Global Journal of Computer Science and Technology, C: Software & Data Engineering, Volume 22 Issue 2

Keep the model and do not make any adjustments if the prediction is accurate. Make adjustments to the model if the prediction is incorrect. The key idea of passive-aggressive classifier is that with each misclassified training data point it gets, the classifier changes its weight vector, attempting to fix it. Initially, the weight W is 0, then the algorithm receives a new doc, then the algorithm applies feature extraction and normalizes the doc. If the dTw value is greater than 0 the algorithm will predict the class as a positive and prediction class will be The core concept of the passive-aggressive classifier is to determine the loss function and act according to it. So, the loss function of the passive- aggressive classifier is Loss: L = max(0; 1 y(dTw)) If there any misclassification, there will be a loss. If the data set has an input class that belongs to a positive class, but initially the output is a negative class. For this misclassification, the passive-aggressive classifier will update its initial weight. IV. O ur W ork A workflow is a sequence of tasks that processes a set of works. In Fig. 4 we have shown our workflow. Firstly, we have collected true and fake news from the various traditional media. As our data are text data so for feature extraction we approach through two different feature extractors one is TF-IDF (Term Frequency and Inverse Document Frequency) and another is count vectorizer. Then we have tested them with four different supervised algorithms and finally discussed the final result based on our data set. a) Data Collection Process The main focus of our work is to collect news from various traditional news media as shown in Fig. 6. We choose the most popular and reputable news websites from across the world to construct a collection of authentic news. Such as Aljazeera, The New York Times, BBC, CNN, and many other news websites as well. For fake news, we have collected from some reputed fact-checking websites such as Lead Stories, PolitiFact, Health Feedback, etc. These fact-checking websites offer a clear and instructive explanation of fake news that has previously been published on other websites. We have the nine metadata from news sites, the title of a news annotated as a headline, the main article referred to the body, URL of the news, the date of publication, publisher name, the type of the article. In the fake news data set, we added two additional attributes, one is a fact-checking website name and another one is URLs of fact articles for the authentication of the news. An example of the data collection process for our data set is shown in Fig. 5. Our collected data set is given in the link 1. We have labeled the true as 1 and the fake news as 0. Fig. 5: Meta data example b) Data Preprocessing Data preprocessing is a data mining approach that includes transforming raw data into a suitable state. Raw data is frequently incomplete, inaccurate, and even absent, and it is prone to have many faults in certain actions or patterns. Preprocessing data is a well-known approach for resolving such issues. Initially, input data is typically provided in the raw state, which is in the texts, sentences, comments, articles, messages etc format. It requires certain clean-up or preprocessing before the data can be moved to machine learning algorithms, so algorithms can focus on key terms instead of terms that adds limited or no importance. So in raw state text data, there is a special character. Non-alphanumeric characters, as we all know, are special characters, these symbols are most frequently found in text data. These symbols contribute little meaning to text y = ± 1 . Algorithm 1 Passive Aggressive Classifier. Initialize w = (0,....,0) Receive a new doc d = d 1 .....d v Apply feature extraction, normalize k d k = 1 Predict positive if d T w > 0 Observe true class : y = ± 1 Want to have: d T w ≥ +1 if positive ( y = +1) d T w ≤ − 1 if negative ( y = − 1) Same as: y d T w ≥ 1 Loss : L = max (0 , 1 − y ( d T w )) Update: w new = w + yLd Fake News Detection: Covid-19 Perspective Global Journal of Computer Science and Technology Volume XXII Issue II Version I 4 Year 2022 ( ) C © 2022 Global Journals • Passive: • Aggressive: −