News for hackers. Stuff that matters.

A comment on Open source Wikipedia search through vector db.

Re:Open source Wikipedia search through vector db by java-man on Thursday August 15, 2024 @02:27PM (#41258986)
A sample search shows it simply does not work.

Here is my example:

query: gazdanov gaito (a russian writer).

the first article with Score:0.7382957 is https://en.wikipedia.org/wiki/Gardno,_Gryfino_County

which has absolutely no relevance to the query. the main article does not even appear in the output.

[ Parent | on HN ]
Re:Open source Wikipedia search through vector db by ykhli on Thursday August 15, 2024 @02:41PM (#41259112)
Oh wow this is such an interesting case. I wonder if it's because of the embedding model used here.

If I search for Leo Tolstoy it works well, but "Amor towles" doesn't lead to the right page.

[ Parent | on HN ]
Re:Open source Wikipedia search through vector db by enesakar on Thursday August 15, 2024 @02:55PM (#41259231)
oh it looks it lists good results only when correct order and capitalized `Gaito Gazdanov`. we will check with other embedding models.
[ Parent | on HN ]
Re:Open source Wikipedia search through vector db by enesakar on Friday August 16, 2024 @12:07PM (#41267624)
hello. we fixed the issue, you can try again.
[ Parent | on HN ]