|Accueil | Aperçu | Revues | Soumettre | Nous Contacter | English|
In this special issue of Biomedical Informatics Insights we present the results of a shared task dedicated to finding emotions in suicide notes with machine learning tools. Shared tasks are not new, but conducting this type of sentiment analysis with this amount of data is.
A total of 1278 notes that were written by people just prior to dying by suicide were annotated by 160 vested volunteers. Each note was read by three different volunteers and then annotated based on an emotional schema that included: abuse, anger, blame, fear, guilt, hopelessness, sorrow, forgiveness, happiness, peacefulness, hopefulness, love, pride, thankfulness, instructions, and information. These annotated notes formed the corpus required by the machine learning methods.
Twenty four teams agreed to analyze these data and then submit a manuscript for review. The systems with the highest precision and recall were submitted by: Open University in Milton Keynes UK, Microsoft Research Asia in Beijing, P.R. China and Mayo Clinic in Rochester NY USA. Each of these groups received a travel stipend provided by Diamond Healthcare, Richmond VA, USA.
Each manuscript was blindly reviewed by three reviewers whose results formed the decision to publish. This is somewhat of a different review process for Biomedical Informatics Insight because we used the participants to blindly review each others manuscripts rather than calling upon the pool of reviewers. The best paper was A Hybrid Model for Automatic Emotion Recognition in Suicide Notes by Hui Yang, Alistair Willis, Anne de Roeck and Bashar Nuseibeh of Open University.
The full articles along with all the articles can be found in Biomedical Informatics Insights.
A shared task of this magnitude does not happen by chance. Rather, it is the tenacity of the steering committee, the vested volunteers, and the staff who made this important activity occur. From it we have learned a great deal about sentiment analysis and the limitations of the data. I invite you to read the articles about this shared task and I encourage you to learn as much as we have.