An Unsupervised Keywords Extraction Approach for Chinese Government Documents

In China, government documents play an important role in making formal regulation, government policy, etc. The number of government official documents is increasing by hundreds of thousands every year. The huge quantity and diversified content make it difficult to accurately search and classify government documents. With accurate keywords of documents, the accuracy of searching and classification can be improved. Generally, the keywords extraction methods can be classified into two categories: supervised learning and unsupervised learning. Due to the lack of labelling data, this paper focuses on unsupervised methods. Based on the characteristics of official documents, this paper proposes a method for automatic keyword extraction based on a method which is a weighted sum of TF-IDF and TextRank. This method can dynamically adjust the weighting coefficients according to the genre and writing features of official documents so that it could take advantages of each of the keyword extraction method. Experimental results show that this method is superior to the traditional unsupervised keyword automatic extraction method.