22003279003 sehrash (1) (1).docx
- 1. Submitted By: Sehrash Safdar Submitted to: Dr. Mujeeb Ur Rehman Program: MSCS Semester: Second (3rd) University of Management And Technology DATA MINING ASSIGNMENT # 1
- 2. Title: Big data ml based fake news detection using distributed learning Summary :- In today era most people rely on social media. Many people share news on social media and it is difficult to identify between real and fake news. The increase of fake news on different platforms of social media has a serious issue many machine learning deep learning and neural networks used to detect fake news is in any form such as audio video and text. Also many users re share retweet and rewrite fake news to prevent the spreading of fake news. we use characterization and detection in real and fake news. Fake news must be authenticate that there is no miss leading news or information. To identify the fake news researchers must understand the news text and news content to categorized. Many researchers work on fake news detection in different languages. For detection of fake news developed a distributed cluster environment to detect fake news. The features are represented in three ways user based contacts based and social content based. In user based features we consider user profileing user credibility and user behavior. In content based the features are linguistic and sensitive style based and visual based the social context based features are network based impact based and temporal based.
- 3. In in this section we check the effectiveness of recurrent neural networks by modeling news articles which is depend on article body content and the article title we make clusters to find real and fake news. To detect fake news we firstly designed a corpus which is include stances and body of corpus than merge the data set include new articles and then labelled the data set now the labelled corpus will be processing firstly we remove stopwatch timings and eliminates special character by using rogex tokenizer for feature extraction We use limatization hashing TF and idf for individual ml classifiers we used decision tree logistic regression and random forest then by merge these classifiers a meta classifier will be designed to predict like dis like UN related and discuss news on the basis of user interaction. The dataset has four classes these are agree disagree dislikes and un related to check the ability of a model of true and fake news we can use evaluation matrices by using supervised learning techniques. To detect fake news we can use social media interactions on new stories a graph kernel based approach used to detect these patterns and behaviour to detect fake news. We first analyzed at them feature extraction unit classifier and then recombined classified classification the data set consist of 75% training and 25% testing data the data set is created by combining 4997 to stances with 1683 bodies based on ids. In data processing the raw data and incomplete data converted into machine readable form. In feature extraction raw data converted into numerical features we use different parameter models and parameter settings such as random forest logistic regression decision for feature extraction the highest classifier performance is 92.45% F1 score which is 7% more than the existing models. In future researchers work with un supervised learning techniques to find fake news on social media platform by using different social interaction such as liking dislikeking reshare or based on user profile. ------------------------------------------------------------------------------