Lexical Normalization and Relationship Alternatives for a Term Dependence Model in Information Retrieval

摘要

We analyze alternative strategies for lexical normalization and term relationship identification for a dependence structured indexing system, in the probabilistic retrieval approach. This system uses a dependence parse tree and Chow expansion. Stemming, lemmatizing, and nominalization processes are tested as lexical normalization, while head-modifier pairs and binary lexical relations are tested as term relationships. We demonstrate that our proposal, binary lexical relations with nominalized terms for Portuguese, contributes to the performance improvement in information retrieval.