Investigation of Fake Profiles in Social Media Content Using Natural Language Processing

Abstract

Data mining and machine learning employed on huge data help to recover the target patterns for the applications of prediction, classification, and pattern recognition. The machine learning technique is the best fit for the binary classification problems. Thus, the proposed work introducesa new multi-model feature classification system for fake profile identification on social media. The work is conducted in two main parts, the first part includes the dataset collection and refinement of different profile attributes. The second part includes, merging of profile attributes and profile content using data profiling. In the dataset collection and refinement phase, dataset is used as the profile attributes and thereafter data includes the contents published by the user. Thereafter, data features are reduced and only essential features are used for experimentation. In the next phase, the extracted published contents are used with the Natural Language Processing (NLP) parser for transforming the data into the vector format. Thereafter, profile attribute and profile content are combined using the data profiling. The profiled data is further used with an improved Artificial Neural Network (ANN) to learn and classify the data. An algorithm is proposed to compare two popular machine learning models i.e., Support Vector Machine (SVM) and ANN to classify the data. The decision scores are finally generated based on classification outcomes. All the experiments have been done using real world twitter published dataset of user profiles which demonstrate and helps in the identification of fake user profiles. Finally, the performance and accuracy of proposed model has been analyzed using certain accuracy matrices such as Precision, Recall, F-Measures. newline