NLP research at UBC

The Natural Language Processing (NLP) group at University of British Columbia conducts research in Computational Linguistics, Text Mining, Machine Learning, and Visual Text Analytics. The NLP@UBC group comprises of two senior faculty members, post-doctoral researchers, graduate students, and undergraduate research assistants. We focus on the following research areas in particular (but not limited to):

  • Neural Topic Segmentation

    Topic segmentation is a fundamental NLP task which is able to reveal the important aspects of a document semantic structure by splitting the document into topical-coherent textual units automatically. However, conventional neural topic segmenters are week at modeling context information. We have explored effective strategies to enhance neural model's capability of context modeling and further improve model's performance.
  • Neural Extractive Summarization

    We have explored neural extractive summarization models, with a focus on long documents. In addition, we have also worked on incorporating discourse parsing to neural extractive summarization.
  • Discourse Parsing

    We have developed a scalable method to automatically generate large discourse treebanks using distant supervision from sentiment-annotated datasets. More details here.
  • Mining and Summarizing Conversations (Emails, Meetings, Blogs, Chats)

    We have worked on several mining tasks, including topic segmentation and labeling, sentiment analysis, controversiality and extracting the conversational structure. We have also developed novel approaches to automatically generating extractive and abstractive summary of conversations.
  • Understanding, Generating and Summarizing Evaluative Text

    In this direction, we have been working on generating evaluative arguments tailored to a model of the user preferences, as well as on generating multimedia extractive and abstractive summaries of large sets of evaluative documents (e.g., customer reviews)
  • Visual Text Analytics for Conversations

    In this area, our aim is to tightly integrate interactive visualization with text mining and summarization techniques for information exploration and scalable decision support. We are focusing on creating visual text analytic systems for several domains including but not limited to meeting , emails, blogs, customer reviews, social media such as Facebook and twitter.


Dowload our new MEGA-DT Discourse Treebank!

ConVis: A Visual Text Analytic System

Pretty picture placeholderMeeting browser

Interactive Visualization of Opinions