November 3 Math Colloquium

Michael W. Berry, University of Tennessee-Knoxville

Toward Unsupervised Learning for Social Media Using Linear Algebra

In large-scale text mining applications such as tweet classification there is need for fast yet robust techniques to summarize or track concepts without prior knowledge of the content. Linear algebra plays a very important role in the design and implementation of the underlying algorithms needed for the automated summarization of time-sensitive documents, especially those from social media. Matrix and tensor factorization methods can greatly facilitate the the extraction of key documents (tweets) that can summarize a current stream and thereby reduce the exhaustive human effort that would be needed to read and synthesize an enormous number of documents. The long term goal of this research is to develop the core numerical algorithms and software needed for unsupervised learning when no prior labels or metadata is available.

Social Media