The importance of the WEB chat in discussion has been well known and used effectively in collaborative learning environment. In order to activate discussions, various functions such as identification of participants’ roles, motivation to participate, and resolution of depressed discussions are needed. This research proposes the function to calculate the ”values of favorable words’ influence” on real time, which are estimated by member of participants’ words and their transitive structures. In this paper, firstly we define the ”values of favorable wards’ influence” as an index for the active direction of discussion and the discussion model based on this index, secondly we introduce an architecture of the discussion supporting system based on this model and its performance and finally through some practical experiments we examine the validity and effectiveness of our discussion model.