Automated analysis techniques for online conversations with application in deception detection

摘要: Email, chat, instant messaging, blogs, and newsgroups are now common ways for people to interact. Along with these new sending, receiving, storing messages comes the challenge of organizing, filtering, understanding them, which text mining has been shown be useful. Additionally, it done so using both content-dependent content-independent methods. Unfortunately, computer-mediated communication also provided criminals, terrorists, spies, other threats security a means efficient communication. However, often textual encoding communications may provide possibility detecting tracking those who deceptive. Two methods understanding, deception in text-based presented. First, message feature uses features or cues CMC combined machine learning techniques classify according sender's intent. The method utilizes classification coupled linguistic analysis extraction number input features. A study deceptive non-deceptive email attained accuracy between 60% 80%. Second, speech act profiling is evaluating visualizing synchronous by creating profiles conversations their participants theory probabilistic Transcripts from large corpus annotated used train language models modified hidden Markov model (HMM) obtain probable acts sentences, aggregated each conversation participant set profiles. Three studies validating detailed as well two showing profiling's ability uncover uncertainty related deception. The introduced here that represent possible direction analysis. Both have applications outside context In addition aiding detection, applicable information retrieval, technical support training, GSS facilitation support, transportation security, assurance.