Developing methods for behavior scenario analysis (on the material of instructional texts)

Abstract

The article discusses the modeling of behavioral scenarios and methods for their extraction from texts published in social media. The authors proceed from the basic (in modern cognitive science) position that the intentional aspect of a person’s behavior is reflected in his speech activity. This gives reason to extract scenarios from the texts generated to carry out some activity. The choice of instructive texts is determined by the fact that their genre, content and communicative intentions of their authors most explicitly represent the program of human behavior. In this article we propose several methods for automatic construction of behavior scenarios, as well as a scenario model in the form of a frame. This model can be used in the system for automatic extraction of scenarios from texts and their further implementation. We describe an experiment on automatic scenario identification in texts and its results. The experiment includes text segmentation after which each segment is associated with a particular step of a scenario. Text segmentation involves dividing a text into semantic units, selecting the most important of them, which express (by their lexical meaning and form) incentive modality, and finding some details to these units. In psychological terms, this means that we extract those segments that contain description of some goals and particular actions that lead to their achievement. Reconstruction of the scenario in this case can be represented as automatic extraction of goals, actors, actions and conditions from selected segments. The result of text segmentation achieves a fairly high F-measure.