Research On Content-Based Personalized Information Recommendation For The Real Estate

Along with the high-speed growth of the Internet, information overload problem turns worse increasingly and personalized information recommendation becomes a hot research topic. This paper takes the real estate information as a research object, putting forward a new recommendation model of personalized information for the real estate. This method extracts structural attributes from the real estate information through natural language processing. Then, the user profile is established through combining key words of user need and structural attributes. Visual attention mechanism is added in this model which makes the information presentation suitable to the user's selective attention.The main research results are:1. Rule-based structural information extraction of the real estate domain In this paper, structural attributes of the real estate, such as price, size, etc., are extracted through the establishment of rule base. Then, incomplete attributes after extract processing were dealt with using rules. The experimental results show that precise rate of structural information extraction is up to 99%～100% and the recall rate are higher than 80%.2. User modeling based on both keywords and attributesThe user profile in this paper is expressed by keywords as well as the relationship between them and the structural attributes. The model achieves personalized recommendation through a search based on keywords and the similarity calculation base on attributes. Experimental results show that the principle of finding a house which is beyond expression in traditional house service can be expressed by this model. This makes a better expression of the user needs.3. Display of recommended information based on attentive mechanismThis paper analyses the feasibility of guiding user's visual attention from visual attention mechanisms. The model of information recommendation display based on visual attention mechanism and corresponding algorithm are proposed according to special location effect in visual selective attention, analysis of sightline track and the model of saliency-based visual attention. The experiment shows that this display method is 30% more efficient than traditional list view method and it provides a new direction for the display of personalized information recommendation.