Dynamic semantic feature fusion-based comment text sentiment analysis method, system and equipment and medium

• Main Authors: MIAO QIGUANG, LI PEIYUAN, KWON EUI-YOUNG, JIANG TAO, SUN PENGGANG, SONG JIANFENG
• Format: Patent
• Language: Chinese; English
• Published: 11.08.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

A comment text sentiment analysis method, system, device and medium based on dynamic semantic feature fusion, the method uses a BERT pre-training model to supplement the sentiment semantic representation ability of traditional word embedding, fully digs sentiment semantic information contained in a text in a task preprocessing stage through fine tuning of the BERT model, and performs sentiment semantic analysis on the text. Dynamically identifying the words, and representing semantics of polysemy words and new words; in addition, a bidirectional gate loop network structure (Bi-GRU) is fused with a self-attention mechanism to obtain a probability value corresponding to each emotion category, and the category with the maximum probability value is selected as a final output result of the BERT-BiGRU model; the system, the equipment and the medium are used for analyzing the emotion of a comment text when a computer program is executed by a processor; the method has the advantages that the emotion polarity containe