We evaluated which settings for the reader and retriever return the best results.
We fine-tuned our reader on our own data created with Haystack's Annotation tool. Following we compared the default reader with the fine-tuned version. Additionally we used three different retriever in combination with the default and fine-tuned reader to determine which delivers the best performance, including the Dense Passage Retriever, the Elasticsearch Retriever and a joined version of both retrievers.
