Autocorrection of Noise Text Based on Modularity Optimization

This paper brings forward an autocorrection algorithm for noise texts based on modularity optimization. By noise texts we mean those documents in text corpus being distributed to a wrong category. Firstly, the document- similarity network is constructed, in which each node represents a document. If two nodes are similar in content, they are connected with a weighted edge, and their similarity is the weight. Secondly, the categories constitute the corresponding community structure in the network. Modularity has been introduced as a measure to evaluate the quality of community structures. In this paper modularity is used to evaluate the quality of categorise. Finally, noise texts are autocorrected by optimizing the modularity. The experimental results indicate that this algorithm can effectively revise the noise texts. This algorithm can also be used in the preprocessing of text classification or taxonomy building.