\n\t\t\t\t\t\t
percentage: {(Math.round(Math.abs(v[1]) / sumsCatObj[activeCategory] * 10000) / 100)}%
*/}\n{ feedbackText }
\n \n \nAccording to
Bot Detective is a Web service that was developed in the context of the thesis of Maria Kouvela, in fulfillment of the requirements for the degree of Master of Data & Web Science, under the supervision of PhD Candidate Ilias Dimitriadis and Professor Athena Vakali. Bot Detective takes into account both the efficient detection of bot users and the interpretability of the results as well. Our main contributions are summarized as follows:
\n\t\t\t\ti. We propose a novel explainable bot-detection approach, which, to the best of authors’ knowledge, is the first one to offer interpretable, responsible, and AI driven bot identification in Twitter. Bot Detective uses a Random Forest classifier and follows an explainable AI state of the art method to provide justifications of high granularity, since explanations are offered for all relevant features that have contributed in the Bot Detective's scoring.
\n\t\t\t\tii. We provide a publicly available bot detection Web service which integrates an explainable machine learning framework along with users feedback functionality under an effective crowdsourcing mechanism. This service covers the growing demand for bot detection services and it offers extended crowdsourcing functionalities and XAI capabilities which advance existing state of the art tools, such as the Botometer.
\n\t\t\t\tiii. We share a new labelled dataset, annotated by exploiting Twitter’s rules and existing tools, which we use to build the proposed service. The dataset consists of thousands of tweets collected using Twitter’s official API and labelled with the use of Botometer and by taking into account that many of the authors of the posting accounts were, at a later time, deactivated by Twitter.
\n\t\t\t\tAlthough several AI driven bot detection methods have been implemented, the justification of bot classification and characterization remains quite opaque and AI decisions lack in ethical responsibility. Most of these approaches operate with AI black-boxed algorithms and their efficiency is often questionable.
\n\t\t\tBot Detective is an ongoing project of the Datalab at Aristotle University of Thessaloniki, Greece.
Team members: Maria Kouvela
Every time you subject an account on Bot Detective, the server needs to make requests to the Twitter API in order to collect the necessary information and for the model to make the predictions requested. Getting information from the Twitter API requires some special tokens, which are available to logged in Twitter users only. By logging in and granting Bot Detective permissions to access public data you provide your account's tokens to use in our service. The API requests are read-only, they make no changes to your account. Your private tokens are never uploaded to our server and thus we have no way of using them after you leave the website.
\n\t\t\tBot Detective produces a score on a 0-5 scale, based on the likelihood of an account being a bot. Higher scores are more bot-like. A score in the middle of the scale means that our Machine Learning model is uncertain about the classification.
\n\t\t\tOur service places emphasis on delivering explainable and comprehensive bot detection characterizations based on the principle that providing meaningful explanations is not just a matter of understanding or gaining user’s trust, but rather it is a matter of social responsibility, which is also imposed by EU laws as well. The goal of the explainability module is to convert the predictions of the classifier into justifications understandable by humans.
\n\t\t\tEvery time you subject a Twitter account on Bot Detective, a feedback button is presented on the screen along with Bot Detective's score and justification results. By providing us with feedback you help us improve our model. The information you provide is anonymous and no personal data is collected.
\n\t\t\tSince the content and posting activity of accounts in Twitter change over time, the results that Bot Detective outputs will also change. The server that runs in the background collects the 20 most recent tweets' information of an inputted account every time, that then the Machine Learning model uses to make its predictions.
\n\t\t\tMaria Kouvela, Ilias Dimitriadis, Athena Vakali
\n\t\t\t\t\tIlias Dimitriadis, Marinos Poiitis, Christos Faloutsos, Athena Vakali
\n\t\t\t\t\tIlias Dimitriadis, Konstantinos Georgiou, Athena Vakali
\n\t\t\t\t\tDescription: Tweets and their authors annotated using
Description: Twitter account IDs that refer to news agencies and companies, labelled as \"cyborgs\"
\n{get(response.data, \"description\")}
\n\t\t\t\t\t\t)}\n\t\t\t\t\t