Raw Data for Text Generation Metrics from LLM-Fine-Tuned Psychology Questionnaires

A 12.0 KB Excel file containing raw data for evaluating text generation metrics from a study that fine-tuned Qwen-2.5 and GLM-4 models on a corpus of 169 professional psychological questionnaires. The research, authored by Zhitao Yuan and uploaded to figshare in April 2026, aimed to overcome bias in traditional scale development using instruction fine-tuning and human feedback reinforcement. The optimized models demonstrated gains in BLEU-4, ROUGE-L, logical consistency, and cultural adaptability.

Use Cases

• Benchmarking text generation models based on reported BLEU-4 and ROUGE-L metrics
• Analyzing improvements in logical consistency for psychological assessment tools based on the 28.6% gain mentioned
• Studying cultural adaptability in AI-generated text based on the over 85% cross-regional expression conversion accuracy
• Replicating or extending the fine-tuning methodology for professional questionnaire development described in the study

Strengths

• Data is derived from a corpus of 169 professional psychological questionnaires
• Model improvements are quantified with specific metrics: BLEU-4 increased by 0.05, ROUGE-L by 0.057
• Logical consistency improved by 28.6% and cultural adaptability achieved over 85% accuracy
• Released under a permissive CC-BY-4.0 license

Limitations

• Row count and column-level documentation are unknown; field semantics must be inferred after download
• The dataset is very small at 12.0 KB, indicating limited scope or raw metric outputs rather than full corpora
• Freshness should be verified as the last update date is in the future (2026-04 -24)

Provenance

Source

figshare, uploaded by author Zhitao Yuan

Collection Method

Likely contains raw evaluation metrics from a study fine-tuning LLMs on psychological questionnaires

Time Range

null

Freshness

Last updated 2026-04-24 17:29:18; freshness should be verified.

Geography

null