Web analytics is aimed at understanding behavioral patterns of users of various web-based applications or services in e-commerce, mass-media, and entertainment industries. Accurately predicting the probability of desired actions on the web (product purchases, membership registrations, newsletter subscriptions, software downloads, accessing certain information resources, clicking ad banners) in specific circumstances would enable us to achieve better personalization and adaptation to diverse customer needs and preferences. The behavior of users may vary depending on the context (e.g. user activity, location, time, access device, weather, holidays) and potentially within the context. Thus, predictions in web analytics are inherently context sensitive, and therefore, complementing the prediction models with context management mechanisms are expected to make them more specialized and predictive analytics decisions for web applications more accurate. In general, the number of contextual factors that may potentially affect human behavior on the web is enormous and it is hardly possible to capture all of them with a model simpler than the universe itself. Therefore, one of the key challenges is to construct the mechanisms, which would identify, what the (current) context is and how to integrate it into prediction models. Another important aspect to be developed is the mechanism of monitoring the stream of user-related and contextual data over time to signal anomalies and changes in predictive model performance. Taking a broad range of practically relevant issues to address within this project, we aim for developing a complete solution integrating predictive analytics, context awareness and change detection mechanisms and allowing straightforward deployment of project results in web-based applications. The techniques we aim to develop will be tested retrospectively on historical data and deployed in real operational settings and validated externally. The particularly planned prediction tasks include but are not limited to ad banner relevance scoring, content matching in online advertising, bidding for sponsored search, content based and collaborative recommendations, and demand prediction for a given time period.

March 2013: Joep Fennema has started his graduation project: Quantifying and removing biases for predictive modelling in computational advertising

February 2013: Zhenyi Rong has started his graduation project: Distributed pattern mining for context discovery and predictive modeling

February 2013: Bram Keijers has started his graduation project: Reference framework for context-aware recommenders

February 2013: Timur Bagautdinov has started his Honors project (CSE Masters) Context-aware management of ad subsystems

February 2013: Christine Gerpheide and Luc Smet have started their graduation project: Honors project (CSE Master) Location-based ranking adjustment

6 February 2013: 1st user committee meeting took place @TUE/MetaForum

October 2012: Vitaly Kliger has completed his graduation project: Identifying and Utilizing Contextual Information for Banner Scoring in Display Advertising

September 2012: Jose Maria Luna visiting researcher: U. Cordoba started subproject on developing Context-mining framework with evolutionary approach

29 March 2012: CAPA kick-off meeting with HardwareInfo

27 March 2012: CAPA kick-off meeting with Kliknieuws

12 March 2012: CAPA kick-off meeting with Adversitement at TUE

1 March 2012: Julia Kiseleva joins CAPA as PhD candidate. Jeroen De Knijf joins CAPA as Postdoc

24 October 2011: We are hiring! The positions have been filled.
We are inviting interested candidates to apply for the open 3-year Postdoc and 4-year PhD positions.

24 October 2011: Essential extracts from the accepted project proposal can be found here. The document is intended to provide the basic information about the CAPA project for the prospective applicants as well as to the individual researchers and organizations interested in collaboration within the scope of this project.

23 September 2011: CAPA project has been accepted by STW! In half a year from now we should be able to start.

Zheyi Rong and Jeroen De Knijf: Direct Out-of-Memory Distributed Parallel Frequent Pattern Mining. In: Proceedings of BigMine@KDD'13, ACM.

Julia Kiseleva, Hoang Thanh Lam, Mykola Pechenizkiy, Toon Calders, Thijs Putman: Benchmark for user's trail prediction. Tech Report.

Julia Kiseleva, Hoang Thanh Lam, Mykola Pechenizkiy, Toon Calders: Discovering temporal hidden contexts in web sessions for user trail prediction. In: Proceedings of TempWeb@WWW'2013, pp. 1067-1074 (WWW Companion Volume)

Julia Kiseleva: Context mining and integration into predictive web analytics. In: Proceedings of DC@WWW'2013, pp. 383-388 (WWW Companion Volume)

Project poster can be downloaded here

The source code for Direct Out-of-Memory Distributed Parallel Frequent Pattern Mining is now available here: ds.tgz.

We have created a benchmark for the next user action prediction task: mp_may_2012.zip. The description is available in Tech report Benchmark for user's trail prediction.

We will keep trying to make the software, source code and datasets created and used within this project available for the research community (as long as there are no NDA, IP, ethical or proprietary concerns). Please check this section later.

You can reach the project leader dr. Mykola Pechenizkiy and the project team by sending an e-mail to capa.stw@gmail.com