Introduction
The world we live in today is completely data-driven. The data landscape has evolved immensely, creating many new opportunities along with complexities. There are a plethora of sources available to gain information. The Internet has enabled the spread of information way faster than it was ever envisaged to be. Platforms like Social Media have become quick tools that enabled broadening the reach of a wide range of information.
This information technology is a double-edged sword. It has got its pros and cons. While it provides quick and easy access to a broad spectrum of information, it also allows the spread of misinformation. As the number of sources of information grows day by day, it becomes crucial to assure the veracity of such information and the sources.
What is Fake News ??
Any kind of news or information that represents misleading information that is aimed at deliberately misinforming the readers can be termed as “fake news”. The general objective of such fake news is to mislead and influence views, for political motives, to damage the reputation of some entity, or as a clickbait for economic or adversary profit. There are many instances where the spread of fake news has created much trouble for many governments including the countries like the U.S. It adversely affects the entire society in general. As per reports, just before the 2016 U.S. Presidential elections, fake news was spread widely which adversely influenced the outcome of elections. Considering the recent times, misinformation regarding the COVID-19 vaccination was circulated leading to chaos among people.
Such fake news is largely shared on Social Media platforms like Twitter, Facebook, and WhatsApp. Studies and researches done examining the rate of spread of fake news have concluded that the tweets containing false information spread six times faster than those
which contained the true information. Due to diverse and innumerable posts and tweets being made each second on such Social Media sites, it isn‟t easy to solve this problem of “fake news”. Hence, fake news makes it harder for people to perceive the actual truth. A Gartner Research has forecasted that “By 2022, most people in mature economies would consume more false information than true information”.
Types of Fake News:
Based on the aim and effect of fake news, it can be categorized into seven types as depicted in the picture.
A misinformation matrix can be designed that can help us understand these types in a better way. The misinformation matrix shown below summarises the different impacts for each of the mentioned types.
How to detect fake news ??
The difficulty in detecting fake news can be attributed to its resemblance to real or true information. As the volume of information grows rapidly, it becomes practically impossible to detect fake news by implementing manual check-ups.
Artificial Intelligence to detect Fake News:
Detecting fake news manually is complex for many reasons. The cumbersome task of a manual approach to recognize fake news can be subjective, time-consuming, and not very efficient. It is in this scenario that the emerging technology of Artificial intelligence can be utilized to detect and flag fake news and such irrelevant information. Organizations like Facebook implement various Artificial Intelligence algorithms to assure the veracity of information.
Artificial Intelligence, being a double-edged sword itself, was also utilized to create fake news. Machine Learning Algorithms with the ability to create fake news, so-called “deep fakes” (photos/videos resembling real people), etc.; were also developed.
A saying given by Sun Tzu (Chinese Military General, strategist, and philosopher) goes, “To know your enemy, you must become your enemy.” This implies that in order to fight one‟s enemy, one has to think like the enemy. The same strategy is being followed by the Artificial Intelligence researchers in developing AI algorithms that detect and flag such AI-created fakes.
Basic AI process flow in analyzing and detecting fake news:
o Data(News) Collection: Extraction of news to be analyzed.
o Data Exploration and Analysis: Usage of statistical techniques to analyze data and form inferences.
o Data Transformation: Replacing data variables by some function of the data
variables so that the Machine can easily interpret and analyze them.
o Feature Extraction: Retrieving useful data attributes that can be helpful in modeling.
o Feature Vectorization: Converting data into some form of vector for easier comparative analysis.
o Building Model: Creating an AI/ML model with certain parameters(hyper parameters).
o Training Model via AI algorithms: Training the built model by feeding it with training data.
o Implementing Model: Deploying the model on data.
o Decision: Making a required decision based on the analysis results from the model.
Some of the AI Techniques to detect fake news:
⮚ Anomaly detection:
Anomalies/Outliers are those data points in the data which differ significantly from the rest of the data. Anomalies/Outliers in data can be detected by basic exploratory data analysis. AI algorithms like KNN, ARIMA, One-class SVM, etc.; are also utilized to detect these anomalies. The anomalies are analyzed in terms of volume and those which differ greatly on comparing with different sources are classified to be fake.
⮚ Stance detection:
The relationship between the title and the matter of the text is analyzed via algorithms like Natural Language Processing, Neural Networks, etc.; and the decision is made whether the fragments of text are :
– Related
– Unrelated
– Neither (To be discussed)
⮚ Analysis of propagation of fake news against that of real news:
The trending news items are taken into consideration and are analyzed comparatively considering factors such as time of release, kinds of sources(classes) and the number, etc. Data analysis is done on the raw data, patterns if any are verified, and algorithms are applied to classify the sets of information.
⮚ Network-based approach:
Social network behavior can be analyzed using AI algorithms. The associated data can also be assessed to retrieve the required features that can help distinguish fake from real.
⮚ Linguistics approach:
The writing style of the author, semantics, syntax, structural patterns etc.; are inspected and comparatively studied, and obtained significant features are then fit into Machine Learning models implementing AI algorithms like SVM, Naïve Bayes, etc.
Algorithms like N-Grams are also implemented which help analyze data using features like the frequency of different kinds of words to assess the relevance of those words to the topic.
⮚ Verification of facts stated:
Knowledge graphs and referential web sources can be utilized to verify the accuracy and authenticity of the piece of information.
⮚ Image Forensics:
AI algorithms like Neural Networks can be used to extract features including image size, image pixels, etc.; to detect the authenticity of the image. Photoshop use, „deepfakes‟, etc.; can be detected using well-trained Machine Learning models.
⮚ Hybrid approaches:
A combination of one or many techniques can be made use of to devise efficient and optimum solutions.
Conclusion:
Dr. Karsten Donnay, of the University of Zurich who is working on a project to utilize AI tools for fake news detection has stated that “fake news is structurally different” which thereby aids in distinguishing fakery including bias from real facts. AI techniques can be used to identify bias and fakery in the daily news, which can aid in presenting factual information in an ideal and neutral way that readers can easily comprehend and trust. When utilizing AI tools one should keep in mind that “Artificial intelligence is only as smart as the programming and data that aids it”. Therefore, the core of the application lies in providing the necessary inputs that help a machine mimic a network of human intelligence in distinguishing fake news from real news.