Investigating Community Detection Algorithms for Meeting Summarization
We extend a fully unsupervised, abstractive meeting summarization framework to use novel clustering methods. We investigate the application of the Word Mover's Distance and variants of it, as well as various clustering methods such as agglomerative clustering, spectral clustering, and $k$-means applied to data generated using multidimensional scaling. Our embedding-based distance approach encorporates exterior knowledge into the clustering stage of the framework.
natural language processing; Computer science; NLP; abstractive; meeting; summarization; unsupervised
Cardie, Claire T.
M.S., Computer Science
Master of Science
dissertation or thesis