Scientific paper retrieval is usually framed as matching a short query string to relevant documents. A new arXiv paper instead treats it as document-to-document retrieval, where the input is a full paper and the goal is to find other papers relevant to that long-form query.
The authors propose Chain of Retrieval, a method that expands the search iteratively across multiple aspects of the query paper, then aggregates the results in a post-order manner. This is meant to capture the multi-faceted nature of a full paper, which may touch on several topics or methods at once.
The abstract is brief and does not include experimental results or comparisons to prior approaches. It only outlines the task and the proposed technique, so the effectiveness of Chain of Retrieval relative to existing methods remains to be seen.