Winner of the Microsoft MARCO track at HopHacks 2019.
MarcoSearch is an efficient way to get multi-perspective information on a topic by extracting information through multi-lingual online resources and translating them back to English.
MarcoSearch works as follows:
- Translate the target query into a set of languages.
- Run a language-specific Google search on the translated set of queries.
- Translate the pages back into English.
- Obtain Microsoft's Genint embeddings for the original search query and all the translated documents.
- Rank the websites based on a similarity score between the query embeddings and the document embeddings.
- Generate the top N results.