The Dynamic Recommendation Algorithm Based On The Time Factor

Information Recommendation is an important means to solve the informationoverload, and has been extensive research and application, but lack of dynamiccharacteristic research restricts the current recommender system development. Thetraditional recommender system is based on the relationship between users and items,by analyzing a user’s behavior history or evaluation data to predict the size of theinterest of the user on a given item, but they ignore the important role of the timefactor in the model. Time information on the recommended system did not fully playout.Above analysis, In order to solve existing recommender system ignoring the timefactor on the recommendation result, leading to the recommendation result lowaccuracy, this paper investigate the temporal recommendation problem by analyzingsome public released data sets. Following are main contributions of this paper:First of all, in the second chapter describes the dynamic recommendation systemstatus, the time factor can be decomposed into three major aspects: Item timeattributes, User interest time attributes, Time hotspots effect. Based on the three timefactors summarized above, the third chapter in this thesis made a detail decompositionand demonstration. Considering the effects that the three aspects of the dynamicnature bring, we can make the final results of the recommendation more rational andmore efficient.Secondly，combined with the analysis of time factor in Chapter three, I proposeda dynamic time graph model(TGM) to predict user’s interest, and applications in theTop-N recommendation issue that recommend the N most likely items of interest tothe user. This model based on user item dichotomy graph, through introducing itemtime node and user time node to reflect the impact of the time factor of therecommendation process, and then the establishment of a dynamic time graph modelon user interest prediction converted the measure of user interest value in items into the similarity between nodes in the time graph. In the graph model, using the pathfusion algorithm to calculate FIG similarity between the user nodes and item nodes,thus achieving the purpose of the prediction of user’s interesting. The same time, thehotspot time effects on the recommended results do specialized treatment.Experimental results show that, based on the analysis of the time factor impacton the recommender system, our method predicting users’ interests in dynamic timegraph model can make better experience than non-temporal methods in therecommender system.