The construction of “new liberal arts” and the advent of the era of artificial intelligence are forcing the traditional training mode of legal talents to change in the direction of interdisciplinary, high-quality composite talents who know how to govern society, among which, the legal evidence-based talents are the talents training direction adapted to the future needs and have a noble sense of social responsibility. At present, the cultivation of evidence-based talents is still in the exploratory stage, and in order to use the specific cultivation method, we need to cooperate with high-quality enterprises and practical departments. Practical exploration reveals that, on the one hand, utilizing the technical resources of enterprises can effectively enhance students’ ability to use evidence-based research methods. On the other hand, in cooperation with practical departments, conducting research on the rule of law around real issues and evaluating the implementation effect of policies can well exercise students’ practical ability and deepen their patriotic sentiment of helping people in the world. In addition, this paper also carries out the vectorized representation of legal text based on the N-Gram model, calculates the similarity of legal cases by combining with Word2Vec, and introduces the graph representation learning to establish a similar case law recommendation model, which helps students to find out the similar cases in the massive cases. It is conducive to students’ cultivation of systematic thinking in evidence-based jurisprudence and enhancement of their ability to discriminate similar cases. Through the practice of the course, the initial feasible experience of “five-stage parallel progress” and “three integration of science and reality” on the cultivation of evidence-based law talents has been formed, and the incentives and assessment mechanisms for the cultivation of industry-academia integration need to be further strengthened in the future.