A Predictive Search Method of FAQ Corresponding to a User's Incomplete Inquiry by Statistical Model of Important Words Co-occurrence

摘要

We address a predictive search of FAQ corresponding to a user's incomplete inquiry that a user is inputting with important words defined in each FAQ. The important words co-occur in a user's inquiries and the rates of the co-occurrences depend on which FAQ the user's inquiry corresponds to. The co-occurrence rates of important words in inquiries are estimated from a statistical model of important words co-occurrence generated with past inquiries and FAQ corresponding to them. When the highest co-occurrence rate of them is larger than a threshold set on each FAQ, the inquiry is regarded as a corresponding FAQ. Experimental results show that the proposed method can improve the recall rate by 40% for short inquiries and the precision rate by 27% for long inquiries.