Constructing Narrative Event Evolutionary Graph for Script Event Prediction (1805.05081v2)

Abstract: Script event prediction requires a model to predict the subsequent event given an existing event context. Previous models based on event pairs or event chains cannot make full use of dense event connections, which may limit their capability of event prediction. To remedy this, we propose constructing an event graph to better utilize the event network information for script event prediction. In particular, we first extract narrative event chains from large quantities of news corpus, and then construct a narrative event evolutionary graph (NEEG) based on the extracted chains. NEEG can be seen as a knowledge base that describes event evolutionary principles and patterns. To solve the inference problem on NEEG, we present a scaled graph neural network (SGNN) to model event interactions and learn better event representations. Instead of computing the representations on the whole graph, SGNN processes only the concerned nodes each time, which makes our model feasible to large-scale graphs. By comparing the similarity between input context event representations and candidate event representations, we can choose the most reasonable subsequent event. Experimental results on widely used New York Times corpus demonstrate that our model significantly outperforms state-of-the-art baseline methods, by using standard multiple choice narrative cloze evaluation.

Constructing Narrative Event Evolutionary Graph for Script Event Prediction

This paper introduces a novel approach to script event prediction by leveraging the inherent structure of narrative event graphs, termed Narrative Event Evolutionary Graphs (NEEGs). Traditional methodologies in script event prediction, such as those based on event pairs or chains, rely heavily on limited contextual information that may impede their predictive accuracy. By constructing NEEGs, the authors present a method that captures dense connections among events, thereby offering an enriched framework for prediction tasks.

The central premise of the paper is the creation and utilization of an event graph structure that incorporates narrative event chains extracted from extensive news corpora, specifically from the New York Times corpus. The NEEG operates as a knowledge base encapsulating the evolutionary patterns and principles of events, thereby offering a more comprehensive approach to understanding and predicting subsequent script events.

To enable inference within this graph model, the authors propose a Scaled Graph Neural Network (SGNN) that dynamically interacts with subgraphs containing context and candidate events, ensuring scalability even with large graph structures. The SGNN chooses subsequent events based on learned event representations and their underlying connectivity within the NEEG, derived from network embedding principles.

Empirical results substantiate the efficacy of this approach, as evidenced by superior performance metrics when compared to state-of-the-art baseline methods, utilizing a multiple-choice narrative cloze evaluation on the New York Times corpus. The SGNN model achieves an accuracy of 52.45% in predicting script events, which is a notable improvement over previous models such as PairLSTM and EventComp.

The implementation of NEEGs and SGNN offers several compelling implications for AI applications, ranging from discourse understanding to intention recognition and dialogue generation. By facilitating a more nuanced comprehension of event contexts through rich graph-based structures, these methods enhance the predictive capabilities and practical utility of AI models dealing with narrative scripts.

Looking ahead, the conceptual framework of NEEGs could spur further research into event prediction across varied domains, with potential integrations into broader AI systems that require narrative comprehension and event-based reasoning. Additionally, the scalable nature of SGNN opens avenues for working with even larger datasets, paving the way for new applications where dense temporal and causal relationships are critical.

Overall, the paper's contributions lie in advancing script event prediction through innovative graph constructions and neural network models, offering insights into the potential of graph-based methodologies in AI research and development contexts.