In the chat log, a string records all words which had been spoken by any given user. Note that words are encoded/masked
like word:xxxxx where xxxxx is simply a number so we do not know the actual words used in the conversation, leaving
no part-of-speech or LIWC features available for analytics. This feature ends up simply being converted to a bag-ofmasked-word vector for classification where the feature is 1 if a particular masked word is used in a conversation, or 0 otherwise.