Information extraction from structured documents using k-testable tree automaton inference Export

@article{citeulike:684051, abstract = {Information extraction (IE) addresses the problem of extracting specific information from a collection of documents. Much of the previous work on IE from structured documents, such as HTML or XML, uses learning techniques that are based on strings, such as finite automata induction. These methods do not exploit the tree structure of the documents. A natural way to do this is to induce tree automata, which are like finite state automata but parse trees instead of strings. In this work, we explore induction of k-testable ranked tree automata from a small set of annotated examples. We describe three variants which differ in the way they generalize the inferred automaton. Experimental results on a set of benchmark data sets show that our approach compares favorably to string-based approaches. However, the quality of the extraction is still suboptimal.}, author = {Kosala, Raymond and Blockeel, Hendrik and Bruynooghe, Maurice and Van den Bussche, Jan}, citeulike-article-id = {684051}, citeulike-linkout-0 = {http://dx.doi.org/10.1016/j.datak.2005.05.002}, citeulike-linkout-1 = {http://www.sciencedirect.com/science/article/B6TYX-4GCX0DW-1/2/dd1de233a6a5c9e68f90a2a5491b85aa}, doi = {10.1016/j.datak.2005.05.002}, journal = {Data \& Knowledge Engineering}, keywords = {inaki, wrapping}, month = {August}, number = {2}, pages = {129--158}, posted-at = {2006-06-05 08:36:57}, priority = {2}, title = {Information extraction from structured documents using k-testable tree automaton inference}, url = {http://dx.doi.org/10.1016/j.datak.2005.05.002}, volume = {58}, year = {2006} }

TY - JOUR ID - citeulike:684051 L3 - citeulike-article-id:684051 N2 - Information extraction (IE) addresses the problem of extracting specific information from a collection of documents. Much of the previous work on IE from structured documents, such as HTML or XML, uses learning techniques that are based on strings, such as finite automata induction. These methods do not exploit the tree structure of the documents. A natural way to do this is to induce tree automata, which are like finite state automata but parse trees instead of strings. In this work, we explore induction of k-testable ranked tree automata from a small set of annotated examples. We describe three variants which differ in the way they generalize the inferred automaton. Experimental results on a set of benchmark data sets show that our approach compares favorably to string-based approaches. However, the quality of the extraction is still suboptimal. IS - 2 JF - Data & Knowledge Engineering EP - 158 TI - Information extraction from structured documents using k-testable tree automaton inference VL - 58 SP - 129 KW - inaki KW - wrapping AU - Kosala, Raymond AU - Blockeel, Hendrik AU - Bruynooghe, Maurice AU - Van den Bussche, Jan PY - 2006/08// UR - http://dx.doi.org/10.1016/j.datak.2005.05.002 ER -