A digital human multi-dimensional semantic sentiment analysis system and method
By employing a multi-dimensional semantic sentiment analysis method and a BERT-CSAM hybrid model architecture, the problems of insufficient granularity and high annotation costs in sentiment analysis during digital human interaction are solved, achieving efficient and accurate sentiment analysis and interactive feedback, thereby improving the emotional expression and interactive experience of digital humans.
Patent Information
- Authority / Receiving Office
- CN ยท China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SUZHOU GOLD MANTIS EXHIBITION DESIGN ENG
- Filing Date
- 2026-01-15
- Publication Date
- 2026-05-29
AI Technical Summary
Existing sentiment analysis technologies suffer from insufficient analytical granularity, difficulty in decoupling semantics and sentiment in digital human interaction scenarios, and high costs in constructing high-quality labeled datasets, making it difficult to achieve multi-dimensional, fine-grained sentiment analysis and efficient data construction.
We employ a multi-dimensional semantic sentiment analysis approach, construct a training dataset through a semi-supervised annotation process, train the model using the BERT-CSAM hybrid model architecture, and combine it with an adaptive learning rate optimization algorithm to achieve multi-dimensional sentiment feature extraction and accurate interpretation, thereby reducing computational resource consumption and improving model efficiency.
It enables multi-dimensional quantification and precise interpretation of textual sentiment, improves the accuracy and subtlety of sentiment analysis, enhances the interactive affinity and immersion of digital humans, and reduces annotation costs and computational resource consumption.
Smart Images

Figure CN122113937A_ABST