An Optimal Deep Learning Approach for Classification of Age Groups in Social Network

Abstract

There is huge amount of data in social networks, where people post their opinion on a topic, or share their information. But people often don't provide their personal data, like gender, age and other demographics. Research can be done on this data to develop applications of sentiment analysis, but the success rate is restricted by the number of words in the dictionaries as they do not consider all the words which reflect the sentiment in our messages as most of the communication on social networks is non-standard language with small messages. Moreover, with contemporary technology it is quite easy to create profile with false age, gender and location which provides criminals an easy way to deceive. Thus we can analyze the text messages posted by the user on social network platform. As per the research done so far, age is one of the important parameter in the user profile which reveals the important information about the typical behavior among same age group users. An analysis is done with more than 4000 tuples which contains relevant parameters like number of friends, length of message, number of likes, number of hash tags and comments are considered for the classification. In this study, we use the user profile information for the prediction of age group, which we collected using Facebook API. In this paper we classified the users into two age groups teenagers and adults using different Machine learning algorithms like deep convolutional neural networks, Multilayer perceptron, Random forest , SVM and Decision trees. Among all these algorithms deep convolutional neural network stands out to be the best among all of them reaching the best performance with an accuracy of 94%.

Keywords

• dcnn
• multilayer perceptron
• decision trees
• facebook
• svm