When transitive scoring is enabled in Semarchy, the platform evaluates all possible connections between records in a match group—not just direct matches defined by rules, but also indirect (transitive) links through other records. For every pair of records in the group, Semarchy computes both the direct score (if a match rule connects them) and all possible transitive scores (via other records), then selects the highest score—whether direct or transitive—for that pair.
If the transitive score between two records is higher than their direct score, the transitive score is used. This means that the confidence score of the match group is calculated as the average of the highest available match scores (direct or transitive) for all unique pairs in the group. In effect, two records that are weakly matched directly, but strongly connected through other records, will have their stronger transitive score reflected in the group's confidence score.
To summarize:
Semarchy uses the highest score (direct or transitive) between each pair when calculating the group confidence score under transitive scoring.
This approach can result in a pair’s transitive score overriding its direct score if the transitive path provides a stronger link.
The group score is then the average of these best scores for all pairs in the group.
This mechanism enables more accurate and intuitive grouping, especially in complex data sets where relationships may not be captured by direct rules alone.
Hyewon Kim
Our current model does not use transitive scoring but I am curious how match and merge decisions would be like with it. I would like to know what happens when a transitive score is higher than its direct score for a matching pair. It could be that there is no match rule connecting the two records but they match indirectly through other records in the match group, or the two records match directly but somehow the transitive score is higher than its direct score. Either way, which score will Semarchy use to calculate confidence score of the match group?