Abstract: Label extraction and matching arc an important part of the query interface understanding. A vision-based label extraction and matching approach was proposed in this paper. First, the factors which affect label matching were deeply analyzed, and then, a method of reconstructing query interface by analyzing its html code was given correspondingly which can restore the visual layout of form effectively. Finally, the clement label matching was realized which comprehensively considers label tag,text semanteme and position feature. Experiments on 277 query interfaces in 8 domains demonstrate the feasibility of our proposed approach.

Key words: Label extraction, Position feature, Form layout, Element label matching