Personalized Web Content Recommendation based on LDA Profile
Hiroshi Fujimoto, Minoru Etoh, Akira Kinno, Yoshikazu Akinaga · 2011
We propose a web content recommendation method based on latent topic modeling such as LDA. The main technical challenge is how to symbolize web access actions, by words, which are monitored through a web proxy log. We have developed a hierarchical URL dictionary and a cross-hierarchical directory matching method which provides automatic abstraction functionality. We also propose a recommendation scheme based on LDA model which recommends unseen contents as well as seen ones in the past. We show recommendation effectiveness of our method by applying to proxy data of 7500 students in Osaka University.