Mirror of Oracle documentation
Converted for search and offline reading. Authoritative source: Oracle. Diagrams and some complex tables are simplified — check the PDF when in doubt.
ow deo
alphabetic or numeric sequence of characters in a description string. For example, the string “ABC vanilla yogurt 500gr” is a description that consists of the following five tokens (or unigraphs): “ABC”, “vanilla”, “yogurt”, “500”and “gr”. At the same time, it consists of the following digraphs: “ABC vanilla”, “vanilla yogurt”, “yogurt 500” and “500 gr”.
The Annotation tab provides two screens to annotate unigraphs and digraphs.
Unigraphs
In the Unigraph screen in Annotation tab, you can assign attribute labels to tokens, run the machine learning algorithm to find new attribute labels, and review and approve machinerecommended labels.
Assigning User Labels and Reviewing Machine Labels
On the top left of Unigraph screen, you see the Annotation section that contains a table listing all tokens along with the following fields:
Table 6-1 Annotation Fields
| Field | Description |
|---|---|
| Frequency | The number of times the token appears across all product description strings |
| User Label | The label you have assigned to the token |
| Machine Label | The label recommended by the machine learning algorithm |
| Approved | A check box used to approve and apply the machine label to the token |
| New Discovery | A Yes/No flag that indicates whether or not the machine-recommended label is a new discovery from the most recent run |
In the Annotation section, shown in Figure 6-4, you can perform one of the following tasks for each token:
Table 6-2 Annotation Section Tasks
| Task | Description |
|---|---|
| Change/assign the user label | Select a value from the drop-down menu to change the label or assign a label to a token. (Values in the drop-down menu are the attribute labels that you defined in the Edit Labels tab.) The user label is applied to all instances of the token across all product description strings. |
| Approve a machine label | Check the Approve check box to assign the machine-recommended value to a token. Once you check the box, the token with the approved machine label is moved from the top tables to the bottom table (the Approved Values section). |
a
a
Digraphs
The Digraph is used for labeling two adjacent tokens (as opposed to a single token) with the same attribute. The Digraph screen consists of an Annotation section on the top left for labeling digraphs, an Approved Values section on the bottom left, and a Description String section on the right for labeling individual instances of a digraph.
Assigning User Labels to Digraphs
On the top left of Digraph screen, you see the Annotation section that contains a table listing all combinations of adjacent tokens (labeled as “token1” and “token2”) where exactly one of the two tokens has been labeled and the other token has no label assigned or approved. The token that already has a label is shown using a bold font.
Other fields in this section are as follows:
Table 6-5 Diagraph Fields
| Field | Description |
|---|---|
| Frequency | The number of times that the two tokens are adjacent to one another across all product description strings. |
| Label | The label you have assigned to or approved for one of the tokens. (The token that is shown in bold font already has this label.) |
| Approved | A check box used to approve and apply the label of labeled token to the unlabeled token. |
If you think the two tokens are of the same attribute type indicated by the label, check the Approved check box to apply the label to both tokens. Once you check the box, the approved row is moved from the top table to the bottom table (the Approved Values section).
The Approved Values section on the bottom left displays the adjacent tokens that you have approved so far. You can uncheck the approve check box in this section to remove the approved row and return it to the top table.
The Description String section on the right side of the Digraph screen displays all description strings that contain the adjacent tokens selected on the left, as well as the product hierarchy. All labeled tokens are colored based on the colors you assigned in the Edit Labels tab. The two tokens selected on the left (in the Annotation section) are shown using a bold font.
You can do one of the following in the Description Strings section:
Table 6-6 Description Strings Section Tasks
| Task | Description |
|---|---|
| Remove the label of the originally unlabeled token for one instance | If you believe the combination of two tokens must not be labeled the same in one or few description strings, un-check the Approved check box to remove the label for the token that was not originally labeled. Note that the token that was originally labeled (i.e., the token displayed in bold font in the Annotation section on the left) remains as labeled and is not affected. |
|
|
Table 6-8 Normalized Tokens
| Field | Description |
|---|---|
| Token | The token identified by the algorithm as misspelled. |
| Normalized token | The recommended correct value for the token or the value that you defined for replacement (i.e., the value/token pair). |
| Frequency | The number of times the token appears across all product description strings. |
| Approved | A check box used to approve and apply the recommended correct value. |
You can perform one of the following in Normalization tab:
Table 6-9 Normalization Tab Tasks
| Task | Description |
|---|---|
| Edit the normalized token | If you do not agree with the recommended correction and want to edit the normalized token, you can edit the text in the Normalized Token column before approving it. |
| Approve/reject the normalized token for all instances | Check the Approved check box to replace all instances of the token with a normalized token. The approved rows will be moved from top table to the bottom table (Approved Normalized Tokens). To undo the approval, uncheck the check box in the bottom table. |
| Approve/reject the normalized token for one instance | When you click on a row in the Normalized Token table, all descriptions that contain the selected token will be displayed in the description strings table on the right. You can approve or reject individual instances of the normalization by checking or un-checking the Approved check box in the right table. |
| Clear all recommended corrections | To clear all recommended normalized tokens, click the Reset button on top of the Normalized Tokens table on the top left. |
| Refresh | It is recommended that you click the Refresh button on the top right after you make changes to a LOV or select/deselect a LOV. |
Results Tab
The Results tab is used to view the table of attributes. You can export the results into a spreadsheet. You can click the Complete button on the top right to change the status of the category in the overview tab to a value of “complete”. This indicates that the attribute extraction process for this category is complete. Other users who log in to the application may decide to work on incomplete categories.
In this guide
- Guide: AI Foundation User Guide
- Previous: 5 Customer Segmentation
- Next: 7 Affinity Analysis