Extracting representative phrases from Wikipedia article sections

Nowadays, Wikipedia has become one of the most important tools for searching information. Since its long articles are taking time to read, as well as section titles are sometimes too short to capture comprehensive summarization, we aim at extracting informative phrases that readers can refer to. Existing work on topic labelling works effectively and performs well on document categorization, but inadequate for granularity of detailed contents. Besides, existing keyphrase construction methods just perform well on very short texts. So we try to extract phrases which represent the target section content well among other sections within the same Wikipedia article. We also incorporate related external articles to increase candidate phrases. Then we apply FP-growth to obtain frequently co-occurring word sets. After that, we apply improved features which characterize desired properties from different aspects. Then, we apply gradient descent on our ranking function to obtain reasonable weighting on the features. For evaluation, we combine Normalized Google Distance (NGD) and nDCG to measure semantic relatedness between generated phrases and hidden original section titles.