Adding facets

Creating a facet from existing fields in a collection

As you apply enrichments to a collection, new fields are added to the index. Information that is recognized by the enrichment is stored in these new fields. You can use the enrichment fields as the source for facets.

For example, if you apply the Keywords prebuilt enrichment, you can create a facet based on the keywords that are found in your collection. As it processes your document, the Keywords enrichment recognizes any keyword mentions that occur and stores information about them in a set of fields that begin with the prefix enriched_{field_name}.keywords. To create a keyword facet, add a facet that takes its categories from the field where keyword mentions are stored, the enriched_{field_name}.keywords.mentions.text field. For more information about the Keywords enrichment, see Applying prebuilt enrichments.

Similarly, you can use the fields that are generated by domain-specific enrichments as the source of the facet. In fact, as you create a Regular expression enrichment, you can define a facet by which to categorize any recognized expression mentions. For more information, see Adding domain-specific resources.

To add a facet from existing fields, complete the following steps:

1. On the Improve and customize page, click Customize display, and then click Facets.

2. Click New facet > From existing fields in a collection.

3. Choose the field you'd like your facet to use, such as enriched_text.entities.type. Add the facet label and choose a filter control option.

4. Optional: You can adjust the maximum number of facet values that are displayed.

   The maximum number is the sum of the number of facet values that are displayed for each facet type. By default, a total of 10 facet values are displayed.

5. As you test the facet, you can adjust the options.

Creating a facet by creating a dictionary

Add a facet to group a set of terms that have special meaning to your use case by creating a dictionary.

For example, an owner of a retail clothing store collects customer reviews and wants to be able to find any recurring complaints to identify retail items to discontinue. The owner can create a dictionary to help recognize and tag mentions of specific articles of clothing in the review text. To support the goal of filtering customer feedback by clothing article type, the owner might add entries similar to the entries listed in the following table.

A review that says, This long-sleeve is so badly proportioned. Who has arms that long! is returned when you filter the documents by the clothing facet.

To create a facet by creating a dictionary, complete the following steps:

1. On the Improve and customize page, click Customize display, and then click Facets.

2. Click New facet > By creating a dictionary.

3. Enter a name for the facet, and then create a dictionary of the terms that you want to categorize.

   After you save the dictionary, the name that you used for the facet label is shown in the list of facets.

4. As you test the facet, you can add more terms to the dictionary you created by selecting Teach domain concepts > Dictionaries.

   The dictionary that you created is shown in the list on the Dictionaries page. For more information, see Dictionaries.

Creating a facet by identifying a pattern

Patterns is a beta feature and supports English language documents only.

IBM Cloud IBM Cloud only

This facet is available only to managed deployments and in Document Retrieval and Conversational Search projects.

The Patterns feature uses pattern induction to help you teach Discovery to recognize patterns in your data. Pattern induction generates extraction patterns from examples that you provide. After you specify a few examples, Discovery suggests more rules that you must verify to complete the pattern.

Pattern recognition works best on text with consistent structure in casing, length, text, or numeric values. Examples of patterns you can teach Discovery to identify in your documents:

• All ISO standard numbers, for example ISO 45001, ISO 22000
• All currency amounts, for example $50.5 million, 29 dollars, $29.00
• All dates, for example 8 September 2019, June 12, 2020

If you need to identify specific terms or text, create a facet as part of a dictionary entry instead. For example, use a dictionary facet for grouping the following types of information:

• All products in the same family, Cloud Pak for Data, Cloud Pak for Automation, Cloud Pak for Security
• All terms in the same category, carburetor, piston, valves

For more information, see Creating a facet by creating a dictionary.

To add a facet by identifying a pattern, complete the following steps:

1. On the Improve and customize page, click Customize display, and then click Facets.

2. Click New facet > By identifying a pattern.

3. On the Create facet from a new pattern page, choose how you want to select documents. You can allow Discovery to select 10 random documents for you or you can select up to 20 documents yourself.

   If you select the documents yourself, follow these guidelines:

   • Choose documents that contain mentions of the pattern in every different format that you want the model to be able to recognize later.

   • If you want the model to understand ways that a term is used that don't fit the pattern, include a document that uses the term in the wrong way so you can intentionally omit its selection.

     For example, the pattern you want to add might capture mentions of clothing. In documents that mention a top to refer to a shirt, you would select top as a clothing mention. But, you might also want to include a document that mentions the top 3 fashion trends and purposely not select top when the term is used in this context. The omission teaches the model that top doesn't fit the pattern when the term is used to mean the best.

   • Include documents that have a maximum of 5,000 characters. Any document that exceeds the limit is truncated to 5,000 characters.

4. Click Next.

5. Select example words or phrases that fit the pattern that you want to define.

   For example, if you want to define a date pattern, start highlighting date mentions in each document. If you make a mistake, hover over the selection and click the x to delete it.

   Be sure to select every mention of the pattern. The model learns from what you omit as much as what you select.

6. Continue selecting examples. After you identify enough examples, Discovery shows a list of suggested examples for you to validate. Choose Yes or No for each one.

   Click the Preview document icon if you want to confirm the example in context.

7. Continue highlighting examples and validating suggestions until a message is displayed to indicate that you provided enough examples.

8. Click the Review examples tab to review the lists of examples.

9. If the examples are correct, click Save pattern.

If the system cannot determine a valid pattern, the Save pattern button is never enabled. The system cannot determine a valid pattern if you choose contradictory examples to illustrate the pattern. If you cannot save your work, click the Reset button to start over. The documents are returned to their original state and any examples that were identified previously are no longer selected.

After you save the pattern, the name that you specified for the facet label is shown in the list of facets.