Automating the Extraction of User Model Information from Consultation Dialogues Automating the Extraction of User Model Information from Consultation Dialogues

This thesis addresses a natural language processing problem posed in the context of so-called Web assistant systems aka live help systems. A recent feature added to a growing number of Web sites, such systems o er user support via text chat with human assistants. To adapt consultation to the individual user, long-term information about his or her skills and interests is collected in a user model. So far, this updating of user models has been a task performed manually by the assistants. The thesis speci es, designs, implements, and evaluates a software component to automate the user model acquisition task. Text phrases containing information about user skills and interests are (1) automatically highlighted in the consultation dialogues and (2) associated with semantic user model concepts. A requirements speci cation points out the particularities of human-human text chat communication. The particularities are considered in the choice of two suitable and feasible information extraction approaches: one keyword based and one partial parsing approach. The partial parsing approach is being designed to ful l the requirements, and then implemented. An evaluation of performance indicates the resulting system is well suited to facilitate manual user modelling, but not a reliable basis for a full automation yet.