First Normal Form Violation in Dimension table

Let’s revisit the author problem in the book dimension. The AUTHOR column contains multiple authors, a first-normal-form violation, which prevents us from querying book or sales by author.

Normalizing and spinning off the authors into a dimension and adding an artificial BOOK AUTHOR fact solves the problem; it is an artificial fact as it does not contain any real business measure.  Note that the Editor which is an author’s role is also “normalized” into its column in the AUTHOR table (It is related to author, but not actually an author’s name).

Note the artificial COUNT measure which facilitates aggregation always has a value of numeric 1.

SELECT name, SUM(COUNT) 
FROM book_author ba, book_dim b, author_dim a
WHERE ba.book_sk = b.book_sk AND ba.author_sk = a.author_sk
GROUP BY name

You might need to query sales by author, which you can do so by combining the queries of each of the two stars (the two facts) on their common dimension (BOOK dimension), producing daily book sales by author.

SELECT dt, title, name, role, sales_amt FROM
 (SELECT book_sk, dt, title, sales_amt FROM sales_fact s, date_dim d, book_dim b
 WHERE s.book_sk = b.book_sk AND s.date_sk = d.date_sk) sales, 
 (SELECT b.book_sk, name, role FROM book_author ba, book_dim b, author_dim a
 WHERE ba.book_sk = b.book_sk AND ba.author_sk = a.author_sk) author
WHERE sales.book_sk = author.book_sk

Single Column with Repeating Value in Dimension table

Columns like the PUBLISHER, though not violating any normal form, is also good to get normalized, which we accomplish by adding an artificial fact, PUBLISHED BOOK fact, and its own dimension, PUBLISHER dimension.

This normalization is not exactly the same as that in normalizing first-normal-form violation; the PUBLISHER dimension can correctly be linked to the SALES fact, the publisher surrogate key must be added though in the SALES fact.

Related Columns with Repeating Value in Dimension table

CATEGORY and SUB-CATEGORY columns are related, they form a hierarchy. Each of them can be normalized into its own dimension, but they need to be all linked into one artificial fact.

Non-Measure Column in Fact table

The ROLE column inside the BOOK AUTHOR fact is not a measure; it violates the dimensional modeling norm; to resolve we just need to spin it off into its own dimension, effectively normalizing the fact table.

Summary

This article shows that both dimensional table and fact table in a dimensional model can be normalized without violating its modeling norm.

If you have read this article, you may be interested to view :

• Solving Many-to-Many Relationship in Dimensional Modeling