Abstract
•Explore whether compounds can help improve Turkish IR systems.•Define the syntactic patterns used to index compounds for the Turkish Language.•Compare the impact of different compounds types on the retrieval performances.•Study the impact of state-of-the-art model on the results of the retrieval with compounds•Show that using compounds as index can improve retrieval performances.
In this article, we describe an empirical evaluation of compounds indexing for Turkish texts. We dive beyond the keyword indexing to propose a framework for Turkish compounds extraction and indexing. We identify twelve Turkish compounds pattern types that we classify in six categories. To extract Turkish compounds, we rely on a light natural language processing approach based on syntactic pattern recognition. We compare different compounds indexing strategies. We also investigate the effectiveness of using one compounds type and the effectiveness of combining different compound types. We conduct experiments over the Milliyet test dataset. The results of our experiments show that using compounds as index terms can improve retrieval performances. However, not all the compound types have a positive impact on the retrieval process.