Objective:

To construct and validate a Chinese cyberbullying victimization lexicon using large language models to improve identification and facilitate intervention.

Approach:

• Lexicon Construction: Developed from social media data, focusing on cyberbullying methods, perceived harm, and coping strategies, with the assistance of large language models.
• Validation Methodology: Evaluated validity through correlations between word-frequency statistics from Weibo posts and expert ratings, utilizing large language models for analysis.
• Model Comparison: Compared outputs of DeepSeek-R1 and GPT-4o with human evaluations in lexicon development tasks, assessing their effectiveness in vocabulary selection and weight assignment.

Key Findings:

• The lexicon includes 442 words across three dimensions: cyberbullying methods, perceived harm, and coping strategies.
• Strong validity in identifying cyberbullying victimization expressions across dimensions (cyberbullying methods: r=0.500, p < 0.001; perceived harm: r=0.408, p < 0.001; coping strategies: r=0.509, p < 0.001; overall r = 0.870, p < 0.001).
• DeepSeek-R1 showed good performance in small-scale text classification (Kappa = 0.775–0.781) but limitations in large-scale processing.

Interpretation:

The study indicates that large language models can assist in structured tasks but require human oversight for complex research phases.

Limitations:

• Significant discrepancies between model outputs and human evaluations in vocabulary selection and weight assignment tasks.
• Limited effectiveness of models in large-scale processing, with Kappa values indicating substantial inconsistencies.

Conclusion:

The research highlights the potential of a human-machine collaborative approach for optimal outcomes in cyberbullying victimization identification.