Reading and highlighting texts take a long time. Using Artificial Intelligence and natural language processing have created Highlighter Suggest, a module that reads and understands the text and returns these suggestions to the user. The user can then accept or reject these suggestions.
We monitor user’s reaction to each suggestion and use this knowledge to train the computer to make better suggestions in the future.
A new button is added to the Highlighter module which invokes a natural language processing module that analyzes the text and returns sets of suggestions for roles, activities and rules. Below we have clicked the suggest button and the result is shown as dotted lines around words. Placing the mouse over a word will show the type etc as the role of a “Person” below:
Clicking Add Item will mark the suggested text.
Benchmarks
The following table compares the performance of the Highlighter Suggest module against highlights that have been manually performed by expert users.
| Case Id | Title | Description Source | # Activities (manual highlights) | #Activities Suggested | False Positives | False Negatives |
|---|---|---|---|---|---|---|
| 1 | Hospital Workflow | Process Description DCR Process Model | 11 | 12 | 1 | 1 |
| 2 | Insurance Workflow | Process Description DCR Process Model | 11 | 11 | 0 | 0 |
| 3 | Computer Repair Service | Process Description DCR Process Model | 14 | 12 | 0 | 2 |
| 4 | Expense Handling Process | DCR Process Model | 6 | 12 | 0 | 0 |
