Top-N Recommendation Algorithm Based On Autoencoder Representation Learning

In today’s information age,a large amount of data will be generated every day,which brings great difficulties for users to find the required content.In order to help users find the content they need from a large amount of data,personalized recommendation system came into being.The core of the algorithm is to analyze the user’s historical behavior,predict the user’s preference,and then mine the content in line with the user’s preference from the data and recommend it to the user.However,due to the limitations of the principle of the algorithm,the traditional recommendation algorithms inevitably have the problems of data sparsity and cold start.In recent years,the emergence of deep learning has pointed out a new direction for the research of recommendation system.However,the existing recommendation algorithms based on deep learning often default that the user preference is unchanged,and there is no modeling for the timing of user scoring sequence,which makes it difficult to extract the relevant information between features in the process of feature extraction.In addition,the existing methods do not consider the problems of data missing and errors in the real scene,so that the model recommendation effect is poor and lacks robustness.To solve the above problems,this text proposes two top-N recommendation algorithms based on autoencoder representation learning.In the first place,a top-N recommendation algorithm based on self-attention autoencoder and LSTM is proposed to solve the problems that the existing recommendation algorithms ignore the dynamic changes of user preferences and don’t build model aiming at times series data,resulting in insufficient extraction of internal correlation of input features.Based on the self-attention mechanism,the algorithm designs the coding module and calculates the correlation of internal features in the time series.At the same time,it stores the user’s preference change information in different periods and forms the feature coding.Then,the effectiveness of LSTM processing timing information is used to construct the prediction module,analyzing the implicit information in the feature code.Later,predict the user’s scoring and ranking of items,and make the top-N recommendation list.Furthermore,a top-N recommendation algorithm based on stacked denoising autoencoder and siamese neural network is proposed to solve the problems of scoring error,information loss and error caused by user’s improper operation in real scenarios.In order to simulate the problem of noise in real data,the recommended method first adds noise data to the original input data,and then inputs the data with noise data into the first layer autoencoder.Later,the second layer autoencoder is used to learn feature coding.The learned feature codes are input into the prediction module combined with LSTM are Siamese neural network to predict user scores and sort the scores to determine the top-N recommendation list.In the end,experiments are carried out on Movie Lens 1M data set and Movie Lens100 k data set to verify the effectiveness of the proposed algorithm.