Fast and large-scale unsupervised relation extraction

A common approach to unsupervised relation extraction builds clusters of patterns expressing the same relation. In order to obtain clusters of relational patterns of good quality, we have two major challenges: the semantic representation of relational patterns and the scalability to large data. In this paper, we explore various methods for modeling the meaning of a pattern and for computing the similarty of patterns mined from huge data. In order to achieve this goal, we apply algorithms for approximate frequency counting and efficient dimension reduction to unsupervised relation extraction. The experimental results show that approximate frequency counting and dimension reduction not only speeds up similarity computation but also improves the quality of pattern vectors.