Pseudorelevance

Pseudorelevance feedback, also known as blind relevance feedback, is an automatic query refinement technique used in information retrieval. It operates by assuming that the top-ranked documents returned by an initial query are relevant and then using their content to modify the query in hopes of improving later retrieval results. This approach provides a way to improve search effectiveness without requiring explicit user judgments.

How it works: after an initial search, the system analyzes the top K documents and extracts terms with high statistical weight. These terms are added to the original query or used to adjust its weighting, often in a manner similar to the Rocchio algorithm. The expanded query is then submitted to the search engine to produce a new ranking. The process relies on the premise that terms common in top results can help bridge vocabulary gaps and better capture the user’s intent.

Benefits and limitations: pseudorelevance feedback can improve recall by introducing related terms and concepts that the user did not originally specify. It is fast and does not depend on explicit feedback. However, it risks query drift if the initial top results are not truly relevant or are biased by the query itself, leading to degraded precision. The effectiveness is sensitive to choices such as the number of top documents (K) and the set of terms incorporated.

Relation to broader IR practice: it is a foundational baseline technique for automatic query expansion and a contrast to explicit relevance feedback from users. It is evaluated using standard metrics such as precision, recall, MAP, and nDCG on benchmark datasets. See also relevance feedback, Rocchio algorithm, and query expansion.