Pseudorelevantiefeedback

Pseudorelevance feedback, also called pseudo-relevance feedback, is a technique in information retrieval where the system assumes that the top-ranked documents returned by an initial query are relevant and uses them to refine the query. The goal is automatic query expansion that can improve subsequent retrieval without requiring explicit relevance judgments from users.

The typical process involves an initial search to obtain a ranked list of documents. A subset of the top results is treated as relevant for the purpose of query modification. Terms are extracted from these documents and reweighted, often using methods similar to TF-IDF or other vector-space measures. A new expanded query is formed by boosting chosen terms or by adjusting term weights, and a second retrieval is performed using this revised query. In many implementations, this follows Rocchio-style ideas, where the query vector is updated toward the centroid of the assumed relevant documents.

Benefits of pseudorelevance feedback include potential gains in retrieval performance, particularly in terms of recall and sometimes precision, especially when the initial query is sparse or ambiguous. However, the approach carries risks: if the initial top documents are not truly relevant, the feedback can drift the query away from the user’s intent, creating topic drift and degraded results. The method also adds computational overhead and depends on choices such as the number of documents considered and the terms selected for expansion.

Pseudorelevance feedback remains a common component of information retrieval research and is often contrasted with explicit relevance feedback, where users provide judgments. See also relevance feedback, query expansion, and the Rocchio algorithm.