Grouping similar product descriptions together

Models
1

I have data with multiple descriptions of the same product, but they may be worded slightly differently, contain different product ids etc. I would like to group them together Would an approach here be

  1. Train a classifier on the products to predict say product category
  2. use the description embeddings to compute cosine similarity. And then
  3. group by some similarity threshold?

Is there a better approach to this?