Revisiting Interest Indicators Derived from Web Reading Behavior for Implicit User Modeling

Today, intelligent user interfaces on the web often come in form of recommendation services tailoring content to individual users. Recommendation of web content such as news articles often requires a certain amount of explicit ratings to allow for satisfactory results, i.e., the selection of content actually relevant for the user. Yet, the collection of such explicit ratings is time-consuming and dependent on users' willingness to provide the required information on a regular basis. Thus, using implicit interest indicators can be a helpful complementation to relying on explicitly entered information only. Analysis of reading behavior on the web can be the basis for the derivation of such implicit indicators. Previous work has already identified several indicators and discussed how they can be used as a basis for user models. However, most earlier work is either of conceptual nature and does not involve studies to prove the suggested concepts or relies on meanwhile potentially outdated technology. All earlier discussions of the topic further have in common that they do not yet consider mobile contexts. This paper builds upon earlier work, however providing a major update regarding technology and web reading context, distinguishing between desktop and mobile settings. This update also allowed us to identify a set of new indicators that so far have not yet been discussed. This paper describes (i) our technical work, a framework for analyzing user interactions with the browser relying on latest web technologies, (ii) the implicit interest indicators we either revisited or newly identified, and (iii) the results of an online study on web reading behavior as a basis for derivation of interest we conducted with 96 participants.