Abstract: Fine grained emotion analysis analyzes the author's emotional tendency from the perspective of the evaluation object and its attributes through the text. Its main tasks include the recognition of the evaluation object and its attributes (topic recognition) and emotion recognition. To solve the problems of poor fine grained emotion recognition and poor interpretability of deep learning methods in previous studies, a fine-grained emotion analysis model integrating ontology and deep learning is proposed. The model uses domain ontology and CNN fusion methods to identify explicit and implicit topics, and combines emotion dictionary and Bi LSTM+Attention model to identify fine-grained emotions of online comment texts. The experimental results show that the proposed fine-grained sentiment analysis method has advantages over other methods in accuracy, recall and F value.

Key words: Deep learning, Ontology, Fine-grained emotion analysis, Online comments