Consistency of textual expression in newspaper articles: an argument for semantically based query expansion

This article investigates how consistent different newspapers are in their choice of words when writing about the same news events. News articles on the same news events were taken from three Finnish newspapers and compared in regard to their central concepts and words representing the concepts in the news texts. Consistency figures were calculated for each set of three articles (the total number of sets was 60). Inconsistency in words and concepts was found between news articles from different newspapers. The mean value of consistency calculated on the basis of words was 65 %; this however depended on the article length. For short news wires consistency was 83% while for long articles it was only 47%. At the concept level, consistency was considerably higher, ranging from 92 % to 97% between short and long articles. The articles also represented three categories of topic (event, process and opinion). Statistically significant differences in consistency were found in regard to length but not in regard to the categories of topic. We argue that the expression inconsistency is a clear sign of a retrieval problem and that query expansion based on semantic relationships can significantly improve retrieval performance on free text sources.