Research On Trust-aware Collaborative Filtering Recommendation With A Denoising Autoencoder

The amount of data created by modern internet surpasses the requirements considerably as the online world is developing at a very high speed,which highlights the problem of the low utilization of these data.As a consequence,an increasing number of researchers begin to search for approaches to drain pragmatic information out of voluminous data.Recommendation system is coming to being one of many solutions,which can improve customer experience and business efficiency by finding data pattern and connecting costumers to specific commodities.Collaborative Filtering is a popular recommendation algorithm which is wildly used in user preference prediction.Nevertheless,conventional collaborative filtering algorithm can provide displeasing results when coming across sparse ratings.Furthermore,considering only linear calculation is involved in such an algorithm,it is difficult to provide satisfying suggestions in a complex environment.Hence,we are presenting two prediction models to demonstrate a possible approach to apply deep denoising autoencoder onto recommendation systems in order to get better performance.In this thesis,we mainly focus on:(1)TDAE,a prediction model based on explicit trust information,which formulas explicit trust as vectors and processes them along with user ratings by using denoising autoencoder,is proposed in this thesis.(2)An enhanced algorithm,TDAE++,is also proposed in this thesis.This improved algorithm extracts implicit trust information by Pearson similarity and models user profiles using the combination of both explicit and implicit trust information to get more reasonable results.Unlike the conventional collaborative filtering algorithm,our approaches can achieve better performance as we use the deep neural network with trust information to exploit both deep learning and social network methodology.Our experiments on data set Filmtrust,Epinions and Douban prove the efficiency of such approaches.Especially,TDAE++ is added to the implicit trust relationship model,which improves the accuracy of prediction to a certain extent.