Good morning folks, I have a question regarding ho...
# community-helps
Sergio Behrends
05/15/2023, 11:17 AMGood morning folks, I have a question regarding how to fine-tune Typesense queries to have better results against plural words. We index product names, so for example, "_JBL Charge 5 Portable Bluetooth Speaker_". When searching for "Speaker" results are accurate, but when searching for "speakerS" (with an S) we are getting pretty bad results. I expected this to be handled as a Typo, but seems that something is off.
Sergio Behrends
05/15/2023, 11:18 AMmessage has been deleted
Sergio Behrends
05/15/2023, 11:18 AMWhen adding "Speakers" we first have listed some Documents, that have an exact match on "Speakers" but then none of the "Speaker" are listed.
Sergio Behrends
05/15/2023, 11:19 AMBy increasing
typo_tokens_thresholdSergio Behrends
05/15/2023, 11:20 AMSpeaker => 97 results
Speakers => 31 results
*Not expecting to be equal, but the difference is pretty big.
Sergio Behrends
05/15/2023, 11:21 AMSo far we implemented a one way synonym speaker -> speakers, to improve the results.
Sergio Behrends
05/15/2023, 11:21 AMIs there a better approach to this?
k
Kishore Nallan
05/15/2023, 2:14 PMHave you tried increasing typo tokens threshold further? I think we might something similar to max_candidates (used for prefixes) for typo candidates.
Kishore Nallan
05/16/2023, 2:35 AMIf you can share a subset of the dataset that reproduces this issue. I can dig into it further and see what we can do to address this.
s
Sergio Behrends
05/19/2023, 9:07 AMResults for speakerS improved after moving
typo_tokens_threshold Copy code
TV Sony 43" Class X85J
LG TV 48" Class C2 Series OLED evo 4K UHD
Samsung TV 43" Class Q60B QLED 4K
Samsung TV 65" Class The Frame QN65LS03BAFXZA Copy code
{
"q": "TVs",
"query_by": "title",
"page": 1,
"per_page": 10,
"exhaustive_search": true,
"min_len_2typo": 8,
"num_typos": 2,
"typo_tokens_threshold": 9
}k
Kishore Nallan
05/19/2023, 9:09 AMLook up min_len_1typo and min_len_2typo params
Kishore Nallan
05/19/2023, 9:09 AMFor shorter strings we don't enable typos because it becomes very noisy sometimes. But behaviour is configurable.
s
Sergio Behrends
05/19/2023, 9:09 AMI see
Sergio Behrends
05/19/2023, 9:10 AMMakes sense. So longer strings like speaker, are solved by the typo_tokens_threshold, and maybe for this short values I will stick with synonyms.
k
Kishore Nallan
05/19/2023, 9:37 AMCorrect
s
Sergio Behrends
05/19/2023, 10:04 AM Copy code
typo_tokens_threshold: 16,
min_len_1typo: 2,
min_len_2typo: 8,👍 1
Sergio Behrends
05/19/2023, 10:05 AM(A screenshot of an internal comparison tool that we built to compare to previous search engine plus to typesense query configurations)