Using Natural Language Processing of Web Logs to Identify Signs of Distress Among In-Treatment Adolescent Cancer Patients, 07-R9737
Printer Friendly VersionPrincipal Investigators
Brett G. Knight
Robert Van Dam
Leanne Segovia
Doretta E. Gordon
Inclusive Dates: 07/01/07 – 01/30/09
Background - Quickly and accurately identifying distress levels of adolescents with cancer is critical to providing timely care. Adolescents with cancer often use writing and journals as a means of expressing their feelings and emotions relative to dealing with the disease and its treatment. The goal of this project was to determine if it is possible to identify accurately the levels of distress of adolescents in cancer treatment by parsing writing samples using natural language processing (NLP) techniques. This study tested this hypothesis by comparing NLP-generated distress level ratings with those identified by trained psychologists.
Approach - The study included an expert analysis session to identify distress diagnosis indicators. These indicators were then used to develop NLP logic to be used to parse the BLOGs (web-based logs) to identify distress signs. These BLOGs were then run through the NLP software and distress indicators were automatically flagged. Licensed psychologists manually coded the same BLOGs based on the agreed distress diagnosis indicators. The two coding results were compared, and the findings were analyzed to generate a final report on the methodology.
Accomplishments - Using supervised NLP techniques to identify stress in adolescent cancer patients looks promising given larger sample sizes. Additionally, the SwRI-owned NLP software tool suite has been greatly enhanced and should prove useful towards future work in this area. Other notable accomplishments are positive feedback from test subjects, and improvements to the most promising NLP method.
