Clustering sentences to discover events from multiple news articles using Buckshot and Fractionation

Abstract

Sentence Clustering is performed based on the key terms in sentences within a document or group of documents. A sentence may come under different topics in a single document with different word of similar meaning which will not be clustered correctly by using hierarchical clustering methods. Hierarchical clustering methods are robust. They are not very efficient as its time complexity is O (n2). To overcome this problem, K-means type algorithms are used, but it handles only few documents. A proposed algorithm uses both hierarchical and partitional clustering method alternatively. It increases the accuracy and reduces the time complexity for multiple news articles. It is applied to group the text spans from multiple news articles that refer to the same event.