Digital tools continue to redefine much of modern student life and learning. Educational administrators could better serve their student communities if they had a clearer view of the emotions and opinions those students are expressing online. Research in the International Journal of Information and Communication Technology, describes a deep learning-based method to analyse and categorize student sentiment in online content. The tools could offer invaluable insights for managing campus dynamics and enhancing the academic environment.
Dan Wang and Li Wang of the Gingko College of Hospitality Management in Chengdu, China, explain how deep learning techniques can be though of a subset of artificial intelligence (AI) technologies with a focus on understanding human language. By analysing content from different online platforms, such as social media, discussion forums, and website comment sections, the team suggests that it is possible to extract a clearer picture of the emotional and ideological landscape of student population.
The approach uses Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. CNNs can identify patterns and extract key features from textual data, while LSTMs are used to understand the relationships between words in long passages of text. By combining the strengths of these tools, it is possible to extract the nuance of ideas and emotions being shared online in the wider student discourse.
A key aspect of the new analytical model is the introduction of an “attention mechanism”. This improves the model’s ability to accurately interpret complex emotional expressions. In online communication, students often use irony, sarcasm, or metaphor to convey sentiments, as do we all. This is difficult to grasp with a simple analytical tool. The attention mechanism allows the system to focus on the most critical words or phrases in a given piece of text and this improves its ability to detect and decode these subtle emotional cues. For instance, the phrase “yeah, right” is familiar American vernacular and is commonly used as a sarcastic riposte to an apparently unbelievable comment. Taken literally, however, it would simply be interpreted as confirmation of the person reading the unbelievable and confirming their acceptance of it.
In addition to the nuances of the model and the AI tools on which it is built, the team has also created a large-scale, annotated dataset of student-generated content. This dataset, drawn from a wide range of platforms, allowed the team to train and validate their model with real data. The same data and model might be used off campus too, to analyse public online sentiment or perhaps within the corporate environment.
Wang, D. and Wang, L. (2024) ‘Deep learning semantic understanding and classification of student online public opinion for new media’, Int. J. Information and Communication Technology, Vol. 25, No. 10, pp.62–76.