VTT is developing methods for automatically recommending articles and advertisements for readers of online newspapers in a project led by the Norwegian Adressavisen newspaper. The recommendations increase the time readers spend on online newspaper websites and enhance the reading experience. One of VTT's roles in the project is to assess how information confidentiality can be ensured. VTT's own UPCV recommendation method has been developed to address this need.
VTT Technical Research Centre of Finland Ltd's expertise in recommendation technologies is internationally recognised both with regard to content-based and behaviour-based technologies. This expertise has been developed, in particular, within the Finnish Next Media programme. In this program VTT has co-operated with NxtMedia Norway –project.
Recommendation methods collect information on visitors' behaviour, process this information and provide recommendations based on it. The purpose of recommendations is to offer even more to visitors than they were originally looking for and help them to only view the information that is relevant for them. This will enhance the reading experience and increase the time visitors spend on the website, as well as their likelihood of returning to it. It will also increase the number of advertisements clicked.
The UPCV method guarantees the confidentiality of information
The collection and analysis of user data requires a high degree of confidentiality of the service as as to avoid the misuse of the collected data. Traditional behaviour-based methods often require long-term collection of data separately on each individual visitor. The method developed by VTT, UPCV (Ubiquitous Personal Context Vectors), addresses this issue in a different way: in UPCV, random values are changed in connection with each user transaction, and these random values are the only data saved on the transactions.
"This means that no data related to the actual transaction needs to be stored. The recommendations are based on the similarities between the random value stacks generated," says the chief developer of the UPCV method at VTT, Senior Scientist Ville Ollikainen.
Another unique feature of the method is that the random values assigned to consumers and the recommended item – all that is required for recommendation – can be separately stored and independently used. Similarities between newspaper articles, for example, can be assessed by comparing their random value stacks. With this method, no user-related data needs to be revealed.
As the amount of information grows, so does the importance of recommendation methods
In Finland, VTT has used the UPCV method for creating a learning recommendation service (suosittelija.fi) for the Helmet library portal, which currently covers 78 public libraries in the Helsinki metropolitan area. For the first time, library users can now receive lending recommendations on the basis of their own and other borrowers' earlier book loans.
"Big Data has long been a hot topic. As the amount of information grows, there is also a growing need for technological solutions that enable individual users to locate the information that is most relevant for them. Recommendation technologies are a good example of this," remarks Caj Södergård, Research Professor at VTT.
In the Norwegian recommendation technology project"Anbefalingsteknologi", VTT will compare recommendation methods with regard to their confidentiality, among other features, including VTT's own UPCV method. In addition to Adressavisen and VTT, project participants include the Norwegian University of Science and Technology NTNU and IT company Cxense.