Maxim Gubin, Alexander Morozov
The article contains results of experimental study of an effect of different morphological analysis methods on quality of informational retrieval system search. It is commonly accepted that morphological analysis can significantly improve recall of information search. Especially this is important for inflective languages such as Russian language. We studied three public available modules which can be used for stemming or normalization. One of them uses dictionary based approach other two are rules-based. We use a system without any morphological processing as a basic line. Also we studied variant with manual suffix stripping using wildcards.
The RIRES search quality evaluation methodology was used. This methodology is similar to TREC methods and uses Russian document and query collections.
Our experiments show that morphological analysis significantly improves not only recall but also search precision. Very close characteristics of search quality was shown by all methods of morphological analysis and manual wildcards stripping.