Mining the web for reciprocal relationships

In this paper we address the problem of identifying reciprocal relationships in English. In particular we introduce an algorithm that semi-automatically discovers patterns encoding reciprocity based on a set of simple but effective pronoun templates. Using a set of most frequently occurring patterns, we extract pairs of reciprocal pattern instances by searching the web. Then we apply two unsupervised clustering procedures to form meaningful clusters of such reciprocal instances. The pattern discovery procedure yields an accuracy of 97%, while the clustering procedures indicate accuracies of 91% and 82%. Moreover, the resulting set of 10,882 reciprocal instances represent a broad-coverage resource.