Toy-Pan12 Dataset

[SarahSalamati]

This table explains the toy dataset.

• √ means either a predatory author or a predatory message
• X means either a non-predatory author or a non-predatory message

  ---

  After analyzing and reviewing the dataset PAN_2012, I realized that an author is tagged a predator if there is at least one predatory message has been said by the author inside a conversation. This method may not be precise because maybe that person chats with one of the close friends. A better way would be to create a threshold and if an author have more than that threshold predatory messages in different conversations then we tag the author as predator. We need to check this on all conversations tho.

@hosseinfani

[SarahSalamati]

While I worked on the dataset PAN_2012, I realized:

• It can be said that most predators tend to talk a lot and their chats are usually long with the victims so that they can create a sense of trust in the person.
• Most of the chats between people are informal.
• According to reading the chats, it can be noticed that most of the predators have a great desire to ask a lot of questions from the victim. So for analyzing our data we need question marks and emojis.
• The other thing we can consider is the Behavioral factor in our research.
• Finding the data as an Author is Non-Predator and Message Predator is very rare and difficult to find it. @hosseinfani

[hosseinfani]

@SarahSalamati Thank you for the findings.

Yes, actually, I'm thinking of modeling predatory behaviour through #messages or time signals.

About your last note, I believe this is due to the way that the human annotators label the predators, that is if there is a predatory message, the author of that message is predator (which might not be correct as we discussed above)