Approximate summaries for why and why-not provenance

Abstract

Why and why-not provenance have been studied extensively in recent years. However, why-not provenance and-to a lesser degree-why provenance, can be very large resulting in severe scalability and usability challenges. In this paper, we introduce a novel approximate summarization technique for provenance which overcomes these challenges. Our approach uses patterns to encode (why-not) provenance concisely. We develop techniques for efficiently computing provenance summaries balancing informativeness, conciseness, and completeness. To achieve scalability, we integrate sampling techniques into provenance capture and summarization. Our approach is the first to scale to large datasets and to generate comprehensive and meaningful summaries. PVLDB Reference Format:Seokki Lee, Bertram Ludäscher, Boris Glavic. Approximate Summaries for Why and Why-not Provenance. PVLDB, 13(6): xxxxyyyy, 2020.

Measuring information gain using provenance

“…Computing G(𝐷 𝑡 , 𝐷 𝑟 ) is challenging because we have to capture both Prov ℎ and Prov 𝑤 and compare (each) value to compute G(𝐷 𝑡 , 𝐷 𝑟 ). To address this challenge, we will investigate various approaches for approximate computation [3,15,16] that achieves high efficiency with bounded errors.…”

“…Furthermore, we need to study what are semantically meaningful explanations and how to generate them. An interesting avenue for future work would be investigating the relationship between our method and provenance summarization [13,15], hypothetical reasoning via provenance compression using abstraction tree [7], and generating workflow provenance summaries [1]. The technique introduced in [13,15] computes pattern-based top-k provenance summaries that are semantically meaningful and concisely represent a large amount of provenance information.…”

Putting Things into Context: Rich Explanations for Query Answers using Join Graphs (extended version)

“…Some of them support provenance for aggregate queries using simplified models and query plan optimizations [27,31,37]. A number of recent papers have proposed summarization techniques to represent provenance approximately [2,32,33,39], or using summarization rules for better usability [3]. Factorized and summarized provenance in natural language [16,17] has also been studied for ease of user comprehension.…”

Provenance-based Data Skipping (TechReport)

“…Closely related are techniques for provenance summarization [9,30,38,57,58,63], intervention-based methods for explaining aggregate query results [79,80,86] and other approaches for generating explanations of outcomes [34,35]. Some of these techniques use declarative descriptions such as selectionpatterns [34,35,58,80]. However, the summaries produced by these techniques are typically not sufficient for our purpose, e.g,.…”