Hi, how should I go about querying eshop products via semantic search by multiple parameters? I considered 1. concatenating all parameters to one string and creating embedding from the string 2. doing a separate query for each parameter and them manually combining the results 3. create embedding from each parameter, and them join them (e.g. average the values) and querying the final embedding The only option possible directly in TS is 1), but I don't like it because if my parameters are e.g.

product_description

(text ~1500 chars) and

target_audience

(categorical male/female/unisex), the relative importance of the features is not preserved.

You can do number 2 via multi search but the aggregation and re ranking has to be done client side. My recommendation is to generate a shorter summary or even keywords from the description and then concatenate that with category and other metadata and use that for embedding.

Hi! I am new in Typesense, and I was exploring the semantic search. My site is a fashion aggregator platform, where I collect fashion clothes from different websites. I had very good results with keyword search and filtering, but not with semantic search...I cannot get accurate results, if I search for instance 'White dress for a wedding party' I got black dresses, even having the color as part of the embedding. But reading your last comment, maybe I should get keywords from the description instead of using several concatenated fields to build the embedding? Here is part of my code, I am using auto-embedded feature, and I have tried with different models including OpenAI but with no luck: { name: "embedding", type: "float[]", embed: { from: [ "name", "description", "color", "metadata", "product_categories", "brand", ], ... (I have also tried with a single field like 'description'...and with no good results either) Thank you so much.