Scalable distributed inference of dynamic user interests for behavioral targeting

A Survey of User Modelling in Social Media Websites

“…Recently, another issue has been addressed by a couple of recent published work (Ahmed, Low, Aly, Josifovski, & Smola, 2011;Gueye, Abdessalem, & Naacke, 2012;Li, Yang, Wang, & Kitsuregawa, 2007;Xiang et al, 2010) that is referred to as the dynamicity problem. We define it as the effect of time on user's preferences and how it can be reflected in their profiles.…”

“…We define it as the effect of time on user's preferences and how it can be reflected in their profiles. Modelling dynamic users' profiles can help in providing more quality services for users, such as, providing the right ad at the right time by emphasising the short term users' interests (Ahmed, et al, 2011). Moreover, recommender systems may use the dynamicity feature in order to enhance the predictions accuracy of users' ratings and in turn enhance the quality of recommender systems (Gueye, et al, 2012).…”

“…Topic modelling techniques are used in order to represent user interest as topics rather than keywords; this method is argued to provide a better performance (Ahmed, et al, 2011;Weng, Lim, Jiang, & He, 2010;Zhong, Fan, Wang, Xiao, & Li, 2012). Ahmed et al (2011) model users' interests as latent topics based on latent Dirichlet Allocation (LDA), where they maintain two distributions, users' distributions over topics and topics' distributions over words.…”

“…Ahmed et al (2011) model users' interests as latent topics based on latent Dirichlet Allocation (LDA), where they maintain two distributions, users' distributions over topics and topics' distributions over words. They used user queries to collect the words of interest for the user in order to enhance advertising targeting.…”

“…They adopt an iteration-based graph visualisation algorithm which allow them to identify active and not active interests, as well as into long-term, mid-term, and short-term ones. Ahmed et al (2011) provided dynamicity at three levels in their model, the global distribution over interests, the user-specific distribution over interests and the topic distribution over words. They also find short term and long term interests, and combine them using weighted average to get the expected user-specific's popularity over interests at a specific time.…”