Detection and analysis of graduate students' academic emotions in the online academic forum based on text mining with a deep learning approach

Abstract

Purpose: The possibility of mental illness caused by the academic emotions and academic pressure of graduate students has received widespread attention. Discovering hidden academic emotions by mining graduate students' speeches in social networks has strong practical significance for the mental state discovery of graduate students.

Design/methodology/approach: Through data collected from online academic forum, a text based BiGRU-Attention model was conducted to achieve academic emotion recognition and classification, and a keyword statistics and topic analysis was performed for topic discussion among graduate posts.

Findings: Female graduate students post more than male students, and graduates majoring in chemistry post the most. Using the BiGRU-Attention model to identify and classify academic emotions has a performance with precision, recall and F1 score of more than 95%, the category of PA (Positive Activating) has the best classification performance. Through the analysis of post topics and keywords, the academic emotions of graduates mainly come from academic pressure, interpersonal relationships and career related.

Originality: A BiGRU-Attention model based on deep learning method is proposed to combine classical academic emotion classification and categories to achieve a text academic emotion recognition method based on user generated content.

Keywords: academic emotion; deep learning; emotion classification; emotion recognition; graduate mental health.