Combining statistical, structural, and linguistic features for keyword extraction from web pages

Keywords are commonly used to summarize text documents. In this paper, we perform a systematic comparison of methods for automatic keyword extraction from web pages. The methods are based on three different types of features: statistical, structural and linguistic. Statistical features are the most common, but there are other clues in web documents that can also be used. Structural features utilize styling codes like header tags and links, but also the structure of the web page. Linguistic features can be based on detecting synonyms, semantic similarity of the words and part-of-speech tagging, but also concept hierarchy or a concept graph derived from Wikipedia. We compare different types of features to find out the importance of each of them. One of the key results is that stop word removal and other pre-processing steps are the most critical. The most successful linguistic feature was a pre-constructed list of words that had no synonyms in WordNet. A new method called ACI‑rank is also compiled from the best working combination.