The Research Of Adaptive Recommendation System Based On Content

With the rapid development of internet technology,people step into the era of information overload from an era of information scarcity gradually. How to get our required information from huge amounts of information in this era quickly became a research hotspot. Due to good performance in information filtering, recommendation system become an effective way to solve information overload problem and bring huge commercial profits. Therefore recommendation system has great research value not only in business application but also in academic research.Among many kinds of recommendation system,recommendation system based on content has been widely used in the field of text recommendation. This paper majors in the construction of the initial user profile and user profile updating in the process of recommendation in content recommendation system. The author puts forward an adaptive recommendation system based on content applied in text recommendation, in order to improve the accuracy and efficiency of recommendation.In the respect of building initial user profile, this paper presents a method of building initial user profile based on Text Rank which make full use of the character of cluster in users’ information. By taking a series of measures like determining the meaning of each word, clustering, establishing graph models, introducing various influence factors to make the Text Rank transition probability matrix better, an accurate initial user profile is built when just having little information, which can improve the accuracy of recommendation effectively.In the respect of updating user template, this paper introduces the concept of pseudo relevance feedback in information retrieval to this recommendation system to update user profile at the same time of using the new data user providing updating the template. A series of operations are taken to get optimal feedback documents, select keywords and update user profile by using improved Rocchio algorithm so that the system can avoid the introduction of the noise term, expand the scope of recommendation and achieve better recommendation results.Experiments show that the accuracy of the recommendation is high in this system.