This paper implements a film review sentiment analysis algorithm based on BiLSTM. BiLSTM is a bidirectional long-short term memory network. It can effectively extract context dependencies by analyzing time series data such as text and video.Film review analysis is a sub field of text processing, which is a typical time series data processing problem. The BiLSTM uses tworeverse LSTM networks to provide additional context information for the model. In order to prevent over fitting, dropout mechanismis also adopted in the network. This algorithm is implemented based on Keras platform. Through experimental comparison, the lastone is the best among the three algorithms: LSTM, BiLSTM and BiLSTM combined with dropout. The algorithm design provides aresearch basis for the broad application prospect of BiLSTM.