This paper first discusses the application of text mining techniques in public security, with the research focus on case classification, feature extraction, and collusion analysis in civil litigation. A plain Bayesian-based first-level classifier is proposed to accurately classify first-level case texts. For all second-level case category document sets, keyword cooccurrence mapping is used to replace word frequency as a metric for feature selection, and a second-level classifier is constructed using a simple vector distance algorithm. Finally, the impact of time-barred benefit abandonment and time-barred benefit loss on the evolutionary system in civil litigation is investigated in depth using the pilot water tax reform in Hunan Province as a case study. The results show that when one of the subjects chooses to discard the statute of limitations benefits, x and y show a trend of steadily increasing convergence to 1 from the initial point, and the final stabilization point tends to (1,1). This paper aims to provide references and suggestions for reforming information and optimizing civil litigation.