#Gift automation

3 messages · Page 1 of 1 (latest)

feral berry
#

Hey there,

I'm currently leading a project that is tasked with finding appropriate gifts for people based on a few descriptive inputs :like interests, hobbies, budget, age, gender, bio, relationship, etc.

I currently don't have an official dataset, but I'm going around scraping different websites and their items - think like Amazon or aliexpress, but in much smaller quantity (around 1k unique items)

After getting this data though, I'm not sure what my next step ideally should be.

My mind is hovering around using an LLM go generate tags or something like a perfect profile (a perfect profile is basically a fake user that has hobbies, budget, age, gender, bio, relationship but is optimized by the LLM go be perfectly suited for the item) per item. Then, I would embed that data into a vector DB, and for real queries, I would query in the vector DB for similar profiles or tags and retrieve the closest result.

I would greatly appreciate any suggestions or techniques I can use to better optimize this process or rethink it altogether.

Thank you very much for your time ❤️

clear sage
#

I would suggest make sure your scraped data is clean and organized, remove duplicates, fix any missing values, and standardize things like prices and categories. Adding extra details, like tags or descriptions, will make your recommendations more accurate.

Using an LLM like GPT to generate tags from item descriptions is a nice idea. You can fine-tune the model or use clever prompts to get high-quality results. If you’re looking for something more efficient, clustering items by features like price or category can also work well.

feral berry
#

I think since my long term goal is to have many more than just 1k items like now, it would make sense to create sub groupings, where a grouping would represent an item of some kind and I would have many copies of that item that are slightly different.

For example a grouping for video games : than inside that grouping I would have specific games, like Minecraft, fifa, etc