Sentiment analysis of English social media based on transfer learning and semantic enhancement

Social media sentiment analysis faces challenges such as scarce annotated data and complex contextual factors, making it difficult for traditional methods to accurately capture users' genuine emotions.This paper proposes an innovative framework integrating transfer learning and semantic enhancement.By pre-training models on massive unlabeled data and combining them with sophisticated contextual semantic enhancement techniques, it enhances the ability to recognise nuanced emotional expressions in English social media texts.Experiments on public datasets like semantic evaluation-2018 convincingly demonstrate that our method achieves an accuracy of 73.12%, representing a 15.32% improvement over traditional approaches.Statistical tests confirm this enhancement is statistically significant (p < 0.01).This research provides more reliable technical support for practical applications such as real-time public opinion monitoring and personalised recommendation systems.