*5.2. Toxic Comments*

The Toxic dataset consists of two datasets as well, Train and Development. Train dataset consists of 159,571 unique comments labeled with 6 different types of toxicity: *toxic*, *severe\_toxic*, *obscene*, *threat*, *insult* and *identity\_hate*. Some example of comments are:


The Development dataset consists of 63,978 toxic comments with the same classes and their respective labels. The class distribution is heavily skewed towards toxic, obscene and insult, in both Train and Development, Figure **??**a.

TOXIC Train includes 29 unique class combinations, while Development has 21. Both frequencies follow a power law distribution, similar to SEM2018, Figure **??**b. The most common class combinations for Train and Development are: obscene and insult, obscene only and insult only. The most uncommon class combinations included threat and severe\_toxic classes. Out of 29 unique classes for Train 12 have more than 100 occurrences, while out of 21 unique class combinations of Development 7 appear more than 100 times.

**Figure 9.** TOXIC Class Distribution (**a**) and frequency of unique class combinations (**b**).
