How the OAR improves data quality
The OAR is powered by a deduplication algorithm, which processes each line of contributed data to detect whether or not facilities already exist in the database.
The OAR’s technical team runs regular training exercises to develop and refine the algorithm, which is based on a statistical model. In addition to these training exercises, the OAR team continually moderates data in the tool to ensure a high quality of data and maintain the trust of our valued users. We also reactively moderate data, when discrepancies are reported to us by users.
How name and address data is processed
Every line of data contributed to the OAR is processed by a deduplication algorithm.
Using the Dedupe Python library, the workflow begins by comparing a new set of records submitted by a contributor to the existing set of mapped facilities. Dedupe can match large lists accurately because it uses blocking and active learning to intelligently reduce the amount of work required.
The OAR uses simple string comparisons for its facility data. Pre-processing strings before comparing them maximizes the quality of match results.
Each entry to the database then falls into one of three categories:
The OAR auto-accepts matches with 80% confidence or greater
It presents matches between 50% and 80% confidence to the contributor as a "potential match", asking for human intervention to “confirm” or “reject” the potential match
Anything below the 50% threshold is automatically created as a new entity in the tool and allocated its own unique OAR ID
See this technical blog for more detail on how the OAR processes data.
How additional data points are processed
You can view a full table outlining how each data point is contributed, processed, searched and displayed in the OAR.
Number of Workers, Type of Processing and Type of Facility have additional processing steps to allow for more powerful and standardized searching.
Number of Workers: Users can contribute raw values or ranges for this field. All contributions are then processed into searchable categories (selected to align with the Transparency Pledge):
- Less than 1000
- 1001 - 5000
- 5001 - 10,000
- More than 10,000
Type of Processing / Facility: Contributors upload all processing and/or facility types to the OAR in a single field. The OAR then processes those values into a set of searchable facility types based on this Facility Type Taxonomy. The searchable facility type categories are:
Raw Material Processing or Production
Textile or Material Production
Printing, Product Dyeing and Laundering
Final Product Assembly
Warehousing / Distribution
Office / HQ
How data is moderated
Alongside the automated work of the OAR’s algorithm, the OAR team continually moderates data in the tool to eliminate any duplicates that may have crept into the database, as well as to promote higher quality data or update GPS coordinates. All data moderation is logged in our publicly available Moderation Log.
How you can help
Spotted a duplicate in our data? Have access to more accurate GPS coordinates for a facility? Contact the team to report improvements or suggested changes to OAR data.
Merge facilities: share OAR IDs of the facilities that you think should be merged, i.e. where you have identified two or more entries in the database which you believe are the same facility.
Split facilities: share OAR IDs of the facilities that you believe should be unmerged/split, i.e. where two or more names and addresses have been identified as a match, but you believe should be listed as separate entities.
Promote alternative name and address: in most instances, the master entry for a facility in the OAR is the original entry for that facility in the database. Share the OAR ID and the alternative name or address that you believe should be promoted to the primary position, i.e. where more complete or accurate address details are available.
Updating GPS coordinates: members of the OAR community with local knowledge of facility locations may have access to more accurate GPS coordinates than those assigned to a facility profile by Google’s geo-coder for the OAR. While considered best-in-class, Google’s geo-coding is not perfect, and so the OAR team welcomes the submission of alternative GPS coordinates. While neither name and address details nor GPS coordinates are ever over-written by the OAR, additional GPS coordinates may be added to a facility’s profile, with the caveat that one facility clearly cannot exist in two physical locations.
Report a facility as closed: you can report a facility as closed on the OAR by selecting "Report as closed" on a facility profile. Read our Facility Closure policy here.