TY - GEN
T1 - Multi-level Correlation Matching for Legal Text Similarity Modeling with Multiple Examples
AU - Huang, Ting
AU - Xie, Xike
AU - Liu, Xiufeng
PY - 2023
Y1 - 2023
N2 - Legal artificial intelligence (LegalAI) is an emerging field that leverages AI technology to enhance legal services. Similar Case Matching (SCM), which calculates the relevance between a candidate and a target case, is a critical technique in LegalAI to enable diverse legal intelligences. Existing approaches mainly rely the on single query texts or specific keywords for retrieval, yet neglected the domain complexity and multi-faceted nature of queries. Thus, a multi-example matching paradigm is motivated where three inherent challenges reveal. 1) Relevance assessment across multiple examples is complex. 2) The inherent lengthy and structured property of legal documents. 3) Lacking datasets containing golden labels for multi-example-based legal text matching. To address these challenges, this paper develops a novel multi-example dataset, and a Multi-level Correlation Semantic Matching (MCSM) is devised to extract similarity between cases given multi-example inputs. The proposed multi-level scheme can be interpreted as two aspects. Firstly, we consider both content and structure correlations to evaluate the relevance. Secondly, by dividing legal documents into distinctive segments, we can hierarchically learn the intra- and inter-segment dependencies to model the long-term dependencies across components of legal documents. An attention mechanism is employed to capture the complex interconnections among these examples and enable an attentive matching aggregation of content and structure. With multiple examples, the MCSM tackles the intricate and diverse nature of legal queries, providing a comprehensive and multi-dimensional description view. Extensive experimental evaluations show that the proposed MCSM outperforms baseline methods.
AB - Legal artificial intelligence (LegalAI) is an emerging field that leverages AI technology to enhance legal services. Similar Case Matching (SCM), which calculates the relevance between a candidate and a target case, is a critical technique in LegalAI to enable diverse legal intelligences. Existing approaches mainly rely the on single query texts or specific keywords for retrieval, yet neglected the domain complexity and multi-faceted nature of queries. Thus, a multi-example matching paradigm is motivated where three inherent challenges reveal. 1) Relevance assessment across multiple examples is complex. 2) The inherent lengthy and structured property of legal documents. 3) Lacking datasets containing golden labels for multi-example-based legal text matching. To address these challenges, this paper develops a novel multi-example dataset, and a Multi-level Correlation Semantic Matching (MCSM) is devised to extract similarity between cases given multi-example inputs. The proposed multi-level scheme can be interpreted as two aspects. Firstly, we consider both content and structure correlations to evaluate the relevance. Secondly, by dividing legal documents into distinctive segments, we can hierarchically learn the intra- and inter-segment dependencies to model the long-term dependencies across components of legal documents. An attention mechanism is employed to capture the complex interconnections among these examples and enable an attentive matching aggregation of content and structure. With multiple examples, the MCSM tackles the intricate and diverse nature of legal queries, providing a comprehensive and multi-dimensional description view. Extensive experimental evaluations show that the proposed MCSM outperforms baseline methods.
KW - Semantic Matching
KW - Legal Artificial Intelligence
KW - Natural Language Processing
U2 - 10.1007/978-981-99-7254-8_48
DO - 10.1007/978-981-99-7254-8_48
M3 - Article in proceedings
T3 - Lecture Notes in Computer Science
SP - 621
EP - 632
BT - Proceedings of the Web Information Systems Engineering – WISE 2023
ER -