FAQ-based information retrieval (IR) or question answering (QA) is a new trend
in IR area. An important problem is to detect the FAQs, which may be of different
forms and use different words. This problem cannot be solved by the existing NLP
tools. We suggest to exploit user logs or user interactions with the system
- as a complement for FAQ discovery. It is assumed that two queries leading
to the same document clicks are related. By exploiting this assumption, we have
been able to create interesting query clusters that form the basis of FAQ discovery.
User-log analysis can also be used to create a live thesaurus that connects terms
used by users with terms used in documents. Such a thesaurus can help binding
a user query with the relevant documents even though they use different words.
We will describe our exploration on these two tasks and present some experimental
results.