Pergunta

While studying relevance feedback(Pseudo relevance feedback), I have learn that the model can go horribly wrong for some queries. Can anyone give reasons why this is?

Foi útil?

Solução

The problem is also called query drift: if the top-k retrieved documents are all (or mostly) about a particular sub-topic, the importance of such sub-topic is boosted by the feedback mechanism.

A textbook example is with a query about "copper mines": if most of the retrieved documents are about "copper mines in Chile", the feedback process will drift the results towards documents on Chile.

Licenciado em: CC-BY-SA com atribuição
Não afiliado a StackOverflow
scroll top