As an Industrial PhD working with Adform (www.adform.com), my research involves investigation and development of new models for predictive targeting in online display advertising based on statistical learning. Our main hypothesis is that the actions a user performs when navigating and interacting with web sites (clicks, page visits, previous conversions, etc.) reveal weak signals which can be used to infer user behavior and goals.
The rarity of user actions poses a challenge for ordinary collaborative filtering (CF) techniques; hence I investigate augmenting CF with side-information about the users and how that can be used for inferring user similarity and thus improve the predictive performance in cold-start settings.
The dynamics of online user behavior is another area of research. Adform’s role as a key international digital advertisement platform, offers unique possibilities for closed-loop experiments. I.e., my research entails investigating active learning strategies to reduce uncertainties of model parameters in the face of data in which our algorithms have been used to decide on test candidates (banners).
Publication: Research › Ph.D. thesis – Annual report year: 2015
Predicting clicks in online display advertising with latent features and side-information: Technical report
Publication: Research › Report – Annual report year: 2014
Publication: Research › Journal article – Annual report year: 2014
Latest activities and conferences