Region-Based Ranking in Association Analysis for News Relation Discovery

Abstract

Using an association-based technique to find associations among news documents can obtain useful news relations. However, existing works may not detect meaningful relations since only single association measure was used to mine news relations. This paper presents a region-based ranking approach to selectively use different association measures for different ranking regions, towards improvement of the ranking mechanism for news relation discovery. To evaluate region-based ranking, the method is compared to the conventional ranking method, which has no region construction. As performance evaluation, the top-k results of each method are compared using rank-order mismatch (ROM). Compared to the non-region method, the region-based method can find meaningful relations among news with the average ROM improvement of 1.21% - 28.32% for confidence and 4.83% - 29.04% for conviction, respectively.