Dr. Maria Liakata's EBI text-mining team are the authors of 'Three Hybrid Classifiers for the Detection of Emotions in Suicide Notes', recently published in the Biomedical Informatics Insights Supplement.  We asked the EBI text-mining team 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?

We found the challenge on emotion detection in suicide notes particularly interesting. Not only was the topic controversial and dealing with a very sensitive and important social phenomenon, but the data itself, the suicide notes, was a unique collection of documents annotated with a complex scheme by multiple non-expert annotators. Our goal was to find out which emotions were prevalent in suicide notes and help make suggestions about how to improve the annotation scheme. We liked the idea that by classifying and analyzing these texts we could learn more about the human condition.

In writing this paper what were the particular challenges you faced?  How did you overcome these challenges?

Dealing with data sparseness for the majority of the emotion categories as well as inconsistencies in the annotations was definitely the biggest challenge. We tackled these issues by suggesting a combination of human written rules and different machine learning classifiers, which took a different view to the task (classification vs. sequence labelling). We wanted to explore if the sequence of categories mattered (i.e. do people in a suicidal state tend to express emotions in a particular order, at which point do they include concrete information etc.) and which type of classifier was most suited to the task.

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?  

Thinking about the complexities in identifying emotions in suicide notes has been the main benefit from working on this topic. Our main contributions involve trying different perspectives, considering multi-label annotation, single label, taking sequence of labels into account vs. text classification. Rather than focussing on feature engineering, we experimented with combinations of different classifiers and came to the conclusion that a combination of separate classifiers for each emotion would be the way to go. More work on how to best combine this is still required. We also expressed our thoughts about the annotation scheme, the need for explicit annotation guidelines as well as the need to distinguish between emotion and non-emotion categories (such as information and instructions). We also discuss how the task could become more focussed by looking at the most salient emotions such as hopelessness.

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?  

We found it very sad that people committed suicide over what often seemed mundane or solvable problems. It was definitely a life-affirming experience for us, which left us with the desire to prevent such wasteful loss of life. If people felt connected and had some emotional support at the right time we believe that many suicides could have been prevented.

share on