SenticNet/personality-detection

The hierarchical architecture combines document-level Mairesse linguistic features with word2vec embeddings processed through convolutional layers, enabling both static and trainable embedding modes. It operates as a binary classifier for each of the five traits independently, with preprocessing that filters emotionally neutral sentences and generates feature vectors from essay text. The implementation uses Theano for computation and supports GoogleNews pre-trained word2vec or randomized embeddings, requiring annotated essay datasets paired with extracted Mairesse linguistic feature sets.

509 stars. No commits in the last 6 months.