Specific temporal association rules and temporal correlations to enlarge and detect inconsistencies in a large growing knowledge base

Large Knowledge Bases construction has been a relevant topic in the last years. Most techniques focus on discovering facts or relations like presidentOf(Bush, United_States). Many of them are associated with a specific temporal scope, thus, they are true only for a specific period of time. Therefore, a fact (e.g. presidentOf (Bush, United_States)) that was true in the past, might not be currently true anymore. Moreover, automatically (or semi-automatically) created large knowledge bases tend to have noise and inconsistencies, and trying to correct them is quite important. In order to deal with such issues, we propose two approaches: TARE and TCI. The first one introduces the concept of extracting specific temporal association rules and utilizes Conversing Learning to check for possible inconsistencies. Based on the specific temporal association rules extracted, TCI searches for temporal correlations between the instances in order to (i) populate the knowledge base with facts and (ii) to identify inconsistencies among the data. Experiments showed that the proposed approaches can help detecting inconsistencies as well as extending the knowledge base. In addition, TCI component was able to detect more possible inconsistencies than Conversing Learning, showing that TCI is an efficient method.