Author profiling includes the study of various profiling traits of an author such as age and gender, using their authored content. This work describes our methodology for the task of cross-genre author profiling for social media data- spanning across mainly three categories- small tweets, medium-sized posts, and long hotel-reviews. It is inspired from the PAN 2016 challenge. Gender and age prediction problem is addressed as a classification task and our approach to this problem focuses on feature engineering. Furthermore, in our final model we combine different models which work well with different types of social media data using minimal, simplistic features. With our approach, we are able to achieve competitive accuracies for gender classification task and an improvement in the age classification task, as we further modified the problem to a more viable and useful task.
The final Jupyter notebook is located in the Joint_Model directory, along with results.