Principled Approaches to Automatic Text Summarization

In this talk, we will discuss approaches to tackle  Automatic Text Summarization. We'll concentrate on content selection, the inherent problem of summarization which is controlled by the notion of information Importance. To this end, we'll introduce a simple and intuitive formulation of summarization as two components: a summary scoring function indicating how good is a text as a summary of the given sources, and an optimization technique extracting a high-scoring summary. We will briefly discuss how one can empirically "learn" the summary scoring function from data yielding new summarization systems and new evaluation metrics. Then, alternatively, we will take a more theoretical strategy and formalize the vague notion of Importance. Intuitively, Importance can be seen as the measure that guides which choices to make when information must be discarded.

Maxime Peyrard is a Ph.D. student at the University of Darmstadt, working on machine learning and natural language processing.
His research has focused on unifying existing summarization approaches into a generic optimization problem. He was also interested in ways to leverage the available human judgments to improve summarization systems and evaluation metrics.
Maxime comes from France and he received a joint Master's degree in Computer Science from the University of Darmstadt and Grenoble INP Ensimag.
He then moved to the United Kingdom for one year to work on the Alexa project at Amazon Cambridge. For his Ph.D., he joined the AIPHES research training group from TU Darmstadt focused on Automatic Summarization.