Chinese dependency relationship is complex and dependency span between the words is large especially in Chinese long sentences. Considering greed problems caused by Arc-eager algorithm for solving the long-distance dependencies in long sentences, this paper constructs a Root Node Finder. It can divide a long sentence into two short sentences. Using HIT Dependency Tree bank as a training test set, this paper uses Arc-eager algorithm and machine learning for dependency analysis of the whole sentence. The results show that the root accuracy of syntactic analysis is 77.25%. Then experiment uses LIBSVM as a binary classifier and adds adding different features for the Root Node Finder. At last, it identifies the optimal combination of features to impact the Root Node Finder. Results show that optimal features added, the root accuracy is 93.05%.