Detecting and Mitigating the Spread of Low-Quality Information in Social Media


Social media (e.g. Facebook, Twitter, Youtube and Instagram) is exploding at an unprecedented speed and is gradually becoming an indispensable part of billions of people. The inherent openness, ease of access, immediacy and timeliness of social media compared with conventional news organizations (e.g. television and newspapers) give people a twilight zone to voice their opinions, share information and consume content on an explosive scale. However, many people are taking advantage of the positive benefits of social media to generate and spread low-quality information (e.g. fake news, biased and partisan stories, hate speeches and self-harmed content), causing negative effects on our society. In this dissertation, we propose three directions to combat the spread of low-quality information in social media. We firstly aim to build a model to detect groups of online users who coordinate with each other to retweet/share low-quality information. Based on the model, security systems are able to stop the spread of poor-quality information. Secondly, we propose to utilize online fact-checkers to fact-check information in social media. It has been shown that social media posts about fake news are four more times to be deleted when being fact-checked by online fact-checkers. Finally, as many innocent online users are not aware of credible information when they are exposed to fake news, we propose to search and incorporate fact-checked information into social media posts to improve their awareness of fake news, leading to the reduction in consumption of misinformation. Based on the three directions, this dissertation makes the following contributions: · The first contribution of this dissertation is a proposed framework to extract groups of users based on their retweeting/sharing behavior and to classify extracted groups into malicious user groups. To deal with large-scale datasets from social media, we also propose a novel distributed algorithm based on Hadoop MapReduce framework to speedup extracting user groups. · Second, we thoroughly analyze online fact-checkers based on their temporal behavior, topical interest and linguistics signals of their fact-checking tweets. Based on the analysis, we propose two novel frameworks to increase online fact-checkers' engagement in fact-checking activities: (1) a recommendation system to personalize fact-checking articles and (2) a text generation framework to generate responses with fact-checking intention. · Finally, we propose a novel neural learning-to-rank framework to search for fact-checked information and integrate the found information into social posts. The search can directly warn fake news posters and online users (e.g. the posters’ followers) about misinformation, discourage them from spreading fake news, and scale up verified content on social media. We utilize multimodal content (i.e. text and images) to increase effectiveness of the framework.

  • etd-5331
Defense date
  • 2021
Date created
  • 2021-01-12
Resource type
Rights statement
Last modified
  • 2023-10-09


In Collection:



Permanent link to this page: