Research On Intelligent Recommendation Algorithm For Online Video

Nowadays,with the continuous development of Internet technology,the types and numbers of various information on the Internet are increasing,making it more and more difficult for Internet users to find the content they want.For information providers,all information is presented to users to make User selection becomes no longer feasible.In this scenario,people need high-efficiency and high-accuracy recommendation algorithms to make information processing more effective.Appropriate recommendation algorithms can benefit information providers and consumers in both directions,which is a win-win algorithm.At the same time,the current deep learning technology has achieved remarkable results in the fields of speech recognition,image processing,etc.due to its deep-level architecture that can be learned and complex models.At present,in the research of deep learning algorithm data feature extraction,there are more researches on feature extraction for the whole sentence and the whole article,and less research on its application in the field of video recommendation.This article mainly focuses on deep learning technology and The recommendation algorithms are linked together to find a suitable model to be applied to the video recommendation algorithm to make the video recommendation algorithm more accurate and efficient.The specific research content is the recommendation algorithm for network video,After analyzing the characteristics of various algorithms.the traditional recommendation algorithm that is suitable for deployment in today's network video recommendation environment is determined.At the same time,the problems are combined with existing research.It is proposed to introduce a penalty item to punish the influence of popular videos liked by different users on the similarity between users,and deploy it in the actual network environment for experimentation,which is used to recommend the algorithm proposed later.Conduct comparative analysis.In order to cope with the increasing number of online videos,Aiming at some problems in the commonly used one-hot encoding in the past,it analyzes the internal reasons leading to these problems,and proposes a random vector encoding method that largely avoids the past one-hot encoding.Some problems with coding.The reasonable use of this method in data encoding reduces the dimensionality of the encoded data and reduces the computational complexity of the entire model.At the same time,for data such as video names that are not suitable for the random vector encoding,a method of processing using a convolutional neural network is proposed,which improves the structure of the image-specific convolutional neural network so that it can be applied to video recommendation field.When constructing the model,this article fully considered the current explosive growth of video website data and the increasing burden of servers.On the premise of ensuring that server data can be fully utilized,the complexity of the algorithm is reduced as much as possible,and a convolutional layer is proposed.The main deep learning model,combined with the random vector coding method described above and the improved convolutional neural network,is implemented in accordance with the characteristics of different data.The dimensionality reduction and simulation are carried out at the cost of the increase of data diversity.Upgrading operation of combined image convolutional network.After training using the Movielens dataset and the real dataset of the Heilongjiang media asset platform,the feature vectors of the user and the video are obtained and then put into the fully connected layer to fit the user's real score for the video.At the same time,in the specific implementation,the actual function used is selected based on the acceleration of the convergence speed and the reduction of the computational cost,which makes the model retraining less expensive,and can be well applied in today's network video update speed and large amount of data environment.