I'm junior data analyst.
I'm looking for method for Sentence-> word recommendation. For example, if I input 'the little mermaid' and book's introduction(sentence), the model can put out 'swim suit' or 'fish doll'.
My knowledge about NLP is beginner level, I even didn't know about BERT. Can I make that kinds of model through BERT or other LLM?
and I don't have any idea of what keyword should I search for. I ask for your advice.
thank you.
There are multiple ways to solve recommendation problem. The standard way as now is to vectorize the list of sentences that you are searching through them for the recommendation task using SBert and store them somewhere like in a vectordb. then when you get your query sentence vectorised that as well and use a vector similarity metric to retrieve the top most similar items.
This is the most efficient way however, it is not the only way. If you need to get a better model with Bert you have to find-tune it. You can also ignore vector db and train a model that runs through your sentences again!!
Similarly LLMs can be used for encoding into vectors as well. But they are usually too expensive for the job.
