1
$\begingroup$

I’m building a recommendation system where each user interacts with sessions (topics with a title and description). I want to represent each user using their last 5 session interactions by creating a user embedding.

My idea is to use a pretrained model to generate embeddings for the session titles and descriptions, then combine the embeddings of the last 5 sessions to create a user representation.

I’d like advice on:

  • How to combine multiple session embeddings effectively. Which pretrained models are best suited for this kind of text encoding.

I'm using titan-embedding-v2 and open for any other embedding models as well

Looking forward to your suggestions. Thank you!

Improve this question
$\endgroup$

1 Answer 1

Reset to default
0
$\begingroup$

Here are some pretrained models for text embedding:

  • Titan Embedding v2: If you're already using it, it's optimized for embeddings and should suffice if its output aligns with your task.
  • OpenAI's Ada-002: Highly versatile and cost-efficient for
    embedding tasks.
  • Sentence Transformers (e.g., all-MiniLM-L6-v2): Provides robust sentence-level embeddings, optimized for semantic similarity.
Improve this answer
$\endgroup$

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.