Every time a retweet is created, Twitter generates a tweet encapsulating the original tweet and automatically adds “RT @author_screen_name:” in the beginning of the retweet text field. Therefore, the author of this original tweet is also added in the retweet entities.user_mentions field.
Despite being an involuntary (i.e., Twitter automatically does it for you) and hidden (you can only see on data, not in the web interface) mention, Twitter forgets to update the mention field, when necessary, leading to inconsistencies. For instance, if the original author changed his/her user.screen_name, Twitter will update the user info in the encapsulated retweeted_status object, but it won’t update the entities.user_mentions field. I imagine this is a rare scenario if you collect tweets using the stream API, however, if you are using any other search methods, this could be problematic.
The figure below shows an example. On the left, a tweet retrieved using REST API pointing to different users at retweeted_status.user.id and entities.user_mentions[].id; on the right the same tweet, but retrieved a while ago before user.id=124053720 had changed screen_name.
