Research And Implementation Of Adaptive Hybrid Recommend System Base On IPTV

IPTV serves as the representative product of the new generation of digital cable TV earning the popularity for the inborn characters of its high Interactive and convenience.IPTV content provider can supply with digital image, video, audio, online game, distance education, advertisement with metadata and high fidelity. In this scenario, users have the propensity to lose orientation among the vast quantities of information. Consequently, it is urgent to provide high quality personal service. And so far the research based on personal service can be mostly classified as recommend system.In this paper, we first analyze the deficiency and disadvantage of the algorithm adopted by the current recommend system including code start problem and the downgrade of quality caused by the solid proportion adopted by hybrid filtering algorithm. We take these problems into consideration, and arrange the paper as follows.First of all, we propose a collaborative filtering algorithm based on demographic characteristics to solve the code start problem. This collaborative filtering algorithm makes full utilization of the information similarity of the demographic characteristics collaborating with the similarity results from PCC calculating to acquire final similarity results. The experiment results indicate that collaborative filtering algorithm based on demographic characteristics can increase the quality of recommendation effectively in the circumstance where both the users' rating and user profile are scarce.Then, we resolve the problem ascribe to the traditional hybrid recommend system by proposing a hybrid self-adaptive recommend system based on gradient-descent method. This algorithm employs self-learning mechanism to change the proportion of the hybrid recommendations automatically. The result procured from the experiment demonstrates that the algorithm proposed increases the recommendation precision on some level and does not take up extra calculating time. And finally, we summarize the characteristics that recommend system based on IPTV should possess- user zero cost self-learning and user no extra operation by analyzing the feature of IPTV platform and comparing it to existing recommend system which takes personal computer as terminal. Due to the feature we analyze, a user personal data mining algorithm is proposed through analyzing user' web access log and procuring users' preference automatically. As far as we are concerned, there is rarely another algorithm based on IPTV and the algorithm is proved doable and effective through one year operating experience online.