Modeling law search as prediction

• DOI: 10.1007/s10506-020-09261-5
• Publication Date: 2020-02-01T06:02:35Z
• Authors: Faraz Dadgostari; Mauricio Guim; Peter A. Beling; Michael A. Livermore; Daniel N. Rockmore
• Keywords: Benchmark (surveying); Semantic Search

Abstract Law search is fundamental to legal reasoning and its articulation an important challenge open problem in the ongoing efforts investigate as a formal process. This Article formulates mathematical model that frames behavioral cognitive framework of law sequential decision The has two components: first, corpus space second, process (or strategy ) compatible with environment. structure “multi-network”—an interleaved distinct networks—developed earlier work. In this Article, we develop formally describe three related models We then implement these on subset U.S. Supreme Court opinions assess their performance against benchmark prediction tasks. first predict citations document from semantic content. second results generated by human users. For both benchmarks, all outperform null learning-based outperforming other approaches. Our indicate through additional work refinement, there may be potential for machine achieve or near-human levels performance.