User demographic prediction from social media content

Abstract

In past few years social media networking has been expanded a lot and because of this it enabled millions of users of all ages to develop and support their personal and professional relations. Although, a common flaw in these digital social websites is that it is easy to provide fake information about your name, age and gender in order to hide one’s identity. It allows the number of criminals such as pedophiles, online Threats, Stalking, Cyberbullying, Buying Illegal things, posting videos of criminal Activities. A vast number of fake profiles on social media makes it almost impossible to make manual analyses. In order to solve the above described problem, we have come up with three different techniques that are tokenization, feature extraction and classification. First when a user will write the comment or a text in the text box the system will process the text i.e. cleaning, tokenizing and removing the punctuations after the cleaning process the text will under go through a text feature extraction technique using TF-IDF algorithm through which the system will extract the features from the given text. After the feature extraction the system will then classify that if the provided text or a comment is written by a female, male or a brand on the bases of trained data. Our experimental results show that the system is generating satisfying results when can be further improved by either adding more data, treat missing and outlier values, feature engineering or algorithm tuning by applying these techniques we can improve the accuracy of the system in coming future. i