IJIREEICE ISSN (Online) 2321-2004 ISSN (Print) 2321-5526 International Journal of Innovative Research in Electrical, Electronics, Instrumentation and Control Engineering Vol. 8, Issue 9, September 2020 DOI 10.17148/IJIREEICE.2020.8905 Copyright to IJIREEICE IJIREEICE 26 This work is licensed under a Creative Commons Attribution 4.0 International License Machine Learning Based News Validation Reetodeep Hazra 1 , Megha Banerjee 1 , Judhajit Sanyal 2 Student, Department of Electronics and Communication Engineering, Techno International New Town, Kolkata, India 1 Assistant Professor, Department of Electronics and Communication Engineering, Techno International New Town, Kolkata, India 2 Abstract: The use of machine learning has become widespread in recent years, especially due to the impact of media content on the general population. In particular, the validation of truth in the context of news has become a critical necessity, due to its ready availability from verified and unverified sources, and its ability to influence people majorly. The present work outlines a LSTM (long short-term memory) based approach to news validation. The results obtained by the model are presented in terms of the training data set and contrasted with the results obtained from the test data set. Appreciable accuracy is achieved through the model, seen through the corresponding loss curves and confusion matrix. Keywords: Machine Learning, LSTM, News Validation, Loss Curves, Confusion Matrix. I. INTRODUCTION In recent years, the focus of a significant amount of research has been on the application of Machine Learning (ML) techniques in gauging social interaction as well as the impact of media on society. The multitude of ML techniques available allow for in-depth analysis of different types of data, among which techniques such as Naïve Bayesian Analysis, Spline Regression, Support Vector Machines (SVM) and neural network-based methods are quite popular. A model which has recently shown promise in the field of sentiment analysis and truth validation is Long-Short Term Memory (LSTM) based neural network. The work presented here outlines the performance of an LSTM network in fake news detection. The paper is arranged in the following manner. Section II presents a survey on the different approaches employed by researchers. Section III presents the proposed model. The results obtained by application of the proposed model are outlined and the appropriate discussions are presented, in Section IV. Section V concludes the paper. II. LITERATURE SURVEY Efforts have been made by researchers in recent times to map user sentiment with respect to different types of media, with one recent crowdsourcing based approach presenting a user approval estimation model for different types of web series from minimal data [1]. In a similar manner user behaviour prediction in terms of attrition probability of employees of a company has been illustrated in [2], using Naïve Bayesian estimation. A regression spline-based estimation technique of customer spending score has been presented in [3]. In recent years, one of the greatest challenges of the populace has been the identification of news items as fake. The spread of fake news and hoaxes in recent years also has its roots in the socio-political unrest sweeping over the world today, and hence many scholars have considered this a critical problem that needs to be addressed immediately. Some of the approaches to fake news detection in recent times have relied on SVM analysis for natural language processing [4], Naïve Bayes classification of news items for validation [5], and machine and user based multi-validation model [6]. One of the most interesting approaches in recent times has been the application of social and content-based models for fake news detection, as resented in [7], where real-world testing of the model has provided promising results. III. PROPOSED MODEL The stacked LSTM network has a visible layer of input, three hidden layers and an output layer with Rectified linear circuit (ReLu) activation unit that predicts single value or binary classification. An LSTM layer requires three- dimensional input and by default delivers a two-dimensional output. In this classifier, the primary hidden layer is an embedding layer with a input sequence length of 300. The input dimension characterizes the size of the vocabulary in the text information. Output dimension was set as 100. It defines the size of the output vectors from this layer for each word. Output of the embedding layer was a 2-dimensional vector with one embedding for each word in the input sequence of