Gender Inference Based on Indonesian Name and Profile Photo
Lindung Parningotan Manik, Arida Ferti Syafiandini, Hani Febri Mustika, Zaenal Akbar, Yan Rianto · 2019
Social sensing is about building reliable system on top of unreliable data by using recent and top-notch technologies. Demographic understanding in a social sensing system requires several tools such as location, age, or gender inference service. This paper presents an alternative method to infer gender of a person based on the name input and/or the profile photo input. The result shows that the formula improves the standalone text classifier and image classifier. The F1 score of the standalone text (name) classifier reaches 83.71% by using the n-gram Logistic Regression model. On the other hand, the F1 score of the standalone image (profile photo with a facial feature) classifier reaches 98.14% by using the Inception convolutional neural network model. If these two classifiers are combined, the F1 score is improved further, reaching 98.62%.