Q2.3: Classically, approaches that use embedding of word used in information retrieval are based on a re-ranking process (i.e., the embeddings are used only to rerank the top-k results provided by a classical IR matching like BM25 for instance). a) Explain why "simple" approaches that represent documents using the embeddings of their words are not able to cope with the retrieval on large corpus of documents. b) Explain why reranking is then a solution to cope with this