Second retrieval approach
Besides the already implemented retrieval approach, which is based on finetuning an embedding model using MNRL, it may be interesting to investigate, how a Bi-Encoder/Reranker, which uses the input texts as queries and the textual descriptions of the class categories as targets performs compared to the regular finetuning.
--> To implement training pipeline and run evaluation experiment.
Second retrieval approach
Besides the already implemented retrieval approach, which is based on finetuning an embedding model using MNRL, it may be interesting to investigate, how a
Bi-Encoder/Reranker, which uses the input texts as queries and the textual descriptions of the class categories as targets performs compared to the regular finetuning.--> To implement training pipeline and run evaluation experiment.