Twitter Role Classification


The main goal of this project is to provide a web application that will host the existing TWIROLE model. According to its originators, “TWIROLE, [is] a hybrid model for role-related user classification on Twitter, which detects male-related, female-related, and brand-related (i.e., organization or institution) users. TWIROLE leverages features from tweet contents, user profiles, and profile images, and then applies the hybrid model to identify a user's role.” The main use of TWIROLE is to aid future evaluation efforts and research studies relative to investigations that rely upon self-labeled datasets. The web application is made to be easy to use and navigate allowing for a wide range of audiences including researchers or common Twitter users. Other goals of the project were to add a new classifier to the model that will improve the accuracy of TWIROLE. The model previously had only one advanced feature that used the k-top words method to analyze users and classify them. It looked at all of a user's tweets and ranked the k-top words used, classifying a user based on the words. The final model includes the previously mentioned advanced feature and an additional advanced feature that analyzes k-top emojis similarly to the k-top words feature. Once features were added, the model was then trained again on the existing data set, improving the accuracy. The website is made up of an HTML page using React on the front end. The backend (TWIROLE) is made using Django to render the HTML page, host images and other resources, and expose a GraphQL API. The front end makes AJAX calls to the GraphQL API which obtains the information that will be displayed on the website. The website is aimed at being simple to manage and update by those with experience in web development specifically Django, React, and GraphQL. The website is hosted by the Digital Library Research Laboratory. using their domain.



Twitter, Machine learning, Python, User Classification, TWIROLE, Marketing, WebDev