Inferring Trust from Message Features Using Linear Regression and Support Vector Machines

Abstract

With the proliferation of social media like Facebook and Twitter more and more people have started depending on it for all sorts of news. Social media (such as Facebook, twitter), topical forums, wikis etc. enable a community of end users to interact or cooperate towards a common goal. However, the effectiveness of disseminating information through social media lacks in quality as a result of less fact checking, low barriers to entry, more biases, and several rumors thereby making social media not only the source of genuine news but also fake news. Thus, in this paper we are dealing with the problem of classifying information on the basis of its reliability using freely available user generated data like tweets on twitter using machine learning algorithms. A variety of machine learning algorithms like linear regression, logistic regression, Naïve Bayes etc. are used to determine the trust carried by the tweets. In addition to basic machine learning algorithms we present a learning model that is based on the combination of naïve Bayes and logistic regression and also naïve Bayes and linear regression. Also a technique has been discussed to determine a threshold parameter which is used for the classification of data in case of SMO and linear regression algorithm in order to increase their accuracy which is achieved for modified linear regression algorithm in most of the cases. The modified linear regression algorithm when used in combination with naïve Bayes achieves a better accuracy than all the other algorithms used in this paper. Also hybrid classification works better than the individual algorithms.