LOKI: Reusing Custom Concepts in Interactive Analytic Workflows

Abstract

Natural language (NL) interaction enables users to be expressive with their queries when exploring data. Users often specify complex NL queries that involve a combination of grouping, aggregations, and conditionals of data attributes and values. Such queries are often reused several times by users during their analytical workflows. Existing systems offer limited support to save these bespoke queries as concepts that can be referenced in subsequent NL queries, leading to users having to respecify these queries repeatedly. To address this issue, we describe a system, LOKI that allows users to save complex and bespoke queries as reusable concepts and use these concepts in other NL queries and analytics tools. For example, users can save an NL query, ''show me the opportunity amount by customer for open opportunities'' in a sales dataset, as a concept 'followup customers' and reference this custom concept in a query such as ''show me the total opportunity amount for followup customers.'' A qualitative evaluation of LOKI indicates the usefulness of supporting the reuse of custom concepts across various analytical workflows. We identify future research directions around in-situ semantic enrichment and dynamic concept maps for data exploration.