Many years ago, we embraced deep learning for dropping feature engineering, now we have prompt engineering 🤦 then IR is back with RAG and query expansion. Seems there isn't much difference between fashion trends and research trends in the way that they always come back 😅
📃Advanced RAG: Query Transformations
We've added a few advanced RAG methods over the past few days that all fall under a similar abstraction: "Query Transformations"
✒️Rewrite-retrieve-read
🧨RAG-Fusion
🏃Step-back prompting
We've written a blog on this "Query Transformation" abstraction, detailing how we see the space
🔎It's worth noting that this not new - search engines like Google have been doing query expansion for years
💬What is new is using LLMs to do it. All of the above methods (and others outlined in the post) uses an LLM to transform a query into a different (or multiple) queries
The main difference is the prompt used!
Full blog: https://t.co/3ZpNPGeaFo
Interested to do a PhD at @ir_glasgow on information retrieval and/or recommender systems? Please consider this opportunity; The School is offering upto 17 studentships this year; do apply early; deadline is 31st Jan
Just over a week left to apply to our @GlasgowCS postdoc position on unbiased learning to rank, counterfactual and reinforcement learning in search & #RecSys - Deadline for applications is Wed 2nd Sep.
Need NLP?
Give #SparkNLP 2.0 a try!
Why?
It makes half the errors that SPACY makes on named entity recognition (NER). Besides, It obtained the best performing academic pre-reviewed results. It is also faster than SPACY to train #NLP models locally.
https://t.co/W4Ssy0iqO4