A Neural Network-based Implicit Information Learning Model And Its Application In Recommendation

With the development of network technology and e-commerce,the information overload problem has become more and more serious.The recommendation algorithm has become an important factor to drive the development of e-commerce,which is one of the most effective methods to solve the problem.At present,most recommendation algorithms are affected by data sparseness problem and ignore the implicit information of rating data when learn users'performance.In order to solve the sparseness problem,this thesis proposes a multi-criteria rating data processing method(i.e.,MRDP)to process the dataset.In order to learn the implicit information of rating data,this thesis proposes a neural network-based implicit information learning model(i.e.,NNIIL)and applies it into recommendation.In order to verify the performance of NNIIL model,this thesis proposes two kinds of group-oriented recommendation:Group-oriented Multi-criteria Recommendation algorithm based on User(i.e.,GMURec)and Group-oriented Multi-criteria Recommendation algorithm based on Item(i.e.,GMIRec).The GMURec algorithm aims to learn users' preference on various criteria of item with NNIIL model,while the GMIRec algorithm aims to learn the attraction of various criteria to user.The main work of this thesis is as follow:(1)In order to solve the sparseness problem,this thesis proposes a method named MRDP to process the dataset.First,this thesis proposes a multi-criteria rating data filling method(i.e.,MRDF)to fill the dataset;Then,this thesis proposes a multi-criteria rating data conversion method(MRDC)to process dataset,which converts the users' rating data into user groups' rating data.Where,each user group can be considered as a kind of users with similar preferences.(2)In order to learn the implicit information from rating dataset,this thesis proposes a model named NNIIL.In order to verify the performance of NNIIL,this thesis proposed two kinds of group-oriented recommendation algorithm:GMURec and GMIRec algorithms.The GMURec aims to learn the users' preference on various criteria of item with NNIIL model,while the GMIRec algorithm aims to learn the attraction of various criteria to user.(3)In order to verify the performance of GMURec and GMIRec algorithms,this thesis designed four sets of experiments on the Tripadvisor dataset.The result shows that learning the implicit information with NNIIL model can improve the recommendation performance,and compared with the GMIRec algorithm,data sparsity has a greater impact on the GMURec algorithm.