| At present,television is gradually developing towards networking and intellectualization.Traditional TV can only receive information from the radio and television service center.Now TV has more interactive functions with the Internet.Thus,TV has changed from one-way reception to two-way interaction.How to make full use of the characteristics of the current television intelligent network is a new direction of current research.At present,there is a serious lack of classified data for TV users.Social media classification data is relatively easy to obtain.Considering the weak correlation between them,social media data may be helpful for TV user classification.1.Users are classified according to the real user data provided by the laboratory.However,due to the small amount of real user classified data,in order to solve the problem of small amount of real TV user classified data,we try to mine a large number of"micro-blog TV user data"with weak correlation as training data.2.Through social network analysis,mining users and information,extracting the relationship between user characteristics and watching hot programs.Through the topic mechanism of microblog,the users under the topic are found and visited in turn.Grab user specific information and store it in database.3.By using the method of neural network,the model between user's watching hot spots and user's characteristics is established.After word segmentation,the micro-blog data is matched with the user's"relevant hobbies"by dictionary,and the user is classified accordingly.According to the classification results,the user's training data is constructed,and the multi-layer neural network is constructed for training.4.In order to solve the problem of incomplete coincidence between"micro-blog user group"and"real TV user group",the user model built above is used to transfer the learning method into parameters and train with real TV user classification data.Generating a new model of TV user data after transfer learning for predicting real TV user classification.5.User portraits of smart TV users in this area are made by using the model trained by transfer learning,and the experimental results are analyzed by comparing the microblog data with the real classified data of TV users.The experimental results show that the model built by neural network and transfer learning can obtain more information close to the real user classification distribution. |
