A Study On Detection Of Gender-deception In Computer-mediated Communication Based On DIA

This thesis makes an attempt to explore the preferential linguistic features to distinguisheach gender on the level of language in order to develop a linguistic tool to judge whetheran anonymous on-line communicator is acting the true gender or is pretending to be of afalse gender between the lines from the perspectives of discourse information analysis.In this field that has rarely been researched in China before, the author designed2steps to identify a feature set to predict the gender of an anonymous communicator. InStep1, the author conducted a questionnaire based on the former studies of genderpreferential language features in order to investigate male and female’s language featuresin computer-mediated communication according to their communicating habit, thusidentifying4features of usage of non-conventional punctuation, conventional usage ofpunctuation, utterance of final particles, personal information. It is further proved thatthough the four features can reflect gender preferences in using language, three of them:non-conventional punctuation, utterance of final particles, and revelation of personalinformation, cannot be used for detecting gender-deception in CMC, except for theconventional use of punctuation, as well as the use of full stop and question mark.In Step2, the author conducted a pilot study that explored the DIA features inrespect of information knot, information development, and information element. Thisexperiment involved20participants with10males and10females,5males pretending tobe female,5females pretending to be male, and the others keeping their true genders.Comparative analysis was given between the chat records. The experiment found theeffective DIA features set that can identify gender difference, and the DIA features setthat can detect gender-deception. It is found that amount of information levels,information knots on Level1,2,3,4,8,9, and10, information knots that have subsequentinformation knots on Level1,2,7,8,9, and information loss can distinguish whether thereis gender-deception in a CMC; information knots of WA, WF, and WO, and informationelement of Entity, Dative, Patient, Goal, and Affected are effective in both reflecting thegender differences and detecting whether a person is forging a false gender identity ornot.In the last part, comparative study of the15chat records collected in real lifeinvolving gender-deception from CLIPS, with the15normal CMC that don’t involve gender-deception were conducted using the features set. Thus further verify theeffectiveness of it.