A hybrid approach for FAQ retrieval tasks

FAQ (Frequently Asked Question) collections are constructed to solve users' problems by providing related information in question-answer pairs. The task of searching a large FAQ collection can be viewed as the automatic retrieval of an FAQ entity that semantically matches a user query written in natural language. However, this task remains unsolved due to various difficulties. This thesis presents an approach comprised of three IR methods: the term expansion technique, differential latent semantic indexing (DLSI) approach, and template structure with related algorithms for solving FAQ retrieval tasks. This thesis is based on studying the advantages and drawbacks of traditional Information Retrieval (IR) techniques already applied to solving other IR problems, early research works related to retrieving FAQs, as well as the template approach that has been used successfully in language tutoring systems. Finally, by giving detailed methodologies of this hybrid approach for solving FAQ retrieval tasks, the effectiveness of the approach is evaluated through a series of experiments with a particular FAQ collection.