Research map / Large language models / Language, safety and interpretability
ZipLJP: Zipped Information Processor for Legal Judgment Prediction
Fanghao Lou, Qiqi Wang, Guanyu Chen, Kaiqi Zhao, Huijia Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add ZipLJP compression to Qwen2.5-3B fine-tuning | Qwen2.5-3B · MultiLJP charge | Accuracy: 88.7 -> 90.1 | Improved |
| add ZipLJP compression to LLaMA3.2-3B fine-tuning | LLaMA3.2-3B · CAIL2018 charge and law article | Accuracy: charge 77.4 -> 78.3; law article 74.3 -> 73.5 | Mixed |
| add ZipLJP compression to LLaMA3-8B fine-tuning | LLaMA3-8B · CAIL2018 prison term | Accuracy: 57.8 -> 55.1 | Hurt |
| add ZipLJP compression to LawyerLLaMA-13B-V2 fine-tuning | LawyerLLaMA-13B-V2 · CAIL2018 charge | Accuracy: 82.4 -> 86.4 | Improved |
| use BGE embedding model instead of Legal-BERT for sentence selection (ablation) | ZipLJP (Qwen2-7B) · MultiLJP law article | Accuracy: 79.7 -> 93.6 | Improved |
| use BGE embedding model instead of M3E (ablation) | ZipLJP (Qwen2-7B) · MultiLJP law article | Accuracy: 93.3 -> 93.6 | Improved |
| use ZipLJP compression instead of BigBird sparse attention (ablation) | Qwen2-7B · MultiLJP law article | Accuracy: 78.1 -> 93.6 | Improved |
| use ZipLJP compression instead of LLMLingua-2 (ablation) | Qwen2-7B · MultiLJP law article | Accuracy: 83.1 -> 93.6 | Improved |
| use ZipLJP compression instead of Qwen-Plus abstractive summary (ablation) | Qwen2-7B · MultiLJP prison term | Accuracy: 57.3 -> 64.5 | Improved |
| vary retained proportion k of sentences (below vs above 0.2) (ablation) | ZipLJP · MultiLJP | Accuracy and F1: k < 0.2 low performance; k > 0.2 similar high performance | Mixed |
| remove the most similar 20% or 60% of text before prediction (ablation) | ZipLJP · MultiLJP | Accuracy and F1 | Hurt |
| add ZipLJP to zero-shot DeepSeek | DeepSeek-V3.2-Exp · MultiLJP law article | Accuracy and Macro-F: Acc. 89.1 -> 91.1; Ma-F 78.7 -> 77.3 | Mixed |
Closest papers by meaning
- DELTA: Pre-Train a Discriminative Encoder for Legal Case Retrieval via Structural Word Alignment (AAAI 2025 · similarity 0.85)
- Lawma: The Power of Specialization for Legal Annotation (ICLR 2025 · similarity 0.85)
- LexEval: A Comprehensive Chinese Legal Benchmark for Evaluating Large Language Models (NeurIPS 2024 · similarity 0.84)
- J&H: Evaluating the Robustness of Large Language Models Under Knowledge-Injection Attacks in Legal Domain (AAAI 2025 · similarity 0.83)
- SILO Language Models: Isolating Legal Risk In a Nonparametric Datastore (ICLR 2024 · similarity 0.83)
- SaulLM-54B & SaulLM-141B: Scaling Up Domain Adaptation for the Legal Domain (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
