Topological data analysis for fake news detection

This project aims to contribute to the under-researched field of topological data analysis (TDA) for text classification through the task of fake news detection. For this task, three individual models have been used: least absolute shrinkage and selection operator (LASSO) using 0th Dimensional Persi...

Full description

Saved in:
Bibliographic Details
Main Author: Deng, Ran
Other Authors: Fedor Duzhin
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2022
Subjects:
Online Access:https://hdl.handle.net/10356/156821
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:This project aims to contribute to the under-researched field of topological data analysis (TDA) for text classification through the task of fake news detection. For this task, three individual models have been used: least absolute shrinkage and selection operator (LASSO) using 0th Dimensional Persistent Image (PI) vectors, Bidirectional Long Short-Term Memory (BiLSTM), and Bidirectional Encoder Representations from Transformers (BERT). Two ensemble models were also used to improve performances by supplementing contextual information from deep-learning models with structural information from PI vectors: BiLSTM + TDA and BERT + TDA. The results suggest that when structural information is given equal or lesser influence than contextual information, the ensemble performs better than the base models on average. This project offers a possible way of utilising TDA features to improve performances in text classification tasks, and a comparison between different models for organizations concerned with false information detection in general.