A Temporal Scope Prediction for Storyline Generation Using Events’ Knowledge Graphs

Abstract

Knowledge graphs (KGs) have become powerful tools for organizing and making sense of complex datasets, especially when time information is included. However, ensuring the accuracy of events in temporal knowledge graphs remains a challenge. To address this issue, researchers proposed a new approach that uses event storyline generation and visualization to analyze trends and situations. The proposed approach involves conducting experiments using traditional embedding techniques and a transformer-based BERT base model to generate temporal graph embeddings. These embeddings result in lower-dimensional representations that are easier to process and input into factorization machines, leading to improved event classification accuracy. The approach was tested using event-based datasets such as ICEWS and Wikidata12k, achieving an accuracy of 80% when compared to the baseline model. This approach shows promise for analyzing trends and situations, with potential applications in industries that require planning, such as disaster planning and cyber-physical systems. By using event storyline generation and visualization, the proposed approach can facilitate downstream applications, such as trend and situation analysis, and improve the accuracy of events in temporal knowledge graphs. This research highlights the significance of knowledge graphs in managing and analyzing vast amounts of data and emphasizes the importance of developing accurate and efficient strategies for decision-making.