The Use Of Learning Automata In Modified Latent Factor Model Recommendation Algorithm

The target of a recommendation algorithm is to recommend to users the items in which they are most likely interested, thus allowing both users and vendors to get maximum benefits. With the advent of the era of e-commerence and big data, whether a recommendation is good or not is more and more becoming a decidion factor of success of a site or application.Latent factor model(LFM) algorithm has became a hot spot in the field of recommendation algorithm, because of its good adaptive and high predictive accuracy, as well as its good performance both in rating prediction and TopN recommendation. However, there’s still room for improvement. In this paper, the research is expanded aroud LFM and its usage in recommendation system.Before applying LFM, the user preference model need to be trained, and currently gradient decent method is employed to train the model. To this end, this paper firstly discussed the training process of LFM. Considering that Learning Automata(LA) can achieve a global convergence and perform good in a noisy environment, this paper proposed a training algorithm named CALA-TM, which is based on continuous action-set learning automata. Then the modified training algorithm is applied in rating prediction. Experimrnt shows that the modifeid training algorithm can improve the prediction accuracy in a sparse data set.Further, consedering that CALA-TM has a slow convergence speed, another modified training algorithm LAGD-TM is given, based on the conjuction of CALA and gradient decent. Experiment shows that LAGD-TM has a faster convergence speed than CALA-TM, and also can achiece a relatively high prediciton accuracy.In addiction, the usage of LFM in TopN recommendation is also discussed in this paper. Using above fidings, the sorting algorith is also modified by taking use of the classification information. Experiment shows that the modified recommendation algorithm can improve the recall in Top20 and Top10 recommendation.