Estimating document reading time based on apparent user reading activity

Abstract

Methods, systems, and media are shown for calculating an estimated amount of time needed to read a document. The estimate is based on historical session data. Reading sessions are identified in part by excluding sessions that include non-reading user behavior signals, such as adding text to the document, printing, or changing formatting. Reading sessions are further identified by generating session clusters based on shared attributes, and for each cluster, determining if a relationship exists between session time and the shared attribute. If a relationship exists, sessions are excluded if they don't fit the relationship. Once reading sessions are identified, coefficients for a formula are generated that are applied to an unseen document to estimate reading time. Each coefficient is associated with an element type. The coefficients are derived by iteratively analyzing documents with one unknown element type and subtracting out contributions of known element types.