Research on User-personalized Image Retrieval Method

Yu Song, Jing-fei Ren, Mao-Zhu Jin, Pei-Yu Ren
2014 International Journal of Multimedia and Ubiquitous Engineering  
With the rapid expansion of information resources, the amount of image data in the network shows an explosive growth trend. The traditional search engines have not considered users' different interests; therefore image retrieval efficiency is reduced. To solve the problem, this paper puts forward a research on user-based personalized image retrieval technologies. Firstly, this paper studies the user interest model, and provides its definitions and application strategies; secondly, it studies
more » ... laborative filtering algorithm based on Kmeans clustering, and solves the problem of sparse resources effectively; Finally, explicit tracking, implicit tracking and relevance feedback methods are adopted to learn and update user interest model constantly to meet the users' needs and improve retrieval accuracy and efficiency. Based on the above studies, this paper presents a kind of user-based personalized recommendation technology, and completes an image retrieval system based on user personalization, proving that this recommendation technology is able to provide users with better personalized recommendation service. intelligently by drawing diagrams and complex network theoretic technology, and under that circumstance the performance of the recommendation system was improved. Among various recommendation technologies, collaborating filtering technology is the most classic personalized recommendation technology. Wang Qian [6] converted users' evaluation on a project into the calculations of users' preference on certain projects, therefore the nearest user group could be calculated. Li Feng [7] and Xia Xiufeng [8] put forward a personalized recommendation technology that based on product feature. Nowadays, the major element that affects the recommendation accuracy of collaborating filtering technology is the so called sparse data, namely when considering about the nearest users' evaluation on resources, the evaluated resources are little compared to the total resources in the system, and this leads to a scarce and sparse evaluation data given by nearby users. The sparse data fails the system from accurately confirming the nearest user groups of the targeted users and therefore it cannot conduct a high-quality and high-efficiency personalized recommendation to targeted users [9]. Among personalized recommendation systems, user interest model [10] is the core of the system when offering personalized recommendation services to the users and it is established by recording all kinds of users' behavior information. As users search more, the system continues to amend user interest model. Hsu adopted user interest model in medical image retrieval and it worked well [11] . In personalized recommendations, relevance feedback technology is used to perfect user interest model so as to better reflect users' demands. The thought of relevance feedback technology is to adjust recommendation mechanism by using information of users' feedback on the results of the system. And the purpose is to provide more accurate and more reliable recommendation service. At present, in terms of image retrieval, the major relevance feedback technology adopted in personalized recommendation technologies is man-machine coordinated and interactive learning method [12] . Yin used users' retrieval logs as the feedback information to conduct image semantic clustering [13] . 43 means-based collaborative filtering technology and user-based personalized recommendation algorithm. This system is a user-based personalized image retrieval system and can provide users with personalized image retrieval and recommendation service. Therefore, in terms of developing model, B/S model is adopted to facilitate the separation of user information and system data. This system employs a Tomcat 7.0 -based and JDK1.7based integrated developing environment; the foreground developing adopts My Eclipse 10.0 developing application program; the system's design language is Jsp and JavaScript, and the background data base is MySql.
doi:10.14257/ijmue.2014.9.6.05 fatcat:gvmiuebjiffo5lholvchjbmv2q