Research Article

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

Table 7

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

Feature set Gender
LR-NPLR-WPSVM-NPSVM-WP

D2V (1-gram) 61.76 67.65 61.76 64.71
D2V (1 + 2-grams) 64.71 70.59 67.65 73.53
D2V (1 + 2 + 3-grams) 61.76 58.82 67.65 64.71
Character 3-grams 76.4776.4776.4776.47
Bag-of-Words 64.71 67.65 64.71 70.59