This study proposes a method called MPGK-LMI for predicting lncRNA–miRNA interactions in animals. The method integrates feature matrices derived from the doc2vec model, meta-path-based similarity matrices, and Gaussian kernel-based similarity matrices. It uses a graph attention network (GAT) to aggregate these matrices and update the feature matrix with neighborhood information. A scoring module is then used for prediction. MPGK-LMI outperforms three state-of-the-art algorithms in terms of performance, achieving AUC of 0.9077, AUPR of 0.9327, ACC of 0.9080, F1-score of 0.9143, and precision of 0.8739. Detailed case studies demonstrate the effectiveness and feasibility of the approach in practical applications. The method provides significant breakthroughs in understanding the functional roles and mechanisms of lncRNA–miRNA interactions. MPGK-LMI also establishes its practicality and reliability in biological research through real-case analysis, offering strong support and guidance for future studies and applications. The method leverages node neighborhood information to its fullest extent, which is crucial for accurate predictions. The study also evaluates the performance of MPGK-LMI under different parameters, including the number of graph attention layers, learning rates, and hidden layer nodes. The results show that MPGK-LMI achieves the best performance when using two graph attention layers, a learning rate of 0.0001, and 512 hidden layer nodes. The case studies on lncRNAs MALAT1 and HOTAIR demonstrate the predictive capacity of MPGK-LMI in identifying potential interactions between lncRNAs and miRNAs. The results provide insights into the intricate regulatory roles of lncRNA–miRNA interactions and contribute to the understanding of the underlying mechanisms and potential therapeutic targets associated with these interactions.This study proposes a method called MPGK-LMI for predicting lncRNA–miRNA interactions in animals. The method integrates feature matrices derived from the doc2vec model, meta-path-based similarity matrices, and Gaussian kernel-based similarity matrices. It uses a graph attention network (GAT) to aggregate these matrices and update the feature matrix with neighborhood information. A scoring module is then used for prediction. MPGK-LMI outperforms three state-of-the-art algorithms in terms of performance, achieving AUC of 0.9077, AUPR of 0.9327, ACC of 0.9080, F1-score of 0.9143, and precision of 0.8739. Detailed case studies demonstrate the effectiveness and feasibility of the approach in practical applications. The method provides significant breakthroughs in understanding the functional roles and mechanisms of lncRNA–miRNA interactions. MPGK-LMI also establishes its practicality and reliability in biological research through real-case analysis, offering strong support and guidance for future studies and applications. The method leverages node neighborhood information to its fullest extent, which is crucial for accurate predictions. The study also evaluates the performance of MPGK-LMI under different parameters, including the number of graph attention layers, learning rates, and hidden layer nodes. The results show that MPGK-LMI achieves the best performance when using two graph attention layers, a learning rate of 0.0001, and 512 hidden layer nodes. The case studies on lncRNAs MALAT1 and HOTAIR demonstrate the predictive capacity of MPGK-LMI in identifying potential interactions between lncRNAs and miRNAs. The results provide insights into the intricate regulatory roles of lncRNA–miRNA interactions and contribute to the understanding of the underlying mechanisms and potential therapeutic targets associated with these interactions.