Graph Grammar Based Web Data Extraction

Web data extraction becomes a hot topic after the invention of World Wide Web, because the large amount of information on the Web makes it challenging to retrieve useful information. Due to the diverse designs and presentations of information on different Web sites, it is hard to implement a general solution to extract data across different Web sites. This paper presents a novel method based on graph grammar to extract the same type of information from different Web sites without the need of training or adjustment. Our approach formalizes a common Web pattern as a graph grammar. Then, based on the visual layout and HTML DOM structure, a Web page is abstracted as a spatial graph that highlights the essential spatial relations between information objects. According to the defined graph grammar, a spatial parsing is performed on the spatial graph to extract structured records. We have evaluated our approach on twenty one different Web sites, and achieved the F1-score as 97.49% which shows promising performance.