With the popularity of short video platforms, recommending personalized short videos for users has become a challenging task. Among them, graph neural network (GNN)based recommendation algorithms have attracted much attention due to their ability to capture the complex relationships between users and items. However, how to better utilize the historical behavioral data of users and how to optimize GNN algorithms are still hot research topics. To this end, this paper proposes a Graph Convolutional Network (GCN) model that incorporates Bidirectional Long Short-Term Memory (Bi-LSTM) and Multi-Head Attention (MHA), referred to as BiL-MA-GCN. BiL-MA-GCN model combines the user's viewing history and the textual information of short videos, and uses GCN networks to describe the relationship between short videos and users to generate a personalized recommendation list. Meanwhile, this paper conducts experiments using three publicly available short video datasets Kuaishou, TikTok and MovieLens. The experimental results prove that the BiLMA-GCN algorithm can effectively improve the accuracy of recommendations.