Knowledge acquisition has been considered as a bottleneck problem in the development of intelligent systems for many years. Especially in the Internet era, a large number of information exists in the form of unstructured text. This paper introduces a knowledge acquisition system for a large Web page corpus based on distributed computing. This system is designed for automatic information mining and acquisition by the weakly supervised learning method. Computers can realize the automatic learning and mining of knowledge, the discovery of new words dictionary, the extraction of entity relation template, the entity recognition and so on. We represent the Ngram model and pairwise mutual information methods for new words recognition and location name entity detection, and the experimental results show the precision are 72.1% and 87.28% respectively.