<p>With the continuous growth of global air traffic, the effective analysis of aviation accidents has become increasingly important. However, existing aviation accident analysis systems face several limitations, such as low accuracy in cause identification when processing complex accident reports, poor performance in recognizing rare categories of aviation accidents, and a lack of efficient automated report generation mechanisms. Therefore, to address these issues, this study develops a comprehensive analysis system aimed at improving the effectiveness and efficiency of aviation accident analysis. The system consists of two core modules: AviClassificator and AviAgent. AviClassificator integrates a fine-tuned TinyBERT word embedder and uses a BiLSTM-attention classifier to classify the causes of aviation accidents. To improve the recognition of rare categories in aviation accidents, this study applies a pre-trained language model (Pre-trained LM) for data augmentation during the training data preprocessing stage. AviAgent is a customized generation system based on guided strategies, capable of automatically generating standardized aviation accident risk analysis reports. Experimental results show that AviClassificator performs well across evaluation metrics such as classification accuracy, macro precision, and Hamming loss, while also demonstrating advantages in identifying rare accident categories, indicating strong classification performance. Meanwhile, AviAgent generates standardized reports to support the efficiency of accident analysis. This study provides a practical framework for aviation accident analysis by improving cause classification and supporting standardized report generation.</p>