Turkish FAQ retrieval system

The number of accessible information resources via internet is rapidly increasing. This rapid increase in information resources starts to be a problem in reaching appropriate data. Even though the search engines most of the time overcome this problem, for the internet users it is a time consuming and frustrating process to search though the results and to find out the necessary data. Because of this, the need for systems that can quickly and correctly provide answers for user questions is raised.The present thesis covers the performance evaluation of a Frequently Asked Question (FAQ) retrieval system that can retrieve the answer for a given natural language user question by matching it with the question answer pairs that are collected from FAQ web pages in Turkish. The aim is to retrieve the correct question answer pair as the first hit. To evaluate the performance, some of the questions from our data collection were selected and rephrased without changing the meaning and a test collection is created. The effect of Turkish language specific analyzing and the effects of searching the user question in different fields are investigated. It is found that Turkish language specific analyzing improves the retrieval performance of the system. Also, including the title of the source to a search field increases the performance of the system.