Sheng Bi

Existing fine-tuning data selection methods for large language models (LLMs) often struggle to balance two aspects: the evaluation of legal fine-tuning data quality and the maintenance of syntactic diversity. The assessment of data quality consists of two dimensions: legal content quality, encompassing logical rigor, legal basis traceability, fact coverage, and actionable guidance, and instructional complexity, which measures how challenging an instruction–response pair is for the model to learn. Syntactic diversity, in contrast, captures the variety of sentence structures and dependency patterns within the dataset, reflecting how sentences differ in grammatical organization, for example through variations between active and passive voice, clause nesting, or modifier placement. To address these issues, this paper proposes LEAD (Legal Efficiency and Diversity), a fine-tuning data selection framework for legal LLMs that integrates dual-objective optimization with syntactic structure clustering. LEAD consists of two modules: the Legal Efficiency Training Score (LETS), which evaluates each sample’s legal quality, and Syntactic Diversity Clustering (SDC), which applies graph-kernel methods to ensure syntactic diversity. Experiments show that using only 5% of the original dataset selected by LEAD, the resulting model outperforms full-data training by 14.9% and achieves an average 12.3% improvement across legal benchmarks, demonstrating strong potential for efficient legal LLM fine-tuning.

Legal Judgment Prediction (LJP) is an essential component of legal assistant systems, which aims to automatically predict judgment results from a given criminal fact description. As a vital subtask of LJP, researchers have paid little attention to the numerical LJP, i.e., the prediction of imprisonment and penalty. Existing methods ignore numerical information in the criminal facts, making their performances far from satisfactory. For instance, the amount of theft varies, as do the prison terms and penalties. The major challenge is how the model can obtain the ability of numerical comparison and magnitude perception, e.g., 400 < 500 < 800, 500 is closer to 400 than to 800. To this end, we propose a judicial knowledge-enhanced magnitude-aware reasoning architecture, called NumLJP, for the numerical LJP task. Specifically, we first implement a contrastive learning-based judicial knowledge selector to distinguish confusing criminal cases efficiently. Unlike previous approaches that employ the law article as external knowledge, judicial knowledge is a quantitative guideline in real scenarios. It contains many numerals (called anchors) that can construct a reference frame. Then we design a masked numeral prediction task to help the model remember these anchors to acquire legal numerical commonsense from the selected judicial knowledge. We construct a scale-based numerical graph using the anchors and numerals in facts to perform magnitude-aware numerical reasoning. Finally, the representations of fact description, judicial knowledge, and numerals are fused to make decisions. We conduct extensive experiments on three real-world datasets and select several competitive baselines. The results demonstrate that the macro-F1 of NumLJP improves by at least 9.53% and 11.57% on the prediction of penalty and imprisonment, respectively.

Purpose: Sensory feedback for prosthetics is an important issue. The area of forearm stump skin that has evoked tactile sensation of fingers is defined as the projected finger map, and the area close to the PFM region that does not have ETS is defined as the non-projected finger map. Previous studies have confirmed that ETS can restore the tactile pathway of the lost finger, which was induced by stimulation of transcutaneous electrical nerve stimulation on the end of stump skin. This study aims to reveal EEG features between the PFM and the NPFM regions of the stumps under the same TENS stimulation condition.Methods: The PFM and NPFM regions of the two subjects were stimulated with the same intensity of TENS, respectively. TENS as target stimuli are modulated according to the Oddball paradigm to evoke the P300 components.Result: The PFM regions of both subjects were able to elicit P300 components, while their NPFM regions were not able to elicit P300 components. However, this P300 appears early and has continuous positive peaks in front of it.Discussion: N30 and P300 can prove that the two subjects with PFM can perceive and recognize ETS. The heteromorphisms of the P300 waveform may be related to the difficulty in subjects’ cognition of ETS or caused by the fusion of P150, P200, and P300.