Journal Paper

#52 

@article{Park2017, title={ConceptVector: Text Visual Analytics via Interactive Lexicon Building using Word Embedding}, author={Deok Gun Park and Seungyeon Kim and Jurim Lee and Jaegul Choo and Nicholas Diakopoulos and Niklas Elmqvist}, url={http://www.umiacs.umd.edu/~elm/projects/conceptvector/conceptvector.pdf}, year={2017}, date={2017-10-01}, journal={IEEE Transactions on Visualization & Computer Graphics}, abstract={Central to many text analysis methods is the notion of a concept: a set of semantically related keywords characterizing a specific object, phenomenon, or theme. Advances in word embedding allow building such concepts from a small set of seed terms. However, naive application of such techniques may result in false positive errors because of the polysemy of human language. To mitigate this problem, we present a visual analytics system called ConceptVector that guides the user in building such concepts and then using them to analyze documents. Document-analysis case studies with real-world datasets demonstrate the fine-grained analysis provided by ConceptVector. To support the elaborate modeling of concepts using user seed terms, we introduce a bipolar concept model and support for irrelevant words. We validate the interactive lexicon building interface via a user study and expert reviews. The quantitative evaluation shows that the bipolar lexicon generated with our methods is comparable to human-generated ones.}, keywords={}, }