Log Posterior Approach in Learning Rules Generated using N-Gram based Edit distance for Keyword Search

摘要

Challenging searching mechanisms are required to cater to the needs of search engine users in probing the voluminous web database. Searching the query matching keyword based on a probabilistic approach is attractive in most of the application areas, viz. spell checking and data cleaning, because it allows approximate search. A probabilistic approach with maximum likelihood estimation is used to handle real-world problems; however, it suffers from overfitting data. In this paper, a rule-based approach is presented for keyword searching. The process consists of two phases called the rule generation phase and the learning phase. The rule generation phase uses a new technique called N-Gram based Edit distance (NGE) to generate the rule dictionary. The Turing machine model is implemented to describe the rule generation using the NGE technique. In the learning phase, a log model with maximum-a-posterior estimation is used to select the best rule. When evaluated in real time, our system produces the best result in terms of efficiency and accuracy.