Using Statistical Features to Find Phrasal Terms in Text Collections

In this work we investigate alternatives to automatically detect phrasal terms, defined here as phrasal verbs, phrasal nouns, phrasal adjectives or phrasal adverbs found in a text. The automatic identification of phrasal terms may have several applications in text processing systems. We approach this problem and present a novel approach for detecting phrasal terms in a collection of documents. Our solution is based on machine learning and uses statistical features of the word n-grams found in the documents. We also investigate the particular impact of adding phrasal terms in the retrieval model of a search engine when processing queries on several data sets. Our results show that we are able to discover valid phrasal terms with a small error rate, achieving detection results ranging from 70% to 94% in terms of F1. Furthermore, the discovered phrasal terms, when used to enhance search tasks, allow improvements in retrieval performance of up to 11% in terms of MAP when considering all queries, and up to 36% in terms of MAP when considering only the queries that contained the detected phrasal terms.