At present, some progress has been made in the research of context-based recommendation algorithm and label-based recommendation algorithm. However, there are some problems such as sparse scoring data for items by users and low precision of recommendation results. In response to the above problems, this paper proposes a recommendation algorithm that integrates time context and tag optimization. The recommendation algorithm is improved by integrating user behavior time interval and user attribute label information. Firstly, Long Short-Term Memory (LSTM) is introduced to study the effect of time interval on tags. Then, each output layer is combined with Latent Dirichlet Allocation (LDA) to weigh the tags with high importance. Finally, the prediction value is obtained by fusing the scoring information. Experiments show that the new algorithm has effectively alleviated the problem of sparse scoring data and improves the precision of recommendation results.