Computational Analysis of Semantic and Sentiment Features in Natural Language for Schizophrenia.

BACKGROUND: Schizophrenia (SZ) patients exhibit abnormalities in language expression, including semantic confusion and apathy, which significantly impair social functioning. The existing studies on speech in schizophrenia patients mainly focused on semantic or grammatical features, and the integration analysis between emotional dynamics and linguistic fluency remained scarce. The present study aims to systematically analyze abnormal emotional expression in schizophrenia patients by employing natural language processing (NLP) models. METHOD: This study recruited 78 patients with SZ and 85 healthy controls (HCs). All participants completed a free speech task, in which they described a positive event and a negative event happened recently, each for one minute. This study employs natural language processing models with metrics such as sentence similarity, emotional conditional entropy, and emotional transition matrices to systematically analyze emotional expression abnormalities in SZ patients during the task. RESULTS: SZ patients showed less task-congruent sentiment polarity: they failed to generate intense positive sentiments during the positive task and exhibited inappropriate positive emotions (e.g., intense positivity) during the negative task. They also had significantly lower mean sentiment scores and emotional conditional entropy, and higher neutral self-transition probability than HCs. These findings indicate distinct abnormalities in emotional coherence, intensity, and dynamics of SZ patients. CONCLUSION: This study presents a novel NLP-based computational strategy utilizing the analysis of linguistic patterns and emotional expression in SZ patients and found that the patients demonstrated distinct abnormalities across three emotional dimensions: coherence, expression intensity, and dynamics. The findings offer an innovative approach to screen schizophrenic disorders.