Plan Generation from Unstructured Documents Through Transformer-Based Extraction of Knowledge Graphs
Paper Number
2112
Paper Type
Complete Research Paper
Abstract
Planning for complex tasks is a key task for knowledge workers that is often time-consuming and depends on the manual extraction of knowledge from documents. In this research, we propose an end-to-end method, called PlanKG, that: (1) extracts knowledge graphs from full-text plan descriptions (FTPD); and (2) generates novel FTPD according to plan requirements and context information provided by users. From the knowledge graphs, activity sequences are obtained and projected into embedding spaces. We show that compressed activity sequences are sufficient for the search and generation of plan descriptions. The PlanKG method uses a pipeline consisting of decoder-only transformer models and encoder-only transformer models. To evaluate the PlanKG method, we conducted an experimental study for movie plot descriptions and compared our method with original FTPDs and FTPD summarizations. The results of this research has significant potential for enhancing efficiency and precision when searching and generating plans
Recommended Citation
Maass, Wolfgang; Agnes, Cicy K.; and Harig, Amin, "Plan Generation from Unstructured Documents Through Transformer-Based Extraction of Knowledge Graphs" (2024). ECIS 2024 Proceedings. 1.
https://aisel.aisnet.org/ecis2024/track03_ai/track03_ai/1
Plan Generation from Unstructured Documents Through Transformer-Based Extraction of Knowledge Graphs
Planning for complex tasks is a key task for knowledge workers that is often time-consuming and depends on the manual extraction of knowledge from documents. In this research, we propose an end-to-end method, called PlanKG, that: (1) extracts knowledge graphs from full-text plan descriptions (FTPD); and (2) generates novel FTPD according to plan requirements and context information provided by users. From the knowledge graphs, activity sequences are obtained and projected into embedding spaces. We show that compressed activity sequences are sufficient for the search and generation of plan descriptions. The PlanKG method uses a pipeline consisting of decoder-only transformer models and encoder-only transformer models. To evaluate the PlanKG method, we conducted an experimental study for movie plot descriptions and compared our method with original FTPDs and FTPD summarizations. The results of this research has significant potential for enhancing efficiency and precision when searching and generating plans
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.