Utilizing Social Breadcrumbs for User Profiling in Personalization

摘要: Personalization efforts aim to alleviate the “information overload” problem in an attempt help users address their information needs best way possible. An increasing number of systems that employ personalization have cropped up recent past with even well-known commercial giants targeting towards enhanced within services e.g. Amazon product recommendations, Netflix movie Google Now etc. A fundamental building block any is user model powers it. User modelling has remained a theme central broad research area most traditional sources for being controversial nature on account loss privacy associated them. With advent Social Web, paradigm shift occurred content generated Web leading it become online gathering point masses. Users now leave traces experiences various platforms referred as “social breadcrumbs” context this thesis. Recent began explore possibility utilizing data creation personalization-centric models; approaches attempted make use bookmarks and social tags modelling. These however are less effective few making bookmarking annotation tools rendering them infeasible large-scale application personalized applications. Given limitations current efforts, we explored network usage patterns personalization-related concerns derive aspects can lead profiles. The analyzed correlations led us proposition Twitter-based which takes into not only language under consideration but also his/her network. More specifically, framework based statistical models proposed. This enables probability distribution words user’s he/she employs over Twitter addition those whom he considers trustworthy (on Twitter). expressive depicted via incorporation two similarity measures whereby common utilized network-based measure, topical interests measures. To our knowledge, work constitutes one first attempts take generation proposed was extensively scenarios, namely search scientific articles’ recommendation, both these fundamentally quite challenging nature. For personalization, behaviors engages in. Adjustment parameters basis behavior-based heuristics demonstrate solution verified extensive offline experimental evaluations. Similarly, recommendation adjusted by taking followed users, replacing topic modelling-based filtering measure helps topics relevant interest. outperforms standard baseline produces rich recommendations articles user.