Research Of The Deep Learning Based Intelligent Recommender System

Currently,recommender system has been widely used and exhibited great commercial benefit to the enterprise.Among various methods for recommender system,collaborative filtering is the most popular one since it is able to achieve better recommendation accuracy than the others.Generally,the methods based on collaborative filtering rely on the historical interactions between users and items heavily.Especially,applying machine learning to collaborative filtering has been dominated for years as its superior prediction performance.Although traditional machine learning methods have acquired great success in recommender system,the following problems still remain: 1)the problem of capturing complex relationships;2)the interpretability of prediction;3)cold start problem;4)static problem.In contrast with the traditional machine learning methods,deep learning has the following advantages: 1)able to handle more complex nonlinear relationships,2)automatically extracting features layer by layer from raw data;3)good versatility;4)Suitable for large-scale data.Thanks to above advantages,deep learning based methods have achieve remarkable successes in different fields.In this thesis,the study focus on deep learning based collaborative filtering methods,where a series of effective solutions are proposed to overcome existing problems.The main works in this paper include:(1)A deep learning based collaborative filtering for rating prediction problem.Existing CF-based methods can only grasp the single type of relation,such as Restricted Boltzmann Machine which distinctly seizes the correlation of user-user or item-item relation.On the other hand,matrix factorization explicitly captures the interaction between them.To overcome these setbacks in CF-based methods,a novel deep learning method is proposed which imitates an effective intelligent recommendation by understanding the users and items beforehand.In the initial stage,corresponding low-dimensional vectors of users and items are learned separately,which captures the semantic information reflecting the user-user and item-item correlation.During the prediction stage,a feed-forward neural network is employed to simulate the interaction between user and item where the corresponding pre-trained representational vectors are taken as inputs of the neural networks.Several experiments based on two benchmark datasets are carried out to verify the effectiveness of the proposed method,and the result shows that our model outperforms previous methods that used feed-forward neural networks by a significant margin and performs very comparable with state-of-the-art methods.(2)A attention based collaborative filtering for rating prediction problem.Neighborhood based collaborative filtering is a method of high significance among recommender systems,with advantages of simplicity and justifiability.However,recently it is receiving less popularity due to its low prediction accuracy in contrast with model-based collaborative filtering systems,but model-based methods also suffer from a drawback worthy of attention that is they can not effectively explain the reason behind their estimation.In order to develop a system with both high accuracy and justifiability,a novel neighborhoodbased collaborative filtering method is proposed which is inspired by the natural mechanism of attention.This method can adaptively find neighborhood items to the prediction in user history without any pre-defined function with respect item correlations.Then the estimation are made based on these relationships.Experiments on several benchmarks are carried out to verify the performance of the proposed method,and the result shows that this method can beat all previous state-of-the-art methods in addition to being able to justify the prediction obtained.(3)A deep reinforcement learning based collaborative filtering for Top-N recommendation problem.Recommendation is a sequential decision making.However,most existing recommender systems adopt a static view,consequently,they suffer from the problems of user cold-start and non-adaptability of changing user interests.To overcome these problems,a novel top-N interactive recommender system based on deep reinforcement learning is proposed.In this model,the processes of recommendation are viewed as a Markov decision process,wherein the interaction between agents and environments is simulated using a recurrent neural network thanks to its sequential nature.In addition,the reinforcement learning is employed to optimize these model for purpose of maximizing the long-term recommendation accuracy.Experimental results show that our model significantly outperforms the baseline methods in terms of accuracy,and solves user coldstart and non-adaptability to interests problems effectively.