Visualizing Dynamics of the Hot Topics Using Sequence-Based Self-organizing Maps

摘要

We are currently working on a SOM-based method for temporal analysis and visualization of "hot topic" trends in news articles. Hot topics are extracted from a document collection by applying PCA to term frequency bag-of-words vectors. Evaluative experiments on three data sets, the largest expands across ten years, show that SBSOM induces a sequential analysis and that the use of label confidence mitigates the performance loss.