Abstract: It is a recognized fact that human beings usually tend to compare one thing with the other which is typical part of human decision making process. But, this is difficult to know; what are to be compared and what can be the alternatives. This alternatives foundation process is very much difficult to cipher out what user really wants. This comparison activity is very common in our daily life but requires high knowledge skill. We present a novel way of automatically mining comparable entities from comparative questions in order to address this difficulty overcoming some performance issues of existing system. The results will be very useful in helping user’s exploration of alternative choices by suggesting comparable entities based on other users’ prior requests. The proposed system overcomes these drawbacks and improves the efficiency of mining comparators. Therefore, in order to assure high precision, recall and ambiguity resolution, we propose a hybrid approach that combines the Bootstrapping, heuristic rules, pattern matching, and association rules combination. Association rules gives Support and Confidence count of each comparable entity pair aims to extract frequently compared entity pair, interesting correlations from archived questions. The primary goal of this project is to identify comparative question and then we pull out its comparable entities from the query by making use of an extensive question record. Initially, the user presents a query as an input; later the system will identify whether the fired question is comparable or not. Once the system verifies that the query is comparative; the required entities are extracted, and the output is presented to the user with the possible alternative options along with their ranking in repository. This approach provides better results compared to the existing approach.

Keywords: Information extraction, bootstrapping, sequential pattern mining, comparable entity (comparator) mining, POS tagger and association rule.