The next speaker in this AoIR 2017 is my DMRC colleague Brenda Moon, whose focus is on reply chains on Twitter. There are a number of ways in which replies are chained together, and in fact the term ‘reply tree’ may be preferable to ‘reply chains’: there may be many replies to the same original tweet only, or a long dyadic interaction over a series of tweets, or various permutations between these two extremes.
Brenda’s work uses the TrISMA dataset of all tweets sent by Australian accounts over several years; this may miss tweets in a reply tree if those tweets are not part of the Australian dataset (e.g. because they were not sent by Australian accounts). Except for this, however, reply trees can be constructed by using the ‘in_reply_to’ field in the tweet metadata, and the entire dataset of more than 2 billion tweets can be processed (with some difficulties) to assign each @reply to a unique reply tree ID.
Some 20% of all tweets belong to a reply tree, and some 15% of these reply to tweets by other Australian accounts. Across the entire dataset, the vast majority of these trees are very short (five steps or less). The trees with with considerable user involvement tend to have relatively few steps (they represent star networks), while trees with few participants can operate over longer instances. The largest tree consists largely of single-level responses to a teen idol promoting his latest album. The longest tree consists of some 230 responses between only two accounts, over the course of some two months.
The focus on Australian accounts’ tweets only may slightly alter these patterns, of course, and patterns in other nations may also differ. In Australia, at any rate, there does not seem to be much variation in these patterns over the years, even in spite of Twitter’s continued tweaks to reply functionality. There’s also a need to explore whether the specific shapes of reply trees are related in any way to the themes or tone of discussions on Twitter.