I have a vector database using the text-ada-002 embedding model. Each vector is just text representing a service or a product.
e.g.
“massage”
“Lights and desk lamps”
“laptop”
etc
I am converting a product name/type into a vector and then using that as the whole query.
However, it seems like irrelevant vectors are being returned. When I search “Computer desk”, it’s returning this
{
id: '...',
score: 0.883229554,
values: [],
sparseValues: undefined,
metadata: {
text: 'Lights and desk lamps'
},
{
id: 'foobar-207',
score: 0.882196,
values: [],
sparseValues: undefined,
metadata: {
text: 'Computers, laptops, computer parts'
}
},
{
id: 'foobar-279',
score: 0.867525816,
values: [],
sparseValues: undefined,
metadata: {
text: 'Computer assembling or repair'
}
},
{
id: 'foobar-260',
score: 0.863857865,
values: [],
sparseValues: undefined,
metadata: {
text: 'non-adjustable tables, desk that does not raise to standing posit
}
},
{
id: 'foobar-278',
score: 0.859145403,
values: [],
sparseValues: undefined,
metadata: {
text: 'Computer parts for upgrade'
}
}
},
As you can see, most of these results are not really relevant. How can I improve the relevancy of the search results?