The paper introduces a novel method called MPGK-LMI for predicting lncRNA–miRNA interactions in animals. This method combines a graph attention network (GAT) with meta-path similarity and Gaussian kernel similarity to enhance the prediction accuracy. The key steps include constructing a meta-path similarity matrix and a Gaussian kernel similarity matrix, which are then integrated with feature matrices derived from doc2vec to update the feature representation. The GAT model captures neighborhood information and integrates different similarity measures to improve the model's performance. The proposed method outperforms existing state-of-the-art algorithms in terms of AUC, AUPR, ACC, F1-score, and precision. The effectiveness of MPGK-LMI is further validated through ablation experiments and a case study focusing on lncRNAs MALAT1 and HOTAIR, demonstrating its ability to accurately predict known interactions and provide insights into their functional roles.The paper introduces a novel method called MPGK-LMI for predicting lncRNA–miRNA interactions in animals. This method combines a graph attention network (GAT) with meta-path similarity and Gaussian kernel similarity to enhance the prediction accuracy. The key steps include constructing a meta-path similarity matrix and a Gaussian kernel similarity matrix, which are then integrated with feature matrices derived from doc2vec to update the feature representation. The GAT model captures neighborhood information and integrates different similarity measures to improve the model's performance. The proposed method outperforms existing state-of-the-art algorithms in terms of AUC, AUPR, ACC, F1-score, and precision. The effectiveness of MPGK-LMI is further validated through ablation experiments and a case study focusing on lncRNAs MALAT1 and HOTAIR, demonstrating its ability to accurately predict known interactions and provide insights into their functional roles.