Likely stories: artificial intelligence, media bias and marginalised audiences

点击翻译按钮获取中文摘要

Can the techniques and distribution models for astroturfing individuals and communities of manipulation be enumerated? Are there mechanisms at the social structural level that can be identified though analysis of online profile behaviour, and of implied and actual metadata - not only through the social graph, but the content of communication? Are there linguistic, time-based or communicative fingerprints that identify sock-puppet accounts within other communities? Can these online identities be recognised or surfaced through machine intelligence and pattern recognition techniques? Is there a taxonomy of metadata around online social communications that can help understand or profile types of participant?The research will aim to identify these criteria by analysing published online data (using sources such as Twitter, Reddit, and other large-scale online platform with a social element) and analysing the interactions, connections, frequency and linguistic patterns of participants to establish if there are any meaningful inferences behind this information, using datascience techniques. Various models and approaches will be run on the same data, and also then compared with how human observers see these types of interaction and how they interpret them. This data can also be used to verify the utility of these approaches against online identities and communications with known intents. From this, we can start to identify likely taxonomies and useful criteria by which we can begin to make assessments about the correlation of online messages with types of behavioural profile to feed a training set and potentially provide the underpinning for machinebased tools to assist individuals or platform providers in making calls about the intent and health of online communities and individual profiles.