Creating a Model that Includes Text Mining

Oracle Data Mining supports unstructured text within columns of VARCHAR2, CHAR, CLOB, BLOB, and BFILE, as described in Table 7-1.

Table 7-1 Column Data Types That May Contain Unstructured Text

Data Type Description

BFILE and BLOB

Oracle Data Mining interprets BLOB and BFILE as text only if you identify the columns as text when you create the model. If you do not identify the columns as text, then CREATE_MODEL returns an error.

CLOB

Oracle Data Mining interprets CLOB as text.

CHAR

Oracle Data Mining interprets CHAR as categorical by default. You can identify columns of CHAR as text when you create the model.

VARCHAR2

Oracle Data Mining interprets VARCHAR2 with data length > 4000 as text.

Oracle Data Mining interprets VARCHAR2 with data length <= 4000 as categorical by default. You can identify these columns as text when you create the model.

Note:

Text is not supported in nested columns or as a target in supervised data mining.

The settings described in Table 7-2 control the term extraction process for text attributes in a model. Instructions for specifying model settings are in "Specifying Model Settings".

Table 7-2 Model Settings for Text

Setting Name Data Type Setting Value Description

ODMS_TEXT_POLICY_NAME

VARCHAR2(4000)

Name of an Oracle Text policy object created with CTX_DDL.CREATE_POLICY

Affects how individual tokens are extracted from unstructured text. See "Creating a Text Policy".

ODMS_TEXT_MAX_FEATURES

INTEGER

1 <= value <= 100000

Maximum number of features to use from the document set (across all documents of each text column) passed to CREATE_MODEL.

Default is 3000.

A model can include one or more text attributes. A model with text attributes can also include categorical and numerical attributes.

To create a model that includes text attributes:

  1. Create an Oracle Text policy object, as described in "Creating a Text Policy".

  2. Specify the model configuration settings that are described in Table 7-2.

  3. Specify which columns should be treated as text and, optionally, provide text transformation instructions for individual attributes. See "Configuring a Text Attribute".

  4. Pass the model settings and text transformation instructions to DBMS_DATA_MINING.CREATE_MODEL. See "Embedding Transformations in a Model".

    Note:

    All algorithms except O-Cluster can support columns of unstructured text.

    The use of unstructured text is not recommended for association rules (Apriori).