Event coreference resolution using particle swarm optimisation

Due to the tremendous increase of documents in the web, people who need information are not ready to spend much time in reading the entire content of documents retrieved. Instead they need precise information. A kind of precise information obtained in our work is event corefering sentences. All sentences referring to the same event instance are called event corefering sentences. Our proposed approach formulates this event coreference resolution as a graph-based clustering model. It constructs the graph based on the sentences in the document with edge weights representing similarity score between each pair of sentences. To reduce the number of singleton clusters and to have a balanced cut, our approach combines minimum conductance with cut clustering to form clusters of corefering sentences. As finding minimum conductance is NP-hard, it uses particle swarm optimisation technique to obtain minimum conductance.