Orthogonal Query Expansion

TL;DR

This paper introduces a novel query expansion method that uses orthogonal queries with low similarity to the original query to better satisfy users' informational needs, especially when queries are poorly formulated.

Contribution

The paper proposes a new approach to query expansion based on orthogonal queries, which are related but have low similarity to the original query, improving search results for ill-posed queries.

Findings

Orthogonal queries intersect minimally with original query results.

The method improves search results over traditional keyword perturbation techniques.

Effective in handling poorly formulated user queries.

Abstract

Over the last fifteen years, web searching has seen tremendous improvements. Starting from a nearly random collection of matching pages in 1995, today, search engines tend to satisfy the user's informational need on well-formulated queries. One of the main remaining challenges is to satisfy the users' needs when they provide a poorly formulated query. When the pages matching the user's original keywords are judged to be unsatisfactory, query expansion techniques are used to alter the result set. These techniques find keywords that are similar to the keywords given by the user, which are then appended to the original query leading to a perturbation of the result set. However, when the original query is sufficiently ill-posed, the user's informational need is best met using entirely different keywords, and a small perturbation of the original result set is bound to fail. We propose a novel approach that is not based on the keywords of the original query. We intentionally seek out orthogonal queries, which are related queries that have low similarity to the user's query. The result sets of orthogonal queries intersect with the result set of the original query on a small number of pages. An orthogonal query can access the user's informational need while consisting of entirely different terms than the original query. We illustrate the effectiveness of our approach by proposing a query expansion method derived from these observations that improves upon results obtained using the Yahoo BOSS infrastructure.