Using word embedding for environmental violation analysis：Evidence from Pennsylvania unconventional oil and gas compliance reports.

With the booming of the unconventional oil and gas (UOG) industry since the "shale gas revolution" around 2010, its inevitable damage to the environment and human health has aroused people's vigilance. We dig into the UOG compliance reports from 2000 to 2019 launched by the Department of Environmental Protection (DEP) in Pennsylvania, USA, to discover the relationship between the Contaminate , Location, and Operation in environmental violations. Specifically, we conduct a detailed statistical analysis; after narrative text preprocessing, the chosen keywords are transformed to embeddings via the Skip-Gram; then the cosine similarity analysis is utilized to reveal the mechanism of violations. The findings are: (1) sewage is the biggest source of pollution, the leakage mostly took place in the storage tank, wastewater management equipment and trucks; (2) oil pollutants mostly happened in vehicles and storage equipment during the transportation stage; (3) methane leakage mostly took placed in abandoned wells; (4) among the location keywords, the most correlative contaminants are oil, sewage, and silt, which mostly took placed in truck and pipeline during transportation and wastewater management stages; (5) illegal methane emission mostly happened during drilling and exhaust phase, indicating that governor and practitioners should be highly vigilant on the path of water pollution and soil erosion, the current monitoring systems was far from enough. • We applied natural language processing to environmental violations analysis. • We construct a novel framework of environmental violation analysis. • The regular law among key factors of violation is obtained based on word embedding. • We provide environmental supervision strategies for the government and practitioners. [ABSTRACT FROM AUTHOR]