Emotion feature extraction method of BI-LSTM-CNN based on orthogonal projection

The invention discloses a BI-LSTM-CNN emotional feature extraction method based on orthogonal projection, and aims to obtain neutral word vectors with weights from a text, obtain emotional features with higher discrimination, and provide powerful technical support for text emotional classification and the like. A traditional deep learning model neglects special meaning words in key local context information, so that acquired emotion features are not rich enough. In order to solve the problem, the method provided by the invention comprises the following steps: projecting a neutral word vector into an orthogonal space of sentiment polar words to obtain a weighted neutral word vector, and extracting text key semantics through a CNN deep learning model; besides, a BI-LSTM-Attention model and a neutral word vector with a weight are utilized to learn semantic features capable of enhancing sentence sentiment from the extracted key semantics, so that the text is more discriminative during sentiment classification. 本发