Hello everyone,
I am conducting some text retrieval experiments with MySQL 5.5.3.
I have a question regarding WITH QUERY EXPANSION search mode.
In this presentation [1] I learned that it applies blind relevance feedback. It runs natural language mode query twice and add top 20 results (controlled by ft_query_expansion_limit variable default 20) to original query.
Manual says: "concatenated with the few most highly relevant documents"
And my question is: Does it append/concatenate whole rows to the original query? so that second query becomes very long (original query + 20 rows)?
Or it selects some important words inside rows?
If someone reply this I will be very happy.
I am asking this because I tried to verify this as follows:
I set ft_query_expansion_limit to 1. Execute natural language query and get first result. Append first row to initial/original query manually. And then re-query this long query using natural language mode.
When i compared this results with (initial/original query with query expansion mode), the results were quite different.
I also tried to ignore initial query so that i queried first document, results were again different.
I am simply trying to re-produce the same results of With Query Expansion using/querying natural language mode query twice.
Than you for your consideration.
[1]
http://forge.mysql.com/w/images/c/c5/Fulltext.pdf