Dr Hui Yang is the author of 'A Hybrid Model for Automatic Emotion Recognition in Suicide Notes', recently published in the Biomedical Informatics Insights Supplement. We asked Dr Yang to tell us about the background of this paper.
To start please tell us about the challenge this year. Why did you decide to become involved, and what goals did you and/or your team expect to accomplish?
The 2011 i2b2/VA/Cincinnati Challenge had two different shared tasks: one was conference resolution in clinical documents, and the other was emotion detection in suicide notes. Our team took part in the second shared task – emotion identification. The reason we participated in this shared task was because sentiment analysis has recently become an interesting research topic in computational linguistics. This challenge provides a valuable research opportunity to look deeply into the possible practical application of sentiment analysis in the real world. This was our first time being involved in this research topic, and we wished the challenge task could help us to better understand what kind of research issues there are in this area, and how hard it is to accurately recognize various emotions.
In writing this paper what were the particular challenges you faced? How did you overcome these challenges?
Since this paper is a challenge paper, we intend to introduce some experience and lessons learned during the system development so that the readers could better understand what kind of problems are presented in emotion analysis, and why some of them are hard to be solved. We wished the reader could further the research of sentiment analysis on the basis of our current work. How to select the particular research issues that we think may be benefit or interest the reader was somewhat difficult. We made use of the audience’s feedback about our presentation on the challenge workshop and compared our work to the work of other participant teams in order to help the preparation of our paper. Our paper focuses on the research issues that either the audience was most interested in or other teams did not address in their work, but was still considered important.
What has been the major benefit for you in the work discussed in your article? How has it contributed to our knowledge of the area?
In this challenge, we explored different methods to solve the problems in emotion identification and attempted to seek effective approaches specific to emotion detection in suicide notes. What we learned from this challenge is: (1) Machine learning approaches have been proven as a more effective method than other approaches (e.g., keyword-based or rule-based approaches), especially when enough training instances are available. (2) The expressions of different emotions vary widely in the text, thus a single language model is unable to deal with the variation of different emotion expressions and multiple language models are required to improve system performance.
As many of the articles appearing in the supplement are quick to acknowledge, suicide is a distressingly common cause of death particularly among younger people. Has this work changed your view of suicide: do you find yourself more or less understanding or sympathetic of people who commit suicide and those they leave behind?
The most touching thing for this challenge were the 160 volunteer annotators who contributed to the annotation of about 1,300 notes. All of these volunteers had survived the loss of a love one to suicide. Their biggest wish is to help other people avoid experiencing a similar loss and pain that they felt. I am very grateful to these volunteer annotators who help curate such a valuable emotion corpus for the research. I also wish that our research helps to build a system that could accurately predict the emotion status of the patient and help prevent the possible intention of suicide so that people do not suffer from the pain of loss of a loved one any longer.