Poster Presentation COSA-IPOS Joint Scientific Meeting 2012

Identifying emotion in text generated from online interventions for people who have survived cancer (#482)

Erin O'Carroll Bantum 1 , Jason E. Owen 2 , Noemie Elhadad 3 , Mitch Golant 4 , Joanne Buzaglo 5 , Janine Giese-Davis 6
  1. Cancer Prevention and Control, University of Hawai'i Cancer Center, Honolulu, HI, USA
  2. Psychology, Loma Linda University, Loma Linda, CA, USA
  3. Department of Biomedical Informatics, Columbia University, New York, NY, USA
  4. Research & Training, Cancer Support Community, Los Angeles, CA, USA
  5. Research & Training, Cancer Support Community, Philadelphia, PA, USA
  6. Psychology, University of Calgary, Calgary, Alberta, Canada

INTRODUCTION:  Emotional expression is a potential mechanism of action by which psychosocial interventions act to improve cancer-related distress and quality of life.  However, most extant research on emotional expression has been limited to self-report.  Giese-Davis (Specific Affect (SPAFF) for Breast Cancer; 2005) has developed an extensive coding system, based on the work of John Gottman, for identifying emotional expression and other interactions between cancer survivors in the context of supportive-expressive group therapy.  Given the dramatic increase in the use of text-based exchanges between cancer survivors and the use of Internet-based interventions, the current study seeks to extend this work by developing a coding system for identifying and classifying emotional expression in transcripts of online cancer support groups. 

METHODS:  The 24 codes in the original SPAFF system (which include nonverbal codes) were revised to include text identifiers.  All 24 codes are used in the current study, and outcome codes used in the training process were analyzed for frequency of each code. 

RESULTS:  Preliminary data were available from 856 assigned emotion codes.  The most frequent emotions expressed were prosocial emotions (i.e., interest, validation, affection), which represented 59.5% of all expressed emotions.  Codes for excitement/joy/delight and humor represented 9.3% and 9.2% respectively, followed by tense humor (4.4%), sadness (3.0%), tension (2.7%), frustration (2.2%), and anger (2.0%).  Excluding prosocial emotions, expressed emotions were slightly more positive (56.8%) than negative (43.2%). 

DISCUSSION:  The eventual goal of this process is to create a computer program that can effectively identify emotional expression in text.  A first step in this process is to manually code emotional expression in a reliable fashion, also allowing us to identify the use of emotion in different methods of communication.  Finding ways to objectively measure mechanisms of action, instead of relying on self-report measures is crucial to providing tailored and potent psychosocial interventions.