A key factor for the acceptance of 3D virtual world mataphor as human machine interface is the suppression of the strangeness factor, which appears due the perception of non natural behaviour in representations of users (avatars) and bots (NPCs). Such perception raises mostly due to subtle differences in expected behavior for a human being and observed behavior in virtual representations (avatars and NPCs). In this work we will call 'persona' the set of exhibited behaviors. Thus, a way of diminishing the strangeness valley is by supplying avatars with convincing personas. Human beings rule their personas with intermediation of their emotional traits. Interactions with environment and other beings supply stimuli that change the emotional state from the equilibrium situation and, consequently, change observed behavior. Idle emotional state can be determined through natural language communication corpora analysis. In this thesis we state that through the analysis of user interactions in social media it is possible to establish a model of personality and persona that can be transposed to their representation in 3D virtual worlds, so such representation exhibits behavior that minimize strangeness. The model uses recent interations of user in social media and his interactions inside the 3D virtual world to establish current behavior and is coherent with OCC paradigm. It is also demonstrated that such model is evolutive: by the continuous analysis of communication corpora the user personality behavior is updated and its persona is continually adjusted. Personality traits are analysed in the 'Big Five' model (neuroticism, agreeableness, conscientiousness, openness) and operationalized in the PAD model (personality, arousal, dominance). A personality field is built such as in the absence of inputs, emotional state lies in the center of this field. Inputs move state along PAD axes and cause changes in the observed behavior. Correspondence between avatar behavior changes and state changes of user regarding its emotional field enhance the compliance between expected and observed behavior, reducing strangeness.