Network-Based Text Analysis to Support Narrative Exposure Therapy

Abstract

Narrative Exposure Therapy (NET) is a short-term therapy for Post-Traumatic Stress Disorder (PTSD) that works by the patient narrating his or her life story in chronological order with a special focus on the traumatic events, thereby allowing the patient to associate the intrusive recollections of PTSD with the temporal and situational context of the past traumatic events and thus to habituate and reduce the avoidance behavior associated with PTSD.This thesis attempts to support NET by automatically analyzing the resulting narration containing the patient’s life line with network-based text analysis methods. Therefore, a text network applicable to arbitrary texts is defined which attempts to incorporate more of the linguistic information extracted by modern Natural Language Processing systems, constructing the network based on the dependencies within each sentence. A set of domain-specific rules is then developed which automatically detects the items in the text that might be helpful for NET by extracting those that have a certain grammatical and lexical structure. Finally, the information from these two processes is combined in two different kinds of networks, the first of which attempts to find the relations between the detected items, while the second starts from the text network and attempts to find further important items with the help of the information gained from the detected items.These methods were implemented in Java and applied to anonymized narrations. To protect the privacy of the patients, the resulting networks could not be included in this thesis. The results are instead illustrated on a narration already publicly available.