The tf-idf is a well-known weighting measure for words in texts. It measures both the frequency and the locality of words. It is often used for information retrieval and text mining. However, a lot of infrequent words have the same tf-idf value. In this study, the words are noun phrases. This paper proposes a novel weighting measure for noun phrases in texts by using the local frequency of nouns that construct a noun phrase. The proposed measure is calculated by combining the tf-idf of a noun phrase and the average of the difference between its frequency and the frequency of nouns within the phrase. The proposed measure was evaluated in experiments on the datasets of 19,997 newsgroup texts written in English and 206 Wikipedia pages written in Japanese. The experiments showed that the number of noun phrases with the same proposed measure is less than the number of noun phrases with the same tf-idf.
