Cognitive semiotic model for query expansion in question answering

Abstract

Query expansion improves performance of informational retrieval stage in question answering pipeline. We state the benefits of a personalized and autonomous query preprocessing and automate a semiotic model to achieve such properties. The model operates as a context-sensitive weighted grammar, along with the algorithm to apply production rules allowing approximate matching. The semiotic model is packed into a regression model to predict relevant terms for a query. ROC-analysis evaluates the regression model and helps to choose the optimal cutoff level. We compare ranking of terms by regression model and ranking based on an external informational retrieval system.