Phrasal Terms in Real-World IR Applications

In this chapter we report our investigation on one important issue in the real-world IR environment, i.e., the usefulness, extraction and usage of phrasal terms. One large-scale empirical study has provided supporting evidence that phrasal terms can improve retrieval effectiveness, especially when their relative proximity information is understood from the naturally running text. To automatically identify significant terms for a predefined topic, we have adopted a “gaining data from data” approach. The algorithm learns to select candidate terms through a meaningful comparison of a focused sample with a large and diverse base sample. When investigating whether the identified terms can be useful for other IR applications, we applied these knowledge resources for document summarization and classification. The initial results look quite promising.