I can see using LLMs to establish better embeddings to run the traditional pricing algo's (time-series, regression, decision-trees off), but it's not going to give you optimised elasticities on its own. Unless I'm grossly mistaken.
Interesting demo but forgive me I isn’t it just a fancy “look up” table ? I mean the predicted price is just checking if a descriptor is im a table or not and then applying a price accordingly? Apologies if I’ve missed the insight here. Appreciate any clarification cheers
Hi, Thanks for your comment. The look up table technique does not work as the items which are put in sale have a description which is given by a seller in free-form. Every item description is generally unique even if it’s the an identical product. Also sellers put many specific things and personalize the description. So the exact description in not available in past sales . Hence it’s required to use a LLM approach Hope this helps . Thanks for watching my channel
Very innovative indeed ! Thanks for creating this video
I can see using LLMs to establish better embeddings to run the traditional pricing algo's (time-series, regression, decision-trees off), but it's not going to give you optimised elasticities on its own. Unless I'm grossly mistaken.
Good point. LLM approach is useful when price depends on text description, which is relevant for market place scenarios
Interesting demo but forgive me I isn’t it just a fancy “look up” table ? I mean the predicted price is just checking if a descriptor is im a table or not and then applying a price accordingly? Apologies if I’ve missed the insight here. Appreciate any clarification cheers
Hi, Thanks for your comment. The look up table technique does not work as the items which are put in sale have a description which is given by a seller in free-form.
Every item description is generally unique even if it’s the an identical product. Also sellers put many specific things and personalize the description.
So the exact description in not available in past sales . Hence it’s required to use a LLM approach
Hope this helps . Thanks for watching my channel
@@DataScienceDemonstrated yes thanks - i missed that point....i'd better watch the video in full :)