In social media environments users interact with thousands of digital objects that reproduce or optimise their cultural practices. However, it is difficult to distinguish between publications, interchanges or messages originally produced by humans in real time and other ones generated automatically through devices created to post, write or spread contents instantly. Automation tools or software that allow accelerated activity on sociodigital platforms are known as bots. Bot is a short term for web robot: an application (app) that performs tasks or runs scripts over the Internet without previous direct inputs (Dunham & Melnick, 2008; Chu, Gianvecchio, Wang & Jajodia, 2012). Most bots are spiders used for web crawling or web indexing, which means systematic browsing of websites to update the contents or indices of a web or search engine, but in the case of social media sites as Facebook, Twitter or Instagram, bots may be downloadable or commercial apps that respond to the intentions of a user who programs or manipulates them to acquire public notoriety, to enforce a marketing plan, or to support or detract a political or ideological cause (Kollanyi, Howard & Wooley, 2016).Despite 51.8 percent of all web traffic is made up of bots (Zeifman, 2017) academics still know little about their behavior. On this exploratory study we identified which apps available to anyone on Internet through Google Play store, the largest online platform to get over one million Android tools (Warren, 2013), are transforming human digital practices into automated interactions. Besides, this project aimed the final construction of a digital methods map: a flowchart of our reflections, activities and products for anyone interested in the detection of social media bots using our methodology.
As a starting point, we decided to work with a sample of downloadable or commercial social media automation apps considering only those tools available on Google Play store used for: 1) posting (or tweeting) automatically; 2) scheduling automatic posts or contents; 3) tagging all friends of a sociodigital network in one single content (post, tweet, link or photo); 4) reposting (or retweeting) the same content or message hundreds of times; 5) spamming through private messages (i.e. Facebook inbox); 6) getting or generating automatic “likes” or “follows” for all posts or contents of the same (or similar) account(s); 7) giving automatic “like” or “follow” to similar public pages; and 8) getting lots of “friends” or “followers” by adding accounts or groups massively.Considering these criteria we built and analyzed a dataset with the top 216 Google Play store automation apps for Facebook, Twitter or Instagram. Later, we used a dataset collected in a previous research to contrast the use of these apps in a case of study: 29 thousand tweets about the last presidential election debate in Argentina (November, 2015). In this last dataset we found which tweets were generated or spread using an automation app, and we matched if the apps detected were part of the top ones in Google Play store.
We entered into the Google Play store website and searched for some popular and high-rated automation apps for social media platforms. We used as generic searching criteria “Facebook automation”, “Instagram automation” and “Twitter automation”. Later, we filtered some of the results using more specific searches, such as “automatic” plus “posting”, “sharing” or “liking”, or either “automatic post schedule”, “getting more friends (or followers)” and “automatic re-post (or retweet)”. After checking each one of the obtained results corresponded to an active and functional app, we made a small list of 17 apps considering: a) name of the app (as shown in Google Play); b) application ID (a two part string with both letters and numbers used for product identification); and c) the product´s URL.4.2. Building and cleaning the dataset We uploaded this list in the Google Play Similar Apps scraper of the Digital Methods Initiative toolkit. This tool extracted app details for each given product and it also got a new dataset of all the “similar” apps for every token in the list. In total, we found that if we searched in Google Play our original 17 apps list, we would get 478 new ones. However, the resultant dataset needed cleaning and attention. Some of the apps were repeated (for example, showing the American and British version of the same product, or the standard and pro versions). In some other cases, the apps had no assigned metadata (lacked of name, developer, country, etc.), so we tagged them manually. In other cases, metadata was unclear or unreadable (showing signs and numbers instead of letters in the name); we found these anomalies corresponded to apps with names that used non-western characters, like greek, russian or japanese ones, so we substituted the name with a transliteration. After cleaning the database we got only 216 different apps, but all of them acquirable and functional. 4.3. Automation apps network analysis We turned our definitive dataset into a CSV (comma-separated values) table and modeled it like a network using Gephi. There:
The central and most relevant nodes on the similar apps network we generated using Gephi showed that the top Google Play store automation tools are:
Although the concepts automation software and bot can be used indifferently, the term bot has acquired negative connotations due to its recent malicious purposes. Since the late nineties some bots have interfered users' privacy through address harvesting, while others have worked as malware programs, viruses, worms or zombies (applications without clear metadata or tracking records) used to infect systems and steal information (Morales, Bataineh, Xu & Sandhu, 2010). In the actual context of social media platforms bots can be understood as automatic uploaders of text, links or digital objects, machines to add thousands of friends automatically or apps to increase the impact of a user through massive reviewing, “liking” or commenting (Edwards, Edwards, Spence & Shelton, 2014). Following this logic, after exploring the #ArgentinaDebate dataset we found:
There is a close relationship between the use of bots and anonymity. The massification of certain positive or negative messages to support or discredit a presidential candidate (speaking about our Argentinian case of study) often implies the concealment of the identity of the author of the texts. We estimated Twitter bots users, in order to avoid tracking or identification, tend to open and manage ghost accounts using fake names and profiles, or showing a real name but looking for a low profile. According to Bryant (2014), a ghost Twitter account is a profile used exclusively for any kind of propaganda, avoiding interaction. After detecting which accounts using bots have a very low rate of followers (zero or close to zero) and a high number of tweets or retweets (more than 50), we introduced some of their users´ names in the online tool Bot-or-not, which analyze the history of every account to get its reliability. Some accounts with a few followers and active tweeting were detected as fake by Bot-or-not.
Recapitulating our methodology and main findings, we developed a methods map with each activity and product, considering: 1) the scraping of Google Play store and assemblage of our first dataset; 2) the analysis of this dataset using a Gephi network and its central nodes; 3) the contrast of the most relevant Google Play store apps (using the “similar searches” criteria) with another dataset (Twitter #ArgentinaDebate tracking); and finally, 4) the study of the relationship between automation apps and ghost accounts.
Actually, there are no clear regulations for the use of automation apps (Edlich & Sohoni, 2017). When someone downloads any kind of social media app on Google Play store there are no agreements detailing the possible purposes behind the uses of these apps, nor any special rules or consent policies. After exploring the Google Play store case we found automation apps are available to any person and public (without filters in the platform), however, the download tendencies are different for every social media user and app:
After exploring which are the companies that develop the highest number of automation apps we discovered some paths for a future study to trace the economy of easy-access bots. The top producers of these apps are large Internet enterprises like Google or UberMedia, along with important social media platforms like Facebook, WhatsApp or Instagram. It could be supposed that the design of bots come mainly from small or clandestine companies, but the same programmers of the largest web services are the ones behind automation apps, offering them for free as part of their downloadable tools. Just in the case of Twitter the most relevant apps are not closely related to the company managing the platform. Nevertheless, the Twitter apps market has empowered medium-size companies such as IFTTT, Botize or Hootsuite too, which have grown considerably in the last five years due to their automation services, and can be easily found on the web. In our #ArgentinaDebate dataset (29000 tweets) IFTTT is the most relevant app, covering almost 52% of the market (132000 tweets), but to assert this is the top automation app used in Twitter nowadays, we would need more cases of study.6.3. Some practices involving social media bots in politics
There were an important number of ghost Twitter users in the #ArgentinaDebate dataset we explored, so we may suppose the use of bots is a popular practice in politics, at least in Latin American cases. Not all the users that support their candidate´s promotion through automatic “tweeting” or “following” were considered as bots in our study. We considered all the non-anonymous users as legitimate citizens on Twitter and we only defined as bots those users with a ghost or anonymous account who were incorporating automation apps to their communication and political practices. In the case of Argentinian presidential debate almost a third part of our whole dataset were tweets produced by bots. Besides, these bots were diverse in their affinity to ideologies or parties, so it is possible to think politics are strongly influenced by automation in general, and all candidates and factions look to get more sympathizers in Twitter through the recruitment of ghost and automatized users.6.4. Some reflections on a new typology of social media automation users
After analyzing briefly the behavior of some bots detected in #ArgentinaDebate dataset we could propose a draft for a new typology of social media bots and automation apps users:
As part of the Methods Maps: Visualizing automation project of the Digital Methods Summer School 2017 at University of Amsterdam, we did not only identified and analyzed bots and automation apps in social media platforms, but also designed a digital methods map to make our study replicable in the future. The objective of our methods map is to give other scholars the opportunity to research automation tools in different temporary and geographic contexts in the future. We only focused on apps available on Google Play store and in an Argentinian political event in Twitter, but anyone could modify or amplify our methods map proposal to study different apps stores, cases of study or agendas, and social media platforms. Additionally, future researchers can use more than one cases of study or platforms, as well as other digital methods different than ours, adding new statistic or cultural analytics programs to our network analysis in Gephi, or diverse visualizations to our Tableau graphics. We only recommend to be careful during the construction and cleaning of the datasets and to explore and test all the considered methods or platforms before doing the final research.
Our evidence suggests that automation has been important in some common practices on social media platforms, especially on those related to political events. Automation apps are not difficult to find, and there are more than 200 available on Google Play store for free or at a low cost. The market around automation apps is so relevant for Internet economy that some of the main tools developers are the largest bussiness that provide web services, such as Google or UberMedia, while other outstanding developers are big social media companies like Facebook or Instagram. Twitter is not a significant designer of automation apps itself, but has allowed the activity and growth of many companies that develop and distribute these apps through websites, such as IFTTT, Botize or Plume. Also, there are many automation tools for Twitter on Google Play store, and all these, just like the ones available on the web, are often used by ghost accounts to schedule messages or activities, to spread a message massively, or to get social or political relevance. Automation has to be pointed out in future researching on the political and cultural uses of social media, taking into account there are no actual or clear regulations on the production, commerce or use of bots, and that many social media or marketing platforms are gaining advantage from this lack of transparence. Some considerations have still to be discussed for further studies, like if bots can be either artificial, corporative or human, because at least in this project, they seem to produce the same or very similar performative effects.
|png||Tools small network.png||manage||260 K||18 Jul 2017 - 04:08||EloyCalocaLafont||Tools cluster|
|png||Sociality and communication cluster.png||manage||232 K||18 Jul 2017 - 01:57||EloyCalocaLafont||Sociality and communication cluster|
|jpg||Ranking of developers.jpg||manage||66 K||24 Jul 2017 - 18:08||EloyCalocaLafont||Ranking of automation apps developers|
|png||Productivity small network.png||manage||58 K||18 Jul 2017 - 04:09||EloyCalocaLafont|
|png||Method Map Hello bots.png||manage||83 K||18 Jul 2017 - 23:19||EloyCalocaLafont|
|Method Map Hello bots_GabrielaSued EloyCaloca.pdf||manage||323 K||18 Jul 2017 - 23:23||EloyCalocaLafont|
|jpg||Ghost accounts using bots.jpg||manage||152 K||18 Jul 2017 - 21:53||EloyCalocaLafont|
|png||Figures 3 and 4 wiki.png||manage||129 K||18 Jul 2017 - 02:42||EloyCalocaLafont|
|png||Figures 1 and 2 for wiki.png||manage||203 K||18 Jul 2017 - 02:01||EloyCalocaLafont|
|png||Figure 5 Most used apps in ArgentinaDebate.png||manage||24 K||18 Jul 2017 - 19:05||EloyCalocaLafont|
|png||Figure 5 Most used apps 1.png||manage||24 K||18 Jul 2017 - 19:07||EloyCalocaLafont|
|png||Bots network complete2.png||manage||110 K||18 Jul 2017 - 01:10||EloyCalocaLafont||Google Play similar apps network|
|csv||argentinadebate-dataset.csv||manage||7 MB||19 Jul 2017 - 20:52||GabrielaSued||Dataset of 29000 twits collected on last Presidential Debate in Argentina (2015-nov)|