Qian Yang: Identifying and Connecting Salient Information via Semantic Text Representations

Today Qian Yang and her advisor Gerard de Melo visited Penn State University and gave us the opportunity to get to know Qian's work before she defends her thesis! 🎓

Qian Yang is defending her doctoral thesis at Tsinghua University at the end of May.

Gerard de Melo recently moved to Rutgers University from Tsinghua University.

The staggering amounts of data available to us online can often appear overwhelming and lead to information overload. To solve this problem, one approach is to create human-readable summaries of longer texts and obtain compressed versions that still provide useful information to the user. Another direction is to extract machine-readable semantic representations of a text, which enable downstream information management applications and systems, especially better search interfaces to find relevant information. In this talk, we will present three works towards these goals.

The first one is PEAK, a automated evaluation metric for textual summaries based on whether they capture salient information. The second one is HiText, a novel method to present salient information in a text to a reader. The last one is a new method to connect semantic triple representations so as to make them more easily searchable and to enable better information management applications.

After the presentation we had a great lunch where David Reitter & Lee Giles along with their students joined us!

You can find more about Qian's work on her webpage.
For more information about Gerard de Melo's work & students visit his webpage.