Online Human-Bot Interactions: Detection, Estimation, and Characterization

“We benchmarked our system using several off-the-shelf algorithms provided in the scikit-learn library (Pedregosa et al. 2011).”

“Sentiment extracted from social media conversations has been used to forecast offline events including financial market fluctuations (Bollen, Mao, and Zeng 2011), and is known to affect information spreading (Mitchell et al. 2013; Ferrara and Yang 2015).”

“…extraction techniques to generate various sentiment features, including arousal , valence and dominance scores (Warriner, Kuperman, and Brysbaert 2013), happiness score (Kloumann et al. 2012), polarization and strength (Wilson, Wiebe, and Hoffmann 2005), and emoti-con score (Agarwal et al. 2011).”

“Features extracted from user meta-data have been used to classify users and patterns before (Mislove et al. 2011; Ferrara et al. 2016a).”

“This system leverages more than one thousand features to evaluate the extent to which a Twitter account exhibits similarity to the known characteristics of social bots (Davis et al. 2016).”

“For each of these accounts, we collected their tweets through the Twitter Search API.”

“Datasets were filtered and cleaned in order to retain only relevant information for further analysis (Di Minin et al., 2018; Varol et al., 2017).”

“3) Fake User Detection: Fake user detection has always been a hot research topic in social networks [61], [62], which can help to avoid the economic loss caused by malicious activities.”

“Social bots (Ferrara et al., 2016; Varol et al., 2017) can be designed to target vulnerable communities (Shao et al., 2018b; Yan et al., 2020) and exploit human and algorithmic biases that favor engaging content (Ciampaglia et al., 2018; Avram et al., 2020), leading to an amplification of exposure…”

“Past research has estimated that 5-8% of twitter accounts are automated [36].”

“The research community offered multiple approaches for identifying automated accounts (Davis et al. 2016; Varol et al. 2017a; Yang et al. 2019; Chavoshi, Hamooni, and Mueen 2016; Subrahmanian et al. 2016; Chen and Subramanian 2018; Kudugunta and Ferrara 2018).”

“Prior research has documented that a high fraction of active Twitter accounts are bots (17); that bots are responsible for much disinformation during election periods (18, 19); and that bots exacerbate political conflict by targeting social media users with inflammatory content (20).”

“Thus, in both platforms we can find a small population of celebrity users and the presence of social bots which in Twitter was estimated to be around 15% (Varol et al. 2017); by contrast, Mastodon profiles with spambot traits are more marginal, less than 5%.”

“This suggests that many of the accounts are either automated or semi-automated, or exhibit unusual tweeting patterns (Varol et al. 2017).”

“Furthermore, our findings on connotations behind deceptive content online will contribute to research on misinformation propagation in social networks by identifying gatekeepers and persistent minorities, e.g., trolls and bots that potentially spread deceptive content [8, 38, 60].”

“A very high percentage of bot accounts in social media, such as between 9 and 15% accounts (equivalent to 48 million accounts) in Twitter according to a recent study [6], makes these platforms highly vulnerable.”

“While there also exist “good” social bots which have no harm to others, 3 Varol et al. [11] report that between 9% and 15% of accounts on Twitter are fake.”

“The number of false negatives in our study most certainly extends beyond these accounts, with similar studies estimating false positives and false negatives in bot detection to hover at around 26% of the data, or 11% and 15%, respectively [13].”

“To battle against the bots’ army, authors in [15] first introduced Botometer (previously known as BotOrNot), the de-facto standard in terms of bot identification in Twitter, as detailed in Section III-C.”

“…media is by using social bots, algorithmically-controlled accounts that emulate the activity of human users but operate at much higher pace (e.g., automatically producing content or engaging in social interactions), while successfully keeping their robotic identity undisclosed [46, 65, 34, 84].”

“[63]mainly usemetadata, network, and other derived statistics to perform bot detection with random forest, aiming to detect different types of Twitter bots.”

“Six types of features are selected for the social-bot-detection process [15]: user meta-data, sentiment, friends, content, network, and timing.”

“A stream of research focused on specific behavioral analytics tasks, such as detecting specific behavioral patterns [6, 27, 38], predicting future behaviors [30, 55], and detecting anomalous ones [7, 11, 13, 45, 53, 54].”

“Bots can emulate all basic human activity on social media platforms, and they become increasingly more sophisticated as new advancements in Artificial Intelligence emerge [41,57,30,80,70].”

“…Spognardi, & Tesconi, 2017; Gilani, Farahbakhsh, Tyson, Wang, & Crowcroft, 2017; Oentaryo, Murdopo, Prasetyo, & Lim, 2016; Stieglitz et al., 2017; Varol, Ferrara, Davis, Menczer, & Flammini, 2017) and conceptual (Gorwa & Guilbeault, 2018; Grimme, Preuss, Adam, & Trautmann, 2017; Stieglitz,…”

“In Varol, Ferrara, Davis, Menczer, and Flammini (2017), authors have presented a complete framework for bot detection and have used an extensive set of features (1,150).”

“…estimates of the ratio of spammers to the total population of social accounts on Twitter (ranging between 4% in Benevenuto et al. (2010) and 15% in Varol et al. (2017)). supervised classification.650 As for MRF results, the average gain in performance (recall and precision) achieved by the…”

“Nowadays, bots can produce credible content with human-like temporal patterns [10–12].”

“determined by a bot-detection algorithm [28] to have been written automatically.”

“…likelihood of a given Twitter user being a bot, based on more than 1,000 features extracted from meta-data, patterns of activity, and tweet content (grouped into six main classes: user-based; friends; network; temporal; content and language; and sentiment) (Davis et al. 2016; Varol et al. 2017) 2 .”

“We have used three datasets from [4], [8], [18] together with our 140 different features to train a model using random forest.”

“In fact, it has been estimated that between 9 to 15 per cent of all users on Twitter are robots (Varol et al., 2017).”

“There are many statistical machines to learn these features automatically as bot detection methods, such as random forest (Varol et al. 2017; Gilani et al. 2017), AdaBoost (Abu-El-Rub and Mueen 2019; Andriotis and Takasu 2018), K-nearest neighbor (KNN) (Valliyammai and Devakunchari 2019) and…”

“It is worth noting again that Botometer does not implement any notion of activity sessions nor does it use any session-based features for bot classification [32].”

“Additionally, Twitter is notoriously populated by bots (Ferrara et al., 2016; Varol et al., 2017), automated accounts that are often used to amplify specific topics or view points.”

“while Dutt et al. [13] focus on the Facebook platform and analyze ads shared by Russian trolls in order to find the cues that make them effective. Finally, a large body of work focuses on social bots [5,10,17,16,46] and their role in spreading political disinformation, highlighting that they can manipulate the public’s…”

“Eliminating tweets by nonhuman entities (eg, social bots) is a challenge in itself given that 9% to 15% of Twitter accounts are bot accounts (equivalent to almost 48 million Twitter accounts), and these bots generate almost 35% of Twitter content [58].”

“Varol et al. [16] found that between 9% and 15% of Twitter accounts are bot accounts.”

“Second, because diversity is a property of the audience and not of its level of engagement, it is less susceptible to manipulation if one can detect inauthentic partisan accounts [38, 41, 42].”

“…to detect spams on Twitter can be classified into following classes: social network graph structure analysis [8, 26, 27], text structure analysis and patterns extraction [10], metadata analysis of user profiles and applied URLs [4, 16, 23] and interactive user behavior analysis [12, 17, 24, and].”

“…the score is calculated using the Botometer service, which assesses the extent to which the account exhibits characteristics similar to social bots [54] Botometer examines six main features: personal data, friends, social networks, temporal The Twitter account’s bot score is calculated using the…”

“Researchers have reported that 9–15% of all accounts on Twitter are social bots [18], and they have been found to be used for infiltrating political discourse, manipulating the stock market, stealing personal information, and spreading misinformation [19].”

“Varol et al. (2017) estimate that the bot population on Twitter may range between 9 and 15%.”

“We used the same annotation guideline of the Varol-2017 (Varol et al. 2017) to verify the likelihood of We used “X” to mask some users accounts and hide sensitive content T S Type Tweet 1 CC – NotBot @_PinealGland: 1984 https:// pic. twitt er. com/ Dgmni SvYON” one of the best books ever 2 HC – Bot…”

“They used four known datasets ICWSM17 (Varol et al., 2017), RAID11 (Liu et al., 2017), WWW17 (Yang et al., 2013), and Cresci-17 (Cresci et al., 2017).”

“Similarly, Ferrara (2017) investigates the disinformation campaign that has been coordinated by means of socialbots dis-guising themselves as legitimate human users during the 2017 French presidential election.”

“This is consistent with previous estimates of 10%-15% of all Twitter users being bots [22].”

“[12] estimated that between 9% and 15% of active accounts were bots (29–49 million accounts out of 328 millions3).”

“A large body of work focused on social bots [4, 10, 14, 15, 53] and their role in spreading political disinformation, highlighting that bots can manipulate the public’s opinion at a large scale, thus potentially affecting the outcome of elections.”

“Other work has shown that random forest models provide competitive performance across benchmark datasets (Varol et al. 2017), though we note that more complex models may provide slightly better performance for some contexts (Kudugunta and Ferrara 2018).”

“[122] find that compared with human users, the interaction selection of social bots is more”

“In the earlier studies on vaccines, social bots—automated accounts controlled and manipulated by computer algorithms [7]—have been accused of participating in vaccine discussions on social media [8,9], including public sentiment manipulation [10,11].”

“Due to their automation, social bots have the potential to manipulate social media discussions [2, 4, 5].”

“The first publicly accessible Twitter social bot detection system was called Bot or Not? and is discussed in (Davis et al. 2016; Varol et al. 2017a).”

“Table 1 illustrates some of the features that have been used in previous papers [6, 17, 18].”

“rms were exploited during the 2016 US Presidential Election [4,1,39,2], as well as numerous other elections [28,24,17,12,9,34] and other real-world events [11,13], by using tools like bots and trolls [14,8,37]. Campaign Detection: In order to combat misinformation, it is necessary to understand the characteristics tha…”

“DARPA organized a Twitter bot challenge to detect “in…uence bots” [39]. Among the participants, the work of [14], used similarity to cluster accounts and uncover groups of malicious users. „e work of [42] presented a supervisedframeworkforbotdetectionwhichusesmorethanthousands features. In a di‡erent a−empt, the work…”

“We may have had a higher proportion of social bots than that observed in everyday life, contributing to the tendency to identify them as human (Varol et al., 2017).”

“[35].”

“First, we estimate the mean values and deviations of the positive and sentiment scores retrieved from the WWW17 and ICWSM17 datasets, which aggregate a variety of user pro-files (humans and bots): a) Humans – genuine, humans , b) Bots – Traditional, Bots, Social, Fake .”

“Varol et al. (2017) have shown that between 9% and 15% of Twitter’s active accounts operating throughout this period were bots.”

“Existing research emphasizes that the online “ popularity ” of online sources can be easily faked through shares and retweets from inauthentic accounts, such as automated “ bots ” , manually operated “ sock-puppet ” accounts or semi-automated pro fi les (Keller et al., 2017; Monsted et al., 2017; Shao et al., 2018; Var…”

“Millions of bots, typically controlled by automated programs or platform APIs [1], attempt to sneak into genuine users as a disguise to pursue malicious goals such as actively engaging in election interference [11, 17], misinformation dissemination [8], and privacy attacks [37].”

“Traditional supervised learning methods such as Support Vector Machines (SVMs) and random forests have long been used to classify malware [43], [64], [5], [22], [50], detect network intrusions [48], [39], [18], [63], and identify fraudulent accounts [7], [54], [72].”

“Since fake news can be viewed as misinformation, work about detecting content polluters [26], social bots [47] and malicious campaigns [48] are also related to our work.”