6. Results
In terms of F1 scores on the first 10, 100, and 500 postings, this model
performs noticeably better than the one described by Losada et al
[2]. Additionally, this model outperforms the F1 scores of Banovic
et al. [15] for all related data subsets. The average time between
postings was the only feature that significantly improved the
performance of the basic LR + TF-IDF model, despite the fact that data
analysis provided fascinating insights as shown in Table 1. Other
additional characteristics, however, only added needless noise that
reduced model accuracy on the test set.