A two-phase bootstrap approach to facilitate conversion from text to semantic graphs

Abstract This paper proposes a two-phase unlearnable approach for converting a text into its semantic graph, addressing the challenge of efficiently generating a fully connected representation. Existing text-to-graph conversion tools struggle with processing long or complex texts, often resulting in fragmented representations. To overcome this, the proposed approach consists of two phases. In the first phase, divide and conquer, the input text is divided into small pieces manageable by the available text-to-graph conversion tool (e.g., Senna), yielding a collection of small subgraphs where each represents a corresponding piece of the input text. In the second phase, focus attention, these subgraphs are appended together using a bootstrapped algorithm to construct a strongly connected single graph that represents the entire input text. In these two phases, both SRL and RDF are considered and thoroughly explained. Accordingly, the corresponding algorithms for divide and conquer and focus attention are bootstrapped, evaluated, and compared. The implementation demonstrates that this approach enhances the accuracy and usability of text-to-graph conversion while remaining simple, fast, straightforward, and practical. This makes it particularly useful for NLP applications such as sentiment analysis and knowledge graph construction.