Abstract:

Aiming at the of poor efficiency while processing a huge quantity of Web resources, particularly data reaources hidden in deep web by problem of current household textile resources from Web acquisition mode, an automatic approach to extract home textile resouces from Web was proposed. In this approach, a domain model was firstly proposed to identify deep Web query interfaces, then the identkfied query interfaces were filled automatically with domain keywords from household textiles, and the household textile resources from deep Web were extracted. In addition, in order to filter noises from response Web pages, pages were divided into different view blocks, a block importance model was proposed and trained by labeled blocks, and the model was utilized to filter the noise information independent from the subject. Experimental results show that in comparison with rule-based approaches, the domain model achieves 8% and 6% improvements in terms of positive predictive value and accuracy for query interface identification. Also, the block importance model achieves average 12.9% improvements at three levels in terms of harmonic average value for filtering noise information.

Key words: household textile, resource database, deep Web, information extraction