Resolving Seemingly Conflicting Fact Statements Caused by Missing Terms

摘要

The Web has become the most useful resource for people to acquire information. However, it still contains much incorrect and inaccurate information. While the problem of resolving conflicting structured Web data has been studied extensively, the corresponding problem involving unstructured text data has rarely been ad-dressed. In this paper, we consider how to resolve seemingly conflicting fact statements (SCFS). A statement is a fact statement if it attempts to state a fact, although the stated fact may not necessarily be correct. In this paper, multiple fact statements are SCFSs if they are identical except that they have different terms at a certain (same) position. In general, SCFSs are not necessarily truly in conflict. Some conflicts may result from imprecise statements and may be resolved by improving the precision of the statement. In this work, we focus on a special type of SCFSs whose corresponding statements are imprecise because they have missing terms. More specifically, given a set of SCFSs, we propose a method to mine the possible missing terms. This method effectively transforms an imprecise fact statement to precise ones. At the same time, our method also resolves the seemingly conflicts among the transformed fact statements. The experimental results show that our best solution for missing term mining on different data sets can achieve close to 90% F-score, which improves a baseline method by about 40%. We also propose a method to place the mined missing term for each SCFS at the right place in the statement and our experiments show that this method has an accuracy of about 73%.