Our G2 QuickView prototype is intended to serve a few purposes. First and foremost, G2 QuickView allows you to easily glance over the kinds of decisions G2 makes. These results are grouped by: Matches, Possible Matches, and Relationships.
Next, and equally important, G2 QuickView allows you to provide accuracy feedback - whether or not you think G2 is right or wrong. The viewer accomplishes this by presenting one "audit" item at a time (an Entity Resolution decision). As items are reviewed, you will mark items "Audited" of note an "Exception", if any.
During this process, various statistics are computed including the percentage of False Positives & False Negatives that G2 is producing on your data. Audit feedback is written to an audit file that contains summary-level information about audit items (no names, addresses, identifiers, phone numbers, or other PII.). Should you decide to share G2 audit files with Senzing, we will use this information to improve G2 accuracy.
IMPORTANT: If you are going to use G2's results to make business decisions, be sure to review a sufficient number of audit pairs. It is essential that you understand the false positive and false negative rates before you consider G2 suitable for your business objectives.
The G2 Quickview tool allows the user to explore and audit how a data source relates to itself, specifically: 1) Duplicates; 2) Possible Duplicates; and 3) Relationships. Additionally, the user can explore and audit how data sources relate to each other, specifically: 1) Matches; 2) Possible Matches; and 3) Relationships.
Duplicates/Matches: Matches are Entity Resolutions that G2 identifies with high confidence. You will want to review these carefully to determine that G2 has identified these correctly. If you happen to find a False Positive (a match with which you disagree), you will want to click the "Disagree" button.
Possible Duplicates/Matches: Possible Matches are G2’s tentatively-identified Matches for which sufficient data is lacking to make a high confidence claim. Upon inspection, you will likely find some true matches for which you will use the "Resolve" button. In other cases, you are likely to identify Entity Resolved items that are unrelated. For those items, you will want to use the "Disagree" button.
There will also likely be instances when you are unsure. In these cases, your choices are: 1) run off and research the items to determine ground truth; or 2) mark these items "audited as true possible matches" and wait for G2 to self-correct as new data becomes available.
Relationships: Relationships are connections that G2 has made between entities. In other words, this entity contains information that relates to another entity, e.g., entities sharing an address or phone number with different names. In your review, this group is very important for identifying those insidious "False Negatives" – true matches that G2 completely misses. Here you will also be able to mark items as "Meaningless Relationships"; feedback that will help G2 make better decisions in the future (as well as change its mind about the past).
Because G2 is a real-time, transactional, ER engine, we will be able to deliver some very exciting and unique features.
For example, we can envision a future version of this viewer where as soon as you provide audit feedback, these "hints" are submitted to your G2 system improving the accuracy of your G2 database in real-time! We can also imagine general discoveries that may become apparent during an audit, thus making it possible to locate and correct a cluster of similarly-affected entities in real-time. This will be part of our "Active Maintenance" capability that will virtually eliminate the need to fully re-load G2.