Cairn Energy measured the accuracy of the Category Assistant by looking at the documents displayed in the Categories folder and assessing the relevance.
The following section summarizes the deployment environment and the results from the Category Assistant. The project team estimates the Category Assistant to be about 90 percent accurate. The Category Assistant retained accuracy after the initial testing and categorization phase into the second phase. The team automatically categorized only a small proportion of the overall company documents. With the addition of further training documents, it is expected that the Category Assistant will continue with the same level of accuracy.
The following table provides information about the documents included in the index for Cairn Energy.
Index Statistics
Sources | Number of Documents | Size | Health |
---|---|---|---|
Intranet | 2,315 | 500 megabytes (MB) | 99 percent |
Public folder | 5,477 | 500 MB | 99 percent |
Share one | 4,295 | 500 MB | 99 percent |
Share two | 9,986 | 4 GB | 98 percent |
Share three | 60,765 | 16 GB | 97 percent |
Document types for the Cairn Energy deployment break into the following percentages:
All documents are in English.
The project team set the Category Assistant prevision level to High Precision. They also enabled the Category Assistant to categorize all documents, regardless of their storage location.
The following table summarizes the results after training the category assistant. The first update shows the number of training documents and the suggested category results after updating the index the first time. The second update shows the results after updating the index a second time.
Results of the Category Assistant
First Update | Second Update | |||
---|---|---|---|---|
Training Categories | Number of Training documents | Number of Suggested Category | Training Documents | Suggested Category |
Finance | 21 | 303 | 48 | 3,712 |
HSE | 24 | 62 | 24 | 62 |
IM | 20 | 50 | 39 | 140 |
Procurement | 20 | 1,920 | 20 | 1,920 |
Risk | 21 | 128 | 21 | 128 |
Procedures | 0 | 0 | 27 | 176 |
Asset1 | 10 | 25 | 10 | 25 |
Asset2 | 10 | 39 | 15 | 101 |
Asset3 | 0 | 11 | 24 | 11 |
Asset4 | 23 | 105 | 30 | 170 |
Asset5 | 0 | 0 | 30 | 597 |
The Procurement and Finance categories present the most impressive results. The Category Assistant identified documents as purchase orders and categorized them all appropriately. With the Category Assistant set to "High Precision," categorization was highly accurate and resulted in very few ineffective results.