Transfer of technology
Towards personally recommended web content
Anne-Marie Kermarrec, Inria researcher
Anne-Marie Kermarrec, a researcher at Inria Rennes - Bretagne Atlantique research centre, is planning to found a start-up company based on her work on a scientific project funded by the European Research Council (ERC). Her aim is to help website publishers offer content that is individually tailored to the requirements of each user.
The story begins in 2007. A lifetime ago in terms of the Internet. Twitter had only just been born. Facebook was taking its first hesitant steps. A director of research at Inria Rennes - Bretagne Atlantique centre, Anne-Marie Kermarrec submitted an application to the ERC for funding for a project known as Gossple. Her idea was to use decentralised peer-to-peer (P2P) technology to reveal the affinities and resemblances between web users. These would then be used to build sets of user profiles corresponding to their preferences, on the basis that it was possible to learn a lot about someone you didn’t know provided that you both shared a number of points in common. The application was approved. A research team was established and they began five years of research during which the web changed out of all recognition, throwing their work into sharp focus. “Social networks became more and more important. People were increasing uploading data ”. Data that included holiday snaps on Flickr, hotel reviews on Tripadvisor, recipes on Allrecipes, CVs on LinkedIn, and old school friends on Friends Reunited. The digital world increasingly became a reflection of our lives and personalities.
At the same time, a new technology, notification, was beginning to emerge. It was no longer necessary to actively search for information using a search engine. Now, data could be fed directly to the user by means of a system of recommendations and suggested links for further browsing. However, for this system to work it was essential that the suggestions fitted perfectly with the requirements of each individual web user. And things were getting even more complicated. “Many sites were using our browsing histories as a means of personalising content ”. This is an old technique used by many internet shopping sites. Having bought a printer, for example, the user is then continually offered ink cartridges. “However, this history is not the only relevant data available ”. Other forms of personalisation include subscriptions to themed data streams. However, these often suffered from poor quality filtering. “If I subscribe to an RSS feed on a subject such as sport, a lot of the data that I receive will be irrelevant ... or no longer of interest to me. Our interests change a lot over time ”. The filters need to be finer, and continuously updated with more recent data.
“Our approach has been different in that we consider data from a large number of users. We associate each person with their implicit social network, i.e. other similar users who may or may not be known to that person. We then use this network to construct a dynamic recommendation network. At the beginning, our project was based on the use of decentralised peer-to-peer techniques. Having worked with P2P algorithms, we are now ideally placed to rebuild an effective system with only partial data ”. The software does not suffer from problems of scale. It also preserves the anonymity of users by separating personal information from the preferences expressed. “Having completed this research, it seemed to me that we could go further. By extending the ERC grant into 2013, we have been able to take advantage of ERC Proof of Concept funding. This additional funding is intended to allow scientists to continue working on their prototypes with the aim of exploiting their work commercially. We are now working towards creating a start-up company. ”
Personalised recommendation plug-in
The solution being offered to web publishers takes the form of a plug-in built into the web page. The visitor will see a selection of personalised links reflecting their dynamically established profile. “A number of user panels have tested our system at the Trento Rise centre of excellence in Italy ”. The business will begin trading in 2014. Who will be the first customers? “It’s too early to say. However, these techniques will be of interest to a number of potential customers, including information providers looking to increase traffic and maximise the length of time readers spend on their pages ”. The charging structure is also still to be defined. “We may charge by means of a commission per click or a subscription. A consultant is helping us to prepare a business plan as part of the ERC Proof of Concept project. This project is also receiving funding from EIT-ICT Labs. Finally, we are receiving tremendous support and encouragement from the Inria technology transfer, patents and IT Translation departments. Scientists are very well supported by the Institute, and that is very motivating ”.