Year: 2023
Type: conference-paper
Abstract: In the ever-evolving landscape of data visualization and extraction, deep learning techniques have become increasingly pivotal.Addressing this, our paper conducts a sensitivity analysis of Named Entity Extraction on text related to the infamous Panama Papers.We combine the Rebel and Natural Language Processing models to extract entities and relations.To visualize the extracted knowledge, we employ the NetworkX library to transform this data into intuitive graphs.These are then compared to a predefined expected graph to assess the influence of various parameters during the creation process.To ensure an accurate comparison, the graphs are transformed into vectors using the Graph2Vec method after undergoing pre-processing tasks, such as the removal of self-loops, isolated nodes, and node renaming.The similarity with our benchmark graph is determined using Euclidean distance metrics.Results highlight the influence of span length, length penalty, and the number of beams on graph generation.Notably, span length emerges as the most impactful factor, in determining graph detail and complexity.
DOI: 10.26493/scores23.12
Authors: Lea Roj, Aleksander Pur, Štefan Kohek, Niko Lukač
Institutions: University of Maribor, Ministry of the Interior
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4387341981