基于因果事件抽取驱动关键法律要素感知的林法类案检索模型构建

Causal Event Extraction Driven Key Legal Element-aware Retrieval Model of Forestry Legal Case

摘要:

林法类案检索旨在找到与输入案例事实相似的历史林法判决案例，在林业智能法律系统中发挥着核心作用。现有的类案检索模型缺乏对法律案文特定结构内关键法律要素的考虑，无法准确利用关键法律要素蕴含的深层语义信息，导致在相似候选案例的检索场景中表现欠佳。在林业法律案文中，关键法律要素通常出现在以林木为主体的因果事件中，为此，提出一种因果事件抽取驱动关键法律要素感知的林法类案检索模型（Causal event extraction-driven key legal element-aware model,CEKLE），该模型在将法律案文拆分为“前言”、“事实”、“分析”、“判决”和“尾文”5部分基础上，重点关注林业法律案文的“事实”与“分析”2部分，结合因果事件抽取，获取相应因果事件，从而准确感知案例关键法律要素位置，充分挖掘关键法律语义信息，以提升林法类案检索准确性。2个数据集上的实验结果表明，在林法类案检索中CEKLE优于最先进的基线模型。

Abstract:

Forestry legal case retrieval aims to identify historical forestry legal judgment cases with facts similar to the input case, which plays a central role in intelligent forestry legal systems. Existing legal case retrieval models failed to adequately consider the key legal elements embedded in the specific structure of legal documents, thus hindering their ability to accurately use the deep semantic information contained in these key legal elements, ultimately leading to inferior performance when retrieving similar candidate cases. In forestry legal case documents, key legal elements usually appeared in various causal events with forest trees as the main body. Based on this, the causal event extraction-driven key legal element-aware model (CEKLE) was proposed, which was a forestry legal case retrieval model with awareness of key legal elements driven by causal event extraction. This model decomposed forestry legal document into five main sections: “Introduction”, “Facts”, “Analysis”, “Judgment”, and “Tail”. On this basis, it focused on the two parts of “Fact” and “Analysis”, by combining causal event extraction, the corresponding causal events can be obtained, so as to accurately perceive the position of the key legal elements of the legal case, fully excavate the key legal semantic information, and improve the accuracy of forestry legal retrieval. The experimental results obtained from two different datasets clearly demonstrated that CEKLE achieved a better performance than the most advanced baseline model in the task of forest legal case retrieval.