Extracting Chinese product features: Representing a sequence by a set of skip-bigrams

A skip-bigram is a bigram that allows skips between words. In this paper, we use a set of skip bigrams (a SBGSet) to represent a short word sequence, which is the typical form of a product feature. The advantage of SBGSet representation for word sequences is that we can convert between a sequence and a set. Under the SBGSet representation we can employ association rule mining to find frequent itemsets from which frequent product features can be extracted.For infrequent product features, we use a pattern-based method to extract them. A pattern is also represented by a SBGSet, and contains a variable that can be instantiated to a product feature.We use two data sets to evaluate our method. The experimental result shows that our method is suitable for extracting Chinese product features, and the pattern-based method to extract infrequent product features is effective.