Research on entrepreneurial propensity identification methods using a fusion CNN-BiLSTM model

Intelligent algorithms in the field of artificial intelligence demonstrate significant potential in complex behavioral recognition tasks. This study constructs an entrepreneurial tendency recognition model integrating CNN and BiLSTM structures, combining local spatial feature extraction with bidirectional temporal modeling capabilities to enhance representation of behavioral evolution processes. The model inputs standardized high-dimensional behavioral sequences, leveraging collaborative modeling through convolutional and LSTM modules, with outputs weighted and fused via an attention mechanism. Experiments conducted on 2,150 university entrepreneurial behavior records demonstrate superior performance compared to GRU, LSTM, and Transformer models, achieving an accuracy of 93.85% and an F1 score of 0.932. This validates the effectiveness and robustness of the proposed fusion strategy for entrepreneurial behavior recognition.