Generative Comprehension and Storyline Construction from Narrative Text: an Artificial Intelligence Approach Based on Cognitive Psychology

Submission note: A thesis submitted in total fulfilment of the requirements for the degree of Doctor of Philosophy to the Research Centre for Data Analytics and Cognition, La Trobe Business School, College of Arts, Social Sciences and Commerce, La Trobe University, Victoria, Australia.

This research aims to advance event-based knowledge ecosystems in modern digital environments, by addressing the challenges of narrative comprehension and storyline construction from heterogeneous streams of text, using a generative approach founded on the theories of cognitive psychology, artificial intelligence and natural language processing techniques. Related research in this space exclusively focuses on extracting knowledge of factual relationships of entities; thereby, the understanding of the narrative remains static, segmented and isolated from context. As a result, extracted information does not support the discovery and representation of the evolutionary nature of associated events described within a narrative. Furthermore, the ideological and socio-cultural aspects of event knowledge are inadequately studied, leading to poorly formed representations of the subjectivity of individual and collective perspectives. Integrating traditional knowledge bases with collective intelligence from highly dynamic collaborative online environments is a further challenge in this space. The proposed generative approach is grounded on the cognitive and linguistic theories of Situation Model, Event-Indexing Model, Context Model and Integration Construction Model, as it extends these theories and interlinks an amalgamation for narrative comprehension that is representative of the multi-dimensional nature of human understanding. Situation models frame how humans extract, integrate and present textual and verbal situational information by encoding text into mental relational structures, and then use it for reasoning and construction of a set of coherent beliefs. The theoretical underpinnings of situational models are further integrated to adapt to the fast-changing nature of social media platforms, denoted as the Global Online Conversation Environment. This consolidated view contains both the situational context of events and the dynamic time-based social views. This consolidation is advanced into a comprehensive conceptional and computational model, referred to as the Cognitive based Event Graph Model (CEGM), that implements a holistic approach of leveraging artificial intelligence to simulate the mental processes and representations in the human mind during narrative comprehension. It establishes the processes of event knowledge construction while retaining the trail of memory processing from shallow and detailed level to a highly abstract level. Based on CEGM, an innovative approach is proposed to automatically construct storylines from the models and discovered insights. This approach will identify, disambiguate, accumulate, and enhance factors indicative of events from streams of text and present a comprehensive model of events with the situational and social awareness. The constructed event knowledge base is designed to support the evolving nature of events embedded in large bodies of narrative text data, and corresponding decision-making processes with situation and social awareness.