The problem I faced was that Twitter changed the name from ‘Favourite’ to ‘Like.’

One of the things that prompted the focus of my study was the number of people who tweeted how important Twitter was in supporting their professional learning. So even before I began the PhD, I started ‘collecting’ tweets which illustrated this by ‘Favouriting’ them; favouriting rather than bookmarking them or archiving them elsewhere, simply because of the ease with which the ‘Favourite’ button could be clicked, whichever interface was being used to view the tweets. The idea of bookmarking tweets by favouriting is common practice and was found to be the dominant functional use of this feature (Meier et al, 2014). I was also aware that there was an IFTTT recipe which could automatically record your favourited tweets directly to a Google sheet; this meant that should I need to, I would have access to those tweets outside of Twitter. Then in late 2015, Twitter changed the Favourite to ‘Like’ and its icon, the star, to a heart.

Although this didn’t radically change my behaviour, I still Liked those tweets in which people described how important Twitter was for their learning, I was given pause for thought. There are a bunch of reasons why people Favourite/Like a tweet and I began to consider what effect(s) my ‘Liking’ might be having on those whose tweets attracted my attention. It is highly likely that Liking a tweet would be perceived differently from the way I was using it, so I set about thinking of an alternative way to catalogue the tweets I wanted to keep.

It took a couple of different strands of thought coalescing to spur me into action. Firstly I thought that one way to resolve my qualms about what my Liking behaviour might be doing to those on the receiving end would be to use a different account; one which appeared plain, bland and neutral. A second idea was that there might be a way of automating the process, rather than me having to manually Like specific tweets. This automated process could simply run in the background, 24 hours a day and catch all those tweets I’d never spot. Furthermore, it needn’t necessarily be restricted to those people who I follow. Essentially I was looking for a bot.

After a brief internet search, I came across a scripting tool even I, with minimal coding skills could manage. Using no more than a simple, editable Google script, the bot would automatically retweet and favourite (Like) and tweets containing a term or hashtag you choose. The PLDBot was born. There were ethical issues to consider at this point since this was not a method I’d specified in my ethics submission. I reasoned that this could be considered part of my ethnographic approach, since it would simply be involved in the same kind of activities  which I had been. However, the main difference of course would be that this bot wasn’t me and instead of being open about what it’s intentions were, this could constitute a switch to what might be construed as covert, rather than open research. To resolve this problem, when I created the new account, I made it clear that this was a bot (name, avatar), explained in the bio what it was doing (acting on my behalf and simply RTing/Linking) and I provided a link back to my research blog so it was (hopefully) obvious who was behind the account and what its purpose was.

Having set up the account I now turned to the script, for which I need criteria to invoke the RT/Like action. Again, trying to stay true to the research I had already been conducting and for which i had ethics approval, I chose a search term I had been using to monitor tweets through Tweetdeck. This was a compound expression and sought tweets that contained the specific term “professional development” AND twitter. Unfortunately, within the script, quotation marks are used in a particular way, so I couldn’t use them to specify “professional development” precisely, so I knew that might be a problem.

I knew from activity on Tweetdeck that more tweets contained professional development than professional learning, so I settled on that simply to explore that things worked. I launched the script and waited. After an hour, PLDBot has RT’d/Liked around thirty tweets; that was an order of magnitude greater than I ever saw on Tweetdeck for the same search term. With my alarm bells still ringing, I took a look at the tweets to see what was happening and quickly realised that the bot was RTing those tweets which had the specific term, professional development (phew!), but didn’t also include twitter, so any tweets with professional development in were being RT’d. I turned off the script and regrouped. It didn’t matter what I did, the ‘AND’ within the search criteria in the script wasn’t being actioned. I checked the FAQs and tried contacting the script author, bit without any luck. I even considered stumping up for the premium version, but all that seemed to offer was a simple, web-based interface rather than having to work in the raw script. Over the course of a week, I tried a few things, none of which resolved the problem of the script failing to execute the search criteria in the explicit way that Tweetdeck or the twitter.com search page worked. In the meantime however, as I experimented, the bot had been RT/Liking a number of tweets, which opened up a couple of fresh ethical issues. The simplest for me to address was that many of the tweets that it was RTing were from commercial enterprises who were marketing their professional development offerings. I’ve nothing against them doing that on Twitter, but I was firmly of the opinion that my research bot shouldn’t be helping profit-making companies with their marketing, so where that had happened, I manually unRT/Liked all of the tweets. Even if I could resolve the problem with the script, I would still have the dilemma of the bot potentially helping companies with their marketing and it being possible to perceive me as a researcher, through the actions of the bot, as promoting their products. The second concern which unfolded was when other Twitter users began to interact with the bot; something which I hadn’t anticipated at all. Granted, it was low level interaction like RTing the tweets the bot had already RT’d. There were even a number of other accounts which began to follow the bot, which really worried me, until I spotted that most of those who followed were doing so immediately after one of their tweets had been Liked by the bot. It looked like they had a script which automatically responded to Likes by following the account which had Liked them. One script being triggered by another! There’s no wonder there’s so many tweets per day!

Having parked the bot for a while, I had the chance to reflect. I’m sure that with a little more (some?) programming skill, I could have set something up to achieve precisely what I needed. Without that, I returned to the manual approach. I set up Tweetdeck with columns to return the two search criteria “professional development” AND twitter, and “professional learning” AND twitter and spotted that the number of tweets per day weren’t ridiculous; around a dozen or so. This meant I could leave those searches running in Tweetdeck and periodically Like or RT tweets which contained the terms, but which weren’t commercial in nature. During a cumulative time over the two phases, the bot bookmarked for me around a couple of hundred ‘Likes’ – tweets that referred to PD or PL in the context of Twitter. I’ve subsequently been able to download them using TAGS so that I can analyse them more carefully. As a slight aside, I also stumbled across a way of downloading all your Liked tweets as a PDF which is formatted similar to a book with about half a dozen tweets on each page and each month acting as a chapter. No images though.

So what?

Well the bot experiment prompted me to reflect on a number of things:

• It required me to respond to the call for Internet researchers to be adaptable, flexible and responsive to the circumstances in which they find themselves.
• It raised some unanticipated ethical issues and reminded me that making an ethics submission at the start of a project does not absolve you of further responsibility. As researchers, we need to maintain an ethical sensibility which recognises that we will “encounter new contexts and ethical issues that emerge in the ongoing evolution of both the internet and our multifaceted efforts to research the communicative engagements these technologies make possible.” (Markham and Buchanan, 2012)
• It made me regret that I’d not thought of it during the planning of my study so that I could have considered incorporating it as one of my main methods. Despite the irritation that many of the bots cause, some actually add value to the interactions and connections within Twitter. If I’d been aware of the teacher bot before planning my project, I’d definitely have been keen to explore the notion of a researcher bot – what it could do on a simple level in terms of gathering information, but also whether there might be value in designing it to interact with members of the community in which it would operate. Interesting possibilities for opening new insights?
• It added further fuel to the fire which ethnography on the Internet is bringing to traditional ethnography. Is it possible, practical, ethical or even desirable to automate or outsource part of the ethnographic gaze?

 

Markham, A., & Buchanan, E. (2012). Ethical decision-making and internet research: Recommendations from the aoir ethics working committee (version 2.0).

Meier, F., Elsweiler, D. C., & Wilson, M. L. (2014, May). More than liking and bookmarking? towards understanding twitter favouriting behaviour. In Eighth International AAAI Conference on Weblogs and Social Media.