Improving Feature Representation Based on a Neural Network for Author Profiling in Social Media Texts

Table 5

Obtained results (accuracy, %) for age and gender classification on the PAN author profiling 2015 English training corpus under 10-fold cross-validation.

Feature set	Age
	LR-NP	LR-WP	SVM-NP	SVM-WP
D2V (1-gram)	66.45	66.45	68.42	69.73
D2V (1 + 2-grams)	71.05	74.34	71.05	72.36
D2V (1 + 2 + 3-grams)	69.73	70.39	68.42	70.39
Character 3-grams	65.78	67.76	66.44	67.10
Bag-of-Words	65.78	65.78	65.13	65.13
Feature set	Gender
	LR-NP	LR-WP	SVM-NP	SVM-WP
D2V (1-gram)	59.87	66.44	56.57	69.07
D2V (1 + 2-grams)	63.15	69.73	61.84	71.05
D2V (1 + 2 + 3-grams)	65.13	69.07	65.78	71.71
Character 3-grams	57.23	61.84	59.21	62.50
Bag-of-Words	60.52	56.57	61.84	55.26