Standing
Graduate (Masters)
Type of Proposal
Oral Research Presentation
Faculty Sponsor
Dr. Hossein Fani
Proposal
Historically, fake news was mainly propaganda. Now it's used to modify people's beliefs and perceptions about specific phenomena, resulting in their change of behaviors. Identifying fake news articles can be difficult for many people.
Algorithms to identify if a news article is fake can be used to help people avoid potential misinformation and lead to a reduction in the spread of such false news and discourage the creation of such articles.
The research project aims to make fake news detection more efficient computationally and storage-wise by combining two methods. The first method performs feature reduction, which means reducing the number of parameters in a dataset (in this case, a news article) to compress it. The feature reduction method we use is called singular-valued decomposition (SVD). The compressed data is passed into a neural network model. Building a neural network model refers to constructing a model in a computer that is similar to how the neurons in our brain pass information. In this study, we use a natural language processing (NLP) architecture of long short-term memory (LSTM) based on neural networks.
For example, an article "Trump has won the 2020 US elections", with a label 1 i.e. fake news. When the article is passed through the proposed work, using SVD the article is reduced to a representation like [1.238, 4.56, 0.87]. This representation has a length of 3, instead of the length of the article. This reduced representation is used to train the LSTM to learn the label of the article.
Availability
March 30
Efficient Fake News Detection Method using Feature Reduction
Historically, fake news was mainly propaganda. Now it's used to modify people's beliefs and perceptions about specific phenomena, resulting in their change of behaviors. Identifying fake news articles can be difficult for many people.
Algorithms to identify if a news article is fake can be used to help people avoid potential misinformation and lead to a reduction in the spread of such false news and discourage the creation of such articles.
The research project aims to make fake news detection more efficient computationally and storage-wise by combining two methods. The first method performs feature reduction, which means reducing the number of parameters in a dataset (in this case, a news article) to compress it. The feature reduction method we use is called singular-valued decomposition (SVD). The compressed data is passed into a neural network model. Building a neural network model refers to constructing a model in a computer that is similar to how the neurons in our brain pass information. In this study, we use a natural language processing (NLP) architecture of long short-term memory (LSTM) based on neural networks.
For example, an article "Trump has won the 2020 US elections", with a label 1 i.e. fake news. When the article is passed through the proposed work, using SVD the article is reduced to a representation like [1.238, 4.56, 0.87]. This representation has a length of 3, instead of the length of the article. This reduced representation is used to train the LSTM to learn the label of the article.