A quadratic lower bound for Rocchio's similarity-based relevance feedback algorithm with a fixed query updating factor

Rocchio’s similarity-based relevance feedback algorithm, one of the most important query reformation methods in information retrieval, is essentially an adaptive supervised learning algorithm from examples. In practice, Rocchio’s algorithm often uses a fixed query updating factor. When this is the case, we strengthen the linear Ω(n) lower bound obtained in [9] and prove that Rocchio’s algorithm makes Ω(k(n − k)) mistakes in searching for a collection of documents represented by a monotone disjunction of k relevant features over the n-dimensional binary vector space {0, 1}n , when the inner product similarity measure is used. A quadratic lower bound is obtained when k is linearly proportional to n. We also prove an O(k(n−k)3 ) upper bound for Rocchio’s algorithm with the inner product similarity measure in searching for such a collection of documents with a constant query updating factor and a zero classification threshold.

• Type: Journal
• Year: 2010
• Where: JCO
• Authors: Zhixiang Chen, Bin Fu, John Abraham
• Keywords: Algorithm, Inner Product Similarity, JCO 2010, Product Similarity Measure