Who is hot and who is not? Profiling celebs on Twitter
Matej Martinc, Blaž Škrlj, Senja Pollak · Zenodo (CERN European Organization for Nuclear Research) · 2019
We describe the system developed for the Celebrity profiling shared task of PAN 2019, capable of determining the gender, birth year, occupation and fame of celebrities given their tweets. Our approach is based on a Logistic regression classifier and simple n-gram features. The best performance is achieved on the task of gender prediction, while predicting fame and occupation are slightly harder for the system. The worst performance is unsurprisingly achieved on the task of predicting birthyear, the hardest classification problem with seventy unbalanced classes. The proposed system was 3rd in the global ranking of PAN 2019 Celebrity profiling shared task.