The recommendation system infers user interests by collecting their historical behavior, thereby generating a list of recommendations. The goal of a video recommendation system is to fulfill user needs by recommending videos that align with their interests. In such a system, the relevant input data consists of the user's historical video-watching behavior sequence, where the order of videos in the sequence represents the chronological relationship of user video consumption. The output data is a probabilistic model that calculates the likelihood of a given user watching a specific video in the next viewing session. Once this model is obtained, probabilities can be computed for all candidate videos, and the recommendation list can be generated by sorting these probabilities in descending order. Here, we propose a video recommendation model based on graph convolutional networks. This recommendation model constructs a heterogeneous graph that includes user and video nodes and behavior edges. It builds an embedding layer to obtain representation vectors for users and videos, designs vectors to characterize users' long-term and short-term interests, and introduces a prediction layer to estimate the probability of a user watching a video.