known to be a challenging problem. The company was using a simple six weeks moving average strategy as a basic predictor that had to be adjusted by domain experts. Having over 60 000 products in stock, it is not surprising that for a domain expert it is hard to make stock management effective, if the basic predictor provides an inaccurate output. Patrick Meulstee in his MSc dissertation (guided by M. Pechenizkiy) developed an ensemble learning method which showed promising results for products having seasonal behavior. This research was continued at WE group by I. Zliobaite, J. Bakker and M. Pechenizkiy. As the main practical result we achieved 5-10% improvement in demand prediction accuracy for products showing different types of sales patterns and developed a method for controlling the risk of having bad predictions.

DWA (http://www.dwa.nl/) is an SME specialized in installing heating and cooling devices in large scale building projects. The installations DWA installs into buildings are very complex and modeling them based on the laws of Physic and the specifications given by the manufacturers of the individual components is a cumbersome and often even impossible task. Therefore, Siem Opschoor proposed in his MSc dissertation (guided by Toon Calders) to use data mining techniques to model the ‘normal’ behavior of such a heating/cooling system during the start-up period when the installation is delivered to the client. Based on the model of ‘normal behavior’ learned by data mining, one can now predict, in an early stage, if the systems starts malfunctioning. This data mining technology has now been implemented successfully by DWA into a monitoring tool.

Recently, in Philips, a game board has been developed to study the learning behavior of kids. On the game board the children can perform a wide variety of pattern matching and memory-oriented tasks, which are logged by the game board. Based on this information the goal is to identify problem solving strategies employed by the children in order to incorporate this information later on in an intelligent automatic tutoring aid. Ying Zhang shows in her MSc thesis (guided by Toon Calders) how data mining and statistical analysis tools can be used to extract useful information from the game board.

Turpin Vision B.V. is a Dutch company that produces the content management system Content-E. Guided by dr. ir. Alexandra Cristea master students added authoring of adaptive content into Content-E. Turpin Vision also became a partner in the EU Socrates-Minerva ADAPT project, studying the addition of adaptive collaborative elements to e-learning environments, including Content-E.

Magnaview B.V. (http://www.magnaview.nl/)is a Dutch company that specializes in interactive data visualizations. (It's main product is also called Magnaview.) Students guided by prof. dr. Paul De Bra worked on the optimization of the data structures and algorithms, resulting in (on average) a 300-fold improvement of the overall performance of the system. As part of the ITEA Passepartout project TU/e and Stoneroos BV collaborated on the creation of iFanzy, an adaptive TV recommender system. Based on ratings (of TV programs, shown at a certain time, on a certain channel) and on a multitude of databases about TV and movie data iFanzy provides recommendations for the TV viewer. iFanzy is being commercialized by Stoneroos and is integrated into (currently 5000) set top boxes, available on the Web and on mobile devices. The research at TU/e was performed by Pieter Belleken and Martin Björkman and guided by dr. Lora Aroyo (and later also by prof. dr. Geert-Jan Houben and prof. dr. Paul De Bra).

The Rijksmuseum (http://www.rijksmuseum.nl/) is partner in the NWO CATCH project CHIP (Cultural Heritage Information Presentation) in which an art recommender and (Web-based and mobile) guided tour generator were developed. Several master students, a PhD student (Yiwen Wang) and programmer (dr. Natalia Stash) worked under the guidance of dr. Lora Aroyo on the recommender strategies, based on available data and metadata about artworks. They also developed a collaborative recommender, making use of similarities in the ’taste’ of different visitors. The Rijksmuseum plans to integrate the recommender and tour generator into its infrastructure once the museum reopens in its main building.