startthat.work home

IS-04 Data fusion with conflict resolution (majority voting; deciding and mediating resolution functions)

Combines records on the same entity into one record, resolving conflicting values by a rule, and produces a fused record per entity with uncertain and contradictory values marked.

Integrate across sourcesEvidence: DocumentedPreparation technique, runs in step 5.2Strongest statement: describe

What it does

Combines records on the same entity into one record and resolves conflicting values by a rule: choose an existing value (deciding) or compute a new one (mediating). Majority voting is the simplest rule. Artefact: a fused record per entity. Read which values were uncertain (a value against a null) or contradictory (different non-null values).

When to use it

Use when sources give different values for the same entity and you need one value with the conflicts shown.

When it does not fit

Fits when entities are reliably identified. Misfit: mis-linked records or dependent sources. Voting suits only independent sources of similar accuracy. Text sources need claims extracted first, which adds uncertainty.

Result format

The result is one fused record per entity, with the rule used for each field named as deciding or mediating. It marks each value as agreed, uncertain (a value against a null) or contradictory (different non-null values).

Requirements on the material

PropertyRequirement typeRequirement
StructureApplies in some casesrequires=structured; requires=semi-structured
Content typeAlways appliesrequires=tabular; requires=text
Unit of analysisAlways appliesrequires=record
Level of detail availableAlways appliesrequires=record-level
Number of sourcesAlways appliesrequires=two or more
Whether sources can be joinedAlways appliesrequires=shared identifier available
Known data problemsApplies in some casesrequires=inconsistent
Source dependenceChanges the orderprefers=independent
Appraises source credibilityApplies in some casesno

Not restricted on: Variable types, Time structure, How the sample was chosen, Amount of material, Producer interest, Origin, Document genre, Language, Personal information, Security classification, Licence.

Minimum materialnone
Additional sources needednone
Not for (rules)none
Not for (text)Records may be mis-linked or sources depend on each other. Majority voting suits only independent sources of similar accuracy, and text sources need claims extracted first.

Choosing and ordering

Kind of techniquePreparation technique, runs in step 5.2
Phase tagverify and prepare
Other phase tagsnone
Kind of findingcontradiction; gap
Question answeredwhat
Strongest statement it supportsdescribe
Needs as input
Makes as output
Must exist firstNone
Loop targetproceed with limit stated

Library checks by stakes

Each stakes level includes the checks of the level below it. Checks that need a person apply at high stakes only, so the process leaves them out at lower stakes (decision 38). Repeatability: deterministic.

StakesCheckPerson involvedCounts as a different method
exploratoryinput validationnoneno
decision-informinginput validationnoneno
decision-informingspot checkoptionalno
high-stakesinput validationnoneno
high-stakesspot checkoptionalno
high-stakesindependent reviewneededno
high-stakesprovenance lognoneno

Failure modes and limits

Model suitabilitymodel plus code
Typical model failureTakes the most common value when sources copy each other, or the most recent value without checking it.
Error sourcesinput data; data transformation; statistical assumptions
Certainty limitsCertainty drops with mis-linked entities, dependent sources and values that change over time (inconsistency and risk of bias).

Evidence and tag basis

Documented. Truth changes over time, so the most common value can be out of date and the most recent can be wrong. Neither equal treatment of sources nor unchecked trust in accurate ones works. The strategy classification comes from a tutorial that draws on earlier work not opened. (Dong and Naumann, VLDB tutorial; Dong and Naumann, fusion strategies; Li et al., A survey on truth discovery, arXiv 1505.02463)

Tag basis: input profile inference; selection inference; sequencing inference; robustness inference. All tag values are inference unless a source is named.