A Lexical Search Model based on word association norms

摘要

This work introduces a lexical search model based on a type of knowledge graphs, namely word association norms. The aim of the search is to retrieve a target word, given the description of a concept, i.e., the query. This differs from traditional information retrieval models were complete documents related to the query are retrieved. Our algorithm looks for the keywords of the definition in a graph, built over a corpus of word association norms for Mexican Spanish, and computes the centrality in order to find the relevant concept. We performed experiments over a corpus of human-definitions in order to evaluate our model. The results are compared with a Boolean information retrieval (IR) model, the BM25 text-retrieval algorithm, an algorithm based on word vectors and an online onomasiological dictionary-OneLook Reverse Dictionary. The experiments show that our lexical search method outperforms the IR models in our study case.