A Study on Structured Text Parsing for Policies Based on BERTopic

This research delves into a structured text parsing methodology rooted in financial policy documents. The primary objective is to offer a novel interpretative approach, leveraging the application of the four constituent elements of textual analysis. We deploy BERTopic, a topic model, and GPT-3.5, a generative model, to autonomously generate topics and elucidate writing purposes within the textual content. Furthermore, we employ positional and grammatical rules for the extraction of pertinent information pertaining to entities, document formats, and institutional references. This enables the construction of a structured descriptive framework grounded in the four constituent elements. Through an empirical analysis of the interpretation of financial policy texts, our research has established that this framework effectively represents policy components in an intuitive manner, while efficiently revealing the distribution of thematic elements within policy documents. Additionally, this framework lends itself well to the batch analysis and interpretation of extensive policy-related texts. By approaching the study from the perspective of applying the four constituent elements of textual analysis, our work elucidates the key themes, entities, writing objectives, as well as format and structural characteristics within financial policy documents, facilitating rapid comprehension of policy-related information by the public. Currently, we are engaged in further research to harness parsed structured data for subsequent analysis.