Providing FAQ lists based on ontology

Researchers have been fascinated in FAQ (Frequently Asked Questions) management systems in recent years. These systems have reduced the cost of supporting productions. The goal of this research is to implement a Persian FAQ generator in PC domain based on ontology. There are several steps for implementation. First, all slangs are converted into formal Persian phrases. Second, our proposed similarity method measures similarity between questions. After creating question similarity matrix, similar questions are gathered in the same clusters. FAQ lists are created based on clusters which have the most number of members. Evaluations indicates that our proposed similarity matrix generates the best results in comparison with Lin, MCS, and LSA matrices. Moreover, our proposed clustering method provides better results compared to hierarchical complete-linkage, hierarchical single-linkage, and Kmeans methods.

I. Introduction

FAQ¹

Frequently Asked Questions

management systems have attracted a great attention in last decades. Nowadays, customer satisfaction and product support is the highest priority for manufacturers. They spend a lot of time and money for training assistants to support their products. Therefore, the so-called Frequently Asked Questions systems were developed that stored users' asked questions. Every user could read questions and find the answer relating to his problem. However, that was a hard work reading all the questions without any search tools. Then, a search tool added to these lists but the problem was retrieving irrelevant answers. Most FAQ sites, however, provide no effective mechanisms to assist the user, who either has to pan through a long list of FAQs or has to rely on the use of the rudimentary keyword method to find relevant questions and answers [21]. Hence, the need for an automated FAQ management system increased significantly. Current systems use semantic methods to increase retrieving accuracy. These methods help users to enter their query in natural language and omit searching throughout all questions in the database.