[Paper Review] Why so? or Why no? Functional Causality for Explaining Query Answers

Alexandra Meliou, Wolfgang Gatterbauer+2 more|ArXiv.org|Dec 29, 2009

Scientific Computing and Data Management15 references20 citations

TL;DR

This paper introduces functional causality as a unified framework for explaining both query answers ('Why so?') and missing results ('Why no?') in databases. By refining Halpern and Pearl's actual causes with functional causes, it enables graded responsibility ranking of data tuples and query operations, supporting provenance, non-answer explanations, and aggregate contributions with polynomial-time algorithms for tractable cases.

ABSTRACT

In this paper, we propose causality as a unified framework to explain query answers and non-answers, thus generalizing and extending several previously proposed approaches of provenance and missing query result explanations. We develop our framework starting from the well-studied definition of actual causes by Halpern and Pearl. After identifying some undesirable characteristics of the original definition, we propose functional causes as a refined definition of causality with several desirable properties. These properties allow us to apply our notion of causality in a database context and apply it uniformly to define the causes of query results and their individual contributions in several ways: (i) we can model both provenance as well as non-answers, (ii) we can define explanations as either data in the input relations or relational operations in a query plan, and (iii) we can give graded degrees of responsibility to individual causes, thus allowing us to rank causes. In particular, our approach allows us to explain contributions to relational aggregate functions and to rank causes according to their respective responsibilities. We give complexity results and describe polynomial algorithms for evaluating causality in tractable cases. Throughout the paper, we illustrate the applicability of our framework with several examples. Overall, we develop in this paper the theoretical foundations of causality theory in a database context.

Motivation & Objective

• Address the need for unified explanations of both positive and negative query results in database systems.
• Overcome limitations in existing provenance and non-answer explanation models by introducing a formal, extensible causality framework.
• Enable graded responsibility scores for data tuples and query operations to rank causal contributions.
• Provide a theoretical foundation for causality in databases that generalizes and extends prior work on lineage and missing results.
• Support explanations for aggregate functions (e.g., SUM, COUNT) by quantifying individual contributions through responsibility metrics.

Proposed method

• Adapt Halpern and Pearl’s actual causality definition to address its shortcomings in database contexts.
• Introduce functional causes as a refined causality definition with desirable theoretical properties (e.g., compositionality, minimality).
• Use potential functions to model causal influence, enabling analytical manipulation and generalization to probabilistic data.
• Define responsibility as a quantitative measure of causal contribution, allowing ranking of causes.
• Formulate causality evaluation as a constraint satisfaction problem over Boolean networks, with tractable cases solvable in polynomial time.
• Apply the framework to query plans and data relations, enabling causal explanations for both input data and relational operations.

Experimental results

Research questions

• RQ1 How can causality be formally defined in a way that unifies explanations for both query results and missing results?
• RQ2 What properties must a causality definition satisfy to be useful in a database context, especially for provenance and aggregates?
• RQ3 Can causality be quantified with graded responsibility scores to rank contributing data tuples or operations?
• RQ4 Which classes of database queries allow for efficient (polynomial-time) causality evaluation?
• RQ5 How can the framework be extended to handle probabilistic data and aggregates?

Key findings

• Functional causality provides a robust, unified framework that generalizes and improves upon prior provenance and non-answer explanation models.
• The framework supports explanations for both the presence ('Why so?') and absence ('Why no?') of query results using the same causal definition.
• It enables ranking of causes by responsibility, with concrete examples showing decreasing responsibility for contributing keywords in a news feed.
• For safe queries and tractable Boolean networks, causality evaluation can be performed in polynomial time.
• The use of potential functions allows for analytical manipulation and paves the way for future extensions to probabilistic databases.
• The framework successfully models causal contributions to aggregate functions such as SUM and COUNT, assigning measurable responsibility scores to individual tuples.