Non-semantic type social network extraction method for play

The invention provides a non-semantic type social network extraction method for a play. According to the method, a semantic technical route which is better in effect and higher in complexity on a small data set is abandoned, stage instructions closely related to role dialogues are selected under a lightweight and easily-extended frame, and then a direct relation among roles is found from the side. Generally speaking, compared with a more precise natural language processing type method, a role network extracted through the method can be sparser; guidance, a training set, a testing set, parameter optimizing and calibrating and the high time and space cost are not needed, and meanwhile the good extendibility and ready-to-use property are achieved. 本发明提供了种面向剧本的非语义式社交网络抽取方法。本发明放弃了小数据集上效果更好、复杂度更高的语义类技术路线，在轻量、易扩展框架下，选取了与角色对白有紧密联系的舞台指示，从侧面来发现角色间的直接关联。般来讲，与更精准的自然语言处理式方法相比，此举所抽取出的角色网络会更稀疏；但他无需指导，无需训练集与测试集，无需参数调优与校准，无需高昂的时空代价，同时还具备优良的可扩展性、即时可用的特性。