The same company, product or place is rarely written the same way twice. Entity matching links records that describe the same real-world thing across websites and systems, removes duplicates and, where needed, builds a single master record. Infovium combines rules, similarity scoring and AI-based methods, with human review where appropriate.
Where entity matching is used
- Product matching across websites: the same item listed by several sellers or marketplaces, for price comparison and assortment analysis, see ecommerce data scraping.
- Company and entity deduplication: directories and lead lists that overlap, see business directory data.
- Joining your records to external data: linking your customer or supplier list to public sources, see data enrichment.
- Master data: one consolidated record per entity.
Matching methods we use
| Method | What it compares | When it fits |
|---|---|---|
| Exact and rule-based matching | Identifiers, normalized names, domains or codes | Records that share a reliable key |
| Fuzzy matching and similarity scoring | Names and addresses that differ in spelling, abbreviation or word order | Records without a shared key |
| AI or embedding-based semantic matching | Descriptions that use different words for the same thing | Product titles and descriptions that vary between sources |
| Human review | Pairs the methods above cannot settle with confidence | Uncertain matches, where review is appropriate for the project |
The method is chosen per dataset and usually combines several of these. We do not publish accuracy percentages because results depend on the data. Quality checks are part of the project scope.
Golden records
When several records describe one entity, a golden record (master record) is built by choosing or combining the best value for each field. How conflicting values are resolved is agreed with you in the project scope.
Illustrative example
| Source A | Source B | Outcome |
|---|---|---|
| Acme Pvt. Ltd., Ahmedabad | ACME Private Limited | Matched, one master record |
| Phone X 128GB Black | Phone X (Black, 128 GB) | Matched as the same product |
| Delta Foods | Delta Food Services | Uncertain, sent to human review |
Examples are illustrative, not client data.
What the output can look like
The final output schema, matching details and review requirements depend on the project and are agreed in the project scope. Options that can be considered include:
- Your records with a match group ID, so records that belong together share a value.
- A deduplicated dataset with one record per entity.
- Golden (master) records built from the matched records.
- A list of uncertain pairs for review.
These are not all provided by default. Match scores or source references are included only where agreed for the project.
How it works
- We review sample records and agree what counts as the same entity.
- Data is cleaned first, see data cleaning and normalization.
- Matching rules and models are applied and tuned on a sample.
- Uncertain matches can go to human review where appropriate.
- Matched or deduplicated data, or golden records, are delivered in the format agreed for the project.
What we need from you
- The records to match, or the sources they will be collected from.
- Which fields identify an entity for you, such as name, address, domain or product code.
- A few examples of records you consider the same, and a few you consider different.
- Whether you want deduplicated records, match groups or golden records back.
Delivery
Output can be delivered as files, through a REST API, over SFTP or FTP, to cloud storage or directly into a database, depending on project requirements, see data delivery. The fields included, such as match groups or scores, are agreed in the project scope.
Related services
- Ecommerce data scraping and product matching
- Business directory data
- AI-ready datasets
Entity matching FAQ
What is entity resolution?
It is the process of deciding which records refer to the same real-world entity, then merging or linking them.
Can you match products across different websites?
Yes. Product matching across websites is a core use case, using identifiers where present and fuzzy or semantic matching where they are not.
Do humans check the results?
Uncertain matches can be routed to human review where appropriate for the project.
Do you promise a match accuracy?
No. Accuracy depends on the data, so we do not quote percentages.
Can I get a sample?
Free samples may be available for suitable projects, and paid pilots can be arranged.
Discuss your matching project
Describe the records you want matched or deduplicated and where they come from.