Calculating User Session Context
Posted: 28 Jan 2021
Date Written: November 5, 2020
Learning session context is important for query understanding and results ranking, and to reason about users, their session experience, and task goals. This talk will present a session context model and demonstrate a research POC that calculates in-session context for LexisAdvance legal research sessions. The system caches the evolving session and with each new user action the session events are re-partitioned using rules based on user-centered search semantics. Concepts extracted from queries and documents play a key role in the learned session representation. Functions are applied to the learned representation to calculate a local and overall session context. Other functions can revise system hypotheses about the user’s current experience, task, and task goal.
The session context module POC is a proposed component to provide input for query expansion and rewriting, personalizing results document representation. and predicting the user’s task domain. Future work will try to predict the next user action and aspects of the user task goal. Eventually, we want to build intelligent and cooperative information search systems. Session context understanding is part of that picture because it is critical for communication. In that light, session context calculations relate to dialog state tracking, conversational search, and automated negotiation of subtask responsibilities.
Keywords: user modeling, search process, AI, information system design, search interaction
Suggested Citation: Suggested Citation