next up previous
Next: Common IR Techniques Up: High Level Overview of Previous: Query Expansion and Relevance

Term Weighting and Term Selection

Relevance Feedback techniques tend to focus on one of two emphases:

Term Weighting
is the process by which terms in the query are weighted or re-weighted.
Term Selection
is the process by which terms are added, or in some case deleted, from the query.

In this paper, term weighting will refer to techniques that weight one term differently from another. A related concept is relative term weighting. Relative term weighting is a technique whereby older terms as a whole are weighted differently than newer terms.

Term selection will refer to techniques that select which candidate terms, all of which have a non-zero weight associated with them, are added or deleted from the query. Typically, term selection is implemented by using a term ranking formula to sort all candidate terms, and then selecting a set number of top candidates. In the literature, this implementation is often referred to as using alternate sorts for term selection.



Erik Selberg
Wed Aug 6 12:24:17 PDT 1997