Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Interpretable Charge Prediction for Criminal Cases with Dynamic Rationale Attention
Authors: Wenhan Chao, Xin Jiang, Zhunchen Luo, Yakun Hu, Wenjia Ma
JAIR 2019 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that besides providing charge prediction interpretation, our approach can also capture subtle details to help charge prediction. ... We train and evaluate our model on real Chinese criminal cases by collecting legal documents from China Judgements Online. |
| Researcher Affiliation | Academia | Wenhan Chao EMAIL Xin Jiang EMAIL School of Computer Science and Engineering, Beihang University, Beijing, China Zhunchen Luo EMAIL Information Research Center of Military Science, PLA Academy of Military Science, Beijing, China Yakun Hu EMAIL Wenjia Ma EMAIL School of Computer Science and Engineering, Beihang University, Beijing, China |
| Pseudocode | No | The paper describes the model architecture and training process using mathematical formulas and descriptive text (e.g., "p(zt|xp, z<t) = sigmoid(W0[ ht; ht; zt 1] + b0)"), but does not contain a dedicated "Pseudocode" or "Algorithm" section, nor any structured code-like blocks. |
| Open Source Code | No | The paper states "We build and release a real dataset of Chinese criminal judgement documents, which can be used to study charge prediction and other related issues in AI&Law.", but does not provide any specific link or statement about releasing the source code for the methodology described. |
| Open Datasets | Yes | We build and release a real dataset of Chinese criminal judgement documents, which can be used to study charge prediction and other related issues in AI&Law. We construct a dataset from China Judgements Online, which contains a large number of documents on various cases throughout the country. Footnote 1: http://wenshu.court.gov.cn |
| Dataset Splits | Yes | 80,000, 10,000 and 10,000 identically distributed documents are randomly selected as training, validation and test set respectively. |
| Hardware Specification | No | The paper mentions using a "deep neural framework" and training with "Adam stochastic optimization method" and "dropout regularization layer" but does not specify any particular hardware like GPU or CPU models used for experiments. |
| Software Dependencies | No | We use Han LP to tokenize the Chinese texts. Core NLP (Manning, Surdeanu, Bauer, Finkel, Bethard, & Mc Closky, 2014) is used to parse the syntax tree... No version numbers are provided for these tools. |
| Experiment Setup | Yes | The hidden size is set to 200 for all GRUs. For CNN RATIONALE ENCODER, we set different ο¬lter heights of 3, 4, 5, with 128 feature maps each. 200 dimensional word embeddings are pre-trained... We choose a batch size of 64 and adopt Adam stochastic optimization method (Kingma & Ba, 2014) to learn the trainable parameters. We apply a dropout regularization layer... and L2 regularization to all trainable parameters. |