A News Event Clustering Algorithm Based on Semantic Relationship Graph

Abstract

Compared with text topic clustering, the granularity of news event clustering is finer. The complexity of the semantic relationship in news texts and their informational redundancy can cause great inconveniences for clustering. Therefore, the representation of events as well as the method for cluster division has become a major focus of researchers. To better tackle the problem, this paper proposes NEC_SRG, a news event clustering algorithm based on the semantic relationship graph. First, the semantic units related to the event topic are extracted from the news. Then, the semantic relationship graph is established based on the connection of words in each semantic unit to represent the event. After that, the cluster of semantic relationship graph is created based on the sub-graph closeness. Finally, the news events are clustered according to the distribution of the semantic units in the graph clustering results. Experiments show that the NEC_SRG algorithm has obvious advantages over similar algorithms.