Consistency Learning and Multiple Rankings Combination for Text Retrieval

Text retrieval is one of the most basic tasks in the field of information retrieval. This paper deals with retrieving relevant documents for text-based queries from a database. Several different methods for retrieving text are explored, and show widely differing performance on different queries. It is shown how each of those methods may be improved through a “consistency learning” framework, where properties of the database and similarities on three different levels, namely documents, words and synonym sets, are exploited to improve performance. Further gains are achieved when all of the basic functions are combined in a metamodel to get better retrieval accuracy than each of the individual models.