Hybrid BiLSTM-CapsNet for dynamic assessment of entrepreneurship ability in business plans

Proper assessment of business plan entrepreneurial potential is instrumental in facilitating innovation and stimulating economic growth. However, accurate assessment of unstructured text data within these plans using conventional techniques is extremely difficult. In an effort to transcend these limitations, the current research introduces a novel two-stage deep learning-based approach that incorporates the latest techniques to release more accurate analysis. The methodology designed involves two fundamental steps: first, text feature description and extraction, and second, intelligent classification. In the first step, the Global Vectors for Word Representation (GloVe) approach is used to transform the text into machine-understandable numerical vectors. Thereafter, in the second step, these representations from GloVe are modeled and analyzed through a hybrid Bidirectional Long Short-Term Memory-Capsule Neural Network (BiLSTM-CapsNet) model. This hybrid deep learning model, leveraging in turn the advantage of BiLSTM’s sequential modeling and hierarchical modeling abilities of CapsNet along with part-to-whole relationships, enables deep text processing, and the prediction is made on the basis of CapsNet output. The proposed BiLSTM-CapsNet approach was evaluated using a large text dataset, and the results show exceptional performance of the model; its accuracy and F-measure reach 95.70% and 0.9568, respectively. This performance confirms the high capability of the BiLSTM-CapsNet approach in detecting complex patterns and extracting key information from text and proves its improved classification capabilities.