Back to search

University of Central Florida

Recurrent Neural Networks Methods For Social Media Challenges: Forum Summarization And Rumor Detection

Abstract

dc:description.abstract

The overabundance of data on social media has posed several challenges to users. First, information overload becomes a barrier hindering the users to arrive at a conclusion. Summary of lengthy content can thus significantly facilitate them to grasp key ideas; not only does it help save time and energy, but it also contributes to an effective decision-making process. Second, social media platforms also encourage fast information dissemination where each user acts as a social sensor that generates and shares content. Yet without sufficient supervision, such rapid information sharing can lead to widespread rumors and fake news. Inspired by those challenges, this study proposes recurrent neural network-based frameworks to address them. For the first focus, Forum Summarization, our study presents summarization models adapted from the hierarchical attention networks (HAN) to build representations to predict summary sentences. Our findings demonstrated that the proposed frameworks significantly improved the classification performance as evaluated by sentence-level scores and the summary quality as evaluated by ROUGE scores. For the second focus, Rumor Detection, we present an ensemble deep neural network framework to classify input microblogging events, to whether they are valid or contain rumorous information. We proposed that, in addition to texts from microblog posts, the context related to users is also key to achieving performance improvement. The context information is obtained beforehand from each user's historical posts shared or composed about the event. What's more, to address the shortcomings of CNN, we present a deep end-to-end neural architecture that leverages capsule networks together with a hierarchical structure of an event to learn effective representations for rumor detection. Results from extensive experiments conducted on two real-world datasets, Twitter and Weibo, show that the proposed approach can accurately detect events that carry misinformation, outweighing a range of competitive baselines.

Degree

thesis:*
Grantor dc:publisher
University of Central Florida

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Tarnpradab, Sansiri
Contributors dc:contributor
  • Hua, Kien

Subjects

dc:subject × 3

Rights

Language dc:language
English

Identifiers

dc:identifier.*
Identifier
CFE0009270; DP0026874

Chain of custody

source
Harvested from
Central Florida
Base URL
stars.library.ucf.edu/do/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Tarnpradab, Sansiri. Recurrent Neural Networks Methods For Social Media Challenges: Forum Summarization And Rumor Detection. University of Central Florida, https://stars.library.ucf.edu/etd2020/1299