Research And Application Of Recommendation System Related Technology

With the advent of the 4G Era,Great development had been made in electronic business,information searching,mobile application and social network in the circumstance of recommended system in the big data Era.With the increasing of the customers and the products of the application,billions of the resource information has showed up and scale of the data expanding continually so that the hitherto challenge is faced in recommendation system.Data sparse,extendibility and coldstart is becoming the main factors that affect the quality of the recommend system.Researchers have come up with varieties of methods to finish these problems off.Similarly,this paper devoted to solve data sparse,cold-start,the dynamic change of the customer character vary with time and the diverse dependency of different customer when treat the same feature.We improved the algorithm and built a renew data model to cope with these problem.The main works as follow:1.A new collaborative filtering recommend algorithm has been proposed combine customer characters and dynamic time.Collaborative filtering is based on the history marking to recommend.Meanwhile,coalescent customer characteristic information and historical marking information a self-adaptation weight model has been introduced in construct a classification tree,the number of the same character also been included.The influence of the recommend results is also analyzed in dynamic of customer character.As the cold-start,the sigmoid function has combined with customer characteristic model and traditional collaborative filtering calculates the similarity between customers.Finally,simulation experiments of algorithm accuracy,precision and recall rate has been introduced in to validate the algorithm efficiency.Results has showed that combine customer character classification with collaborative filtering recommend of dynamic time not only can promote the recommend accuracy,precision and recall rate,but also can relieve the cold-start problem of recommend system faced.2.A recommendation algorithm based on the noise reduction from the encoding is proposed.autoencoders neural network is that can extract data characters from low level to high level and find the potential relativity behind the samples.To diminish redundancy of the data,ZCA whiting technology has been used to transit or decorrelation the original data.Simultaneously,auto-encoders model has been improved to strengthen the robustness,then the recommend system based on the noise reduction and auto-encoders has come out.The main function is conquering the direct copy and print out problem when print in the samples of auto-encoders neural network by and the random noises.It is validated that the recommend accuracy is largely enhanced in this means.