Glossary
Data conflict
A data conflict is a set of incompatible claims about the same subject, property and relevant time.
Truth discovery
A data conflict exists only after the compared statements have been aligned. The subject must be the same entity, the property must have the same meaning, and the values must concern an overlapping time. Different units, reporting dates or definitions can create apparent disagreements without a genuine contradiction. A multivalued property also requires care: two distinct values may both be valid.
A useful conflict record preserves every claim with its wording, source, date and interpretation. Normalisation may reveal that two values are equivalent, while temporal qualification may show that both were correct at different times. If the conflict remains, review can compare source scope, directness and corroboration. Discarding the losing claim erases the reason a decision was needed.
For an organization that acts on a value, unresolved disagreement is itself material information. A defensible decision explains whether a value was selected, deferred or marked unknown, and why. Data conflict is an input to truth discovery; it is not synonymous with an error, and entity resolution must precede it when the subject identity is uncertain.
Example
Two sources assign different current addresses to Halden Therapeutics AG (fictional); first their effective dates and legal-entity identifiers are checked.
Aligning before declaring a conflict
Most apparent conflicts disappear on alignment. The alignment has four steps. First, confirm that both records concern the same subject, which is the job of entity resolution. Second, confirm that the property means the same thing in both sources: a field called status can describe a legal state, an operational state or a workflow state. Third, normalise units, scales, spellings and calendars. Fourth, qualify by time: a value correct on one date and replaced by another is a history, not a disagreement, and the distinction between when something was true and when it was recorded is made precise in the entries on valid time and transaction time.
After alignment, three outcomes remain. The values may be equivalent, in which case the conflict is only one of representation. They may be compatible, as with a multivalued property or two statements at different levels of specificity; the article on truth discovery with hierarchical claims shows how a method that treats a city and a country as rival values will misjudge a source that is merely less specific. Or they may be genuinely incompatible, which is the case that needs a decision.
Recording and resolving a conflict
A conflict record keeps every claim with its wording, source, observation date and the interpretation applied during alignment. Resolution then takes one of a small number of forms: select a value with a stated basis, such as a primary source overriding a secondary one; keep several values with their qualifications; defer pending review; or declare the value unknown. The database-integration literature has long catalogued such strategies under the name of conflict resolution functions, and Knowledge Vault shows the probabilistic variant used in knowledge fusion, in which no single winner is declared and each candidate carries a calibrated probability.
What a defensible process never does is delete the losing claim. If the winner is later shown to be wrong, the loser is the first place to look, and if the basis for the choice was a rule, the rule can only be audited while the alternatives remain. The same principle applies to the output: a field populated for every record, with no visible unknowns, is evidence that conflicts were suppressed somewhere upstream.
A worked example
Two sources give different registered offices for Halden Therapeutics AG (fictional): Basel in a commercial register extract dated January, and Zurich on the company's website retrieved in June. The first check is the entity: both carry the same registry number, so identity is settled. The second is the property: the register entry is the registered office, the website shows the head-office address, which may legitimately differ. The third is time: if the website is more recent, a change of registered office after January is also possible. The record that results states that the two sources describe possibly different properties, retains both values, and sets a review for the next register extract. No value has been overwritten, and no unsupported certainty has been created.
Related terms and boundaries
A data conflict is the input to truth discovery and the output of claim comparison. It depends on entity resolution for identity and on valid time for the period each claim covers. The related practice of survivorship rules addresses the narrower case in which duplicate records of a single entity must be reduced to one surviving value per field. The distinction between a conflict that is resolved and one that is merely hidden is the distinction between a defensible record and a convenient one.
Checks a reviewer can run
Open a conflict and confirm that every claim is still present with its wording and date. Ask what alignment steps were applied and whether any conflict was closed because the values turned out to be equivalent. Count conflicts that were resolved by a rule, by a person, or left open, and ask how long open ones have waited. Choose a displayed value and ask whether the record shows that an alternative existed.
Common misreadings
- Calling values conflicting before aligning entities and dates.
- Overwriting a losing claim.
- Assuming every differing value is erroneous.
Related terms
Appears in
- Majority voting vs truth discovery
A count of agreeing sources is a useful baseline; truth discovery also considers how those sources behave.
- Reporting source conflicts
Preserve competing values and their grounds until a scoped decision can be defended.
- Scoping a first record
Start with one contested entity and a bounded decision, then test whether its evidence can be defended.
References
- Li et al. (2016), A Survey on Truth Discovery, ACM SIGKDD Explorations 17(2).
- Dong et al. (2014), Knowledge Vault: A Web-Scale Approach to Probabilistic Knowledge Fusion, KDD.
- Hogan et al. (2021), Knowledge Graphs, ACM Computing Surveys 54(4).
- Wikidata, Help:Ranking (preferred, normal and deprecated statements).
Found an error on this page? See corrections for what counts as one, how to report it and the dated log.
Enquiry
Where does "Data conflict" bear on a decision of yours?
This entry sits under truth discovery. If a decision in your organization turns on it, send the entity concerned, the sources you fuse to describe it and who may read which of them. We reply with what a first record would cover.
Or write to [email protected] in your own words.