Real-Estate Recommendation Engine Based on User Navigation Behavior
Pavel Chen · 2011
Content recommendation has become popular in web applications of various types. It is used in e-commerce web pages, social networks and search engines. Different approaches are utilized which rely mostly on traditional server-side web log mining, page semantic analysis. However there is more information that can be used for recommendation of higher quality: user navigation behavior.In this thesis work we propose a novel approach for recommendation which is based on pointing devices actions, e.g. clicks, right-clicks, trails, text selections, etc., combined together with user defined weights of various characteristics of presented content. Actions are captured on client-side with recommended content being instantly displayed on a user screen. Particularly this work is targeted at real estate search engines where real estate properties are being recommended based on how users navigate search results. The given problem was seen as a classification problem where user navigation behavior data served as a feature set, and machine learning algorithm was selected to solve it. Results of initial training data capture and analysis (84% classification accuracy) and then further model validation (79% of positive responses) were successful and this concept proved to be promising.