Scaling up Pattern Induction for Web Relation Extraction through Frequent Itemset Mining

摘要: In this paper, we address the problem of extracting relational information from Web at a large scale. particular present bootstrapping approach to relation extraction which starts with few seed tuples target and induces patterns can be used extract further tuples. Our contribution in paper lies formulation pattern induction task as well-known machine learning problem, i.e. one determining frequent itemsets on basis set transactions representing patterns. The mining is not only elegant, but also speeds up step considerably respect previous implementations procedure. We evaluate our terms standard measures seven datasets varying size complexity. particular, by analyzing rate (extracted per time) show that reduces complexity quadratic linear (in occurrences generalized), while mantaining quality similar (or even marginally better) levels.