Back to feed
arXiv cs.CL
arXiv cs.CL
7/31/2026
HSS-Synth: Humanities and Social Sciences Data Synthesis for LLMs

HSS-Synth: Humanities and Social Sciences Data Synthesis for LLMs

Short summary

HSS-Synth is the first data synthesis pipeline for humanities and social sciences, covering 14 fields with a subject-centric paradigm. It constructs seed documents from web corpora, backtranslates them into diverse instructions with Q&A alignment checks, and uses teacher-forced answering to anchor semantics and reduce hallucinations. The pipeline yields 237K instruction-tuning samples; fine-tuned Qwen3-8B-Base sets a new SOTA across 16 benchmarks, approaching the official Qwen3-8B. Code is publicly available.

  • •First HSS data synthesis pipeline covering 14 fields with subject-centric paradigm
  • •237K instruction-tuning samples; Qwen3-8B-Base sets SOTA on 16 benchmarks
  • •Teacher-forced answering reduces hallucinations and preserves tone; code is open-source

Generated with AI, which can make mistakes.

#research-breakthrough#open-source
Read full article at arXiv cs.CL

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

BridgeAlign: Bridging Preference Alignment for Humanities and Social Sciences
arXiv cs.CL
The original title is "DataPrep-Bench: Benchmarking LLMs as Training Data Preparators"
arXiv cs.LG
Beyond the Hivemind: Escaping LLM Homogeneity via Meta-Persona Anchoring and Sequential Temperature Scaling
arXiv CS.AI
Before the Action: Benchmarking LLMs on Prospective Hypothesis Discovery
arXiv cs.CL