Improving SQL Data Analysis through Database Diagrams

Data modeling and database diagrams for SQL are essential tools for enhancing SQL data analysis. This comprehensive guide explores the importance of these techniques in structuring and visualizing data for better analysis. Learn how to implement data models that reflect real-world scenarios and create diagrams that simplify complex data relationships.

In this guide, we will explore the fundamentals of data modeling and database diagrams, demonstrating their importance in SQL analytics. We will provide practical examples and best practices to help you implement these techniques effectively, ensuring your data analysis processes are both efficient and insightful.

1. Introduction to Stack Overflow Datasets

Stack Overflow is a vibrant community for discussing programming questions, solving technical problems, and sharing knowledge. It offers various datasets that are perfect for practicing data-related tasks. For our analysis, we'll focus on the posts_questions, posts_answers, and comments tables to uncover trends, identify popular topics, and gain insights into frequently discussed questions. Link to the dataset on BigQuery.

2. Basic Exploration

To start, we need to understand the structure of the data. We'll perform exploratory data analysis (EDA) by examining the key tables.

Posts_questions Table

SELECT
  q.tags AS question_tag,
  q.id AS question_id,
  q.title AS question_title,
  q.body AS question_body
FROM
  `bigquery-public-data.stackoverflow.posts_questions` AS q
LIMIT 1000

This posts_questions table contains all the questions posted on Stack Overflow. Each row represents a unique question and its various attributes, such as ID, title, and tags.

Posts_answers Table

SELECT
  a.id AS answer_id,
  a.parent_id,
  ARRAY_AGG(STRUCT(id, body)) AS answers,
  a.body AS answer_body
FROM
  `bigquery-public-data.stackoverflow.posts_answers` AS a
LIMIT 1000

The posts_answers table contains all the answers related to the question table. We use answer_id to specify the unique answer and parent_id which will be used as the key constraint with question_id from the question table.

Comments Table

SELECT
  *
FROM
  `bigquery-public-data.stackoverflow.comments`
LIMIT 1000

The comments table contains all the comments posted on both questions and answers specified by post_id. Also, it contains the creation date, user ID and score of a comment. 

3. Joining Tables

Joining posts_questions with posts_answers

WITH agg_answer AS (
  SELECT
    parent_id,
    ARRAY_AGG(STRUCT(id, body)) AS answers
  FROM
    `bigquery-public-data.stackoverflow.posts_answers`
  GROUP BY parent_id
)
SELECT
  q.tags AS question_tag,
  q.id AS question_id,
  q.title AS question_title,
  q.body AS question_body,
  a.*
FROM
  `bigquery-public-data.stackoverflow.posts_questions` AS q
LEFT JOIN
  agg_answer AS a
ON
  q.id = a.parent_id
WHERE
  REGEXP_CONTAINS(q.tags, 'tensorflow')

From this joins, we know that parent_id from the answer table can be used as the key for joining with a question table. Also, the result will filter only ‘tensorflow’ related question since the tags column is filtered with reg_exp_contains.

Here's example of how to use Datascale to improve SQL understanding: Link to this query

• Where AI query guide helps explain the logic inside. This also works for non-technical users to read and understand your query.

  • "Q" means question that this query answered.

  • "TL;DR" is a quick summary.

  • "Summary" explains logical flow

  • "Key topics" helps categorize analyses when you have many of them!

• ER diagram is where we model your joint relationship into ER digram.

  • Now you can easily see which keys are used.

Joining posts_questions with comments

WITH agg_comments AS (
  SELECT
    post_id,
    ARRAY_AGG(STRUCT(id, text)) AS comments
  FROM
    `bigquery-public-data.stackoverflow.comments`
  GROUP BY post_id
)
SELECT
  q.tags AS question_tag,
  q.id AS question_id,
  q.title AS question_title,
  q.body AS question_body,
  c.*
FROM
  `bigquery-public-data.stackoverflow.posts_questions` AS q
LEFT JOIN
  agg_comments AS c
ON
  q.id = c.post_id
WHERE
  REGEXP_CONTAINS(q.tags, 'tensorflow')

Form the above SQL, we know all the comments contained in the question table. We use the post_id from the comment table joined with the unique ID from the question table.

These two basic analyses will be used as a guide for the advanced analysis which is the data that answer the business question.

4. Advanced Analysis

With these joins, we can perform advanced analyses such as tracking trends over time, identifying the most discussed topics, and measuring user engagement.

Trend Analysis

SELECT
  EXTRACT(YEAR FROM q.creation_date) AS year,
  EXTRACT(MONTH FROM q.creation_date) AS month,
  COUNT(q.id) AS num_questions
FROM
  `bigquery-public-data.stackoverflow.posts_questions` AS q
WHERE
  REGEXP_CONTAINS(q.tags, 'tensorflow')
GROUP BY
  year, month
ORDER BY
  year, month

The trend analysis can be used to predict customer interest in various periods. The value is extracted as year, month, and count of question ID.

Top 10 Most Discussed Questions

WITH question_answers AS (
  SELECT
    q.id AS question_id,
    q.title AS question_title,
    COUNT(a.id) AS num_answers
  FROM
    `bigquery-public-data.stackoverflow.posts_questions` AS q
  LEFT JOIN
    `bigquery-public-data.stackoverflow.posts_answers` AS a
  ON
    q.id = a.parent_id
  WHERE
    REGEXP_CONTAINS(q.tags, 'tensorflow')
  GROUP BY
    question_id, question_title
),
question_comments AS (
  SELECT
    q.id AS question_id,
    COUNT(c.id) AS num_comments
  FROM
    `bigquery-public-data.stackoverflow.posts_questions` AS q
  LEFT JOIN
    `bigquery-public-data.stackoverflow.comments` AS c
  ON
    q.id = c.post_id
  WHERE
    REGEXP_CONTAINS(q.tags, 'tensorflow')
  GROUP BY
    question_id
),
final AS (
  SELECT
    qa.question_id,
    qa.question_title,
    qa.num_answers,
    qc.num_comments,
    (qa.num_answers + qc.num_comments) AS total_interactions
  FROM
    question_answers AS qa
  JOIN
    question_comments AS qc
  ON
    qa.question_id = qc.question_id
  ORDER BY
    total_interactions DESC
  LIMIT 10
)
SELECT * FROM

The query find the ranking of the 10 most user-interested questions for use in business analytics. Also, we use CTEs to simplify the query. Link to the query.

Here's the visualization in ER Digram 🎉

How would the AI summarize this query? ✨

5. Enhancing Analytics with Datascale's SQL Diagram

While BigQuery provides a powerful platform for data analysis, integrating this process with Datascale can significantly enhance productivity as a Second-brain for your SQL.

For the ease of storing SQL code after analysis, we can store our SQL code to be our backup in a shared workspace. Moreover, it is easier to manage and govern the SQL code by storing it in a single source of truth. 🎉

Example of SQL Notes workspace

There's few more SQL utils feature that might be related to your use cases too, e.g.,

• Version history for your query

• Validating anti-patterns query using SQLFluff

Data Modeling & Lineage Digram

After storing the SQL, the platform helps generate the lineage and ER diagram which will display the relationship between tables.

AI Data Dictionary & Metadata

Since the modeling only shows the relationship at the table level, we also provide AI-generated data dictionary to understand context and description for each column.