7
SELECT p.title, p.year, pa.author_order, a.name
FROM Papers p, Paper_Authors pa, Authors a, Conferences c
WHERE p.id = pa.paper_id AND pa.author_id = a.id
AND p.conference_id = c.id
AND c.acronym = ‘SIGMOD’
ORDER BY p.year DESC
Find all SIGMOD papers and their authors:
10
Join results are hard to interpret
Due to duplications produced from joining
tables in many-to-many relationships
Challenge 1
More natural representations exist
e.g., Nested tables, Comma-separated lists
But not directly supported by relational DB,
due to 1NF
11
Challenge 1
Nested tables
More natural representations exist
e.g., Nested tables, Comma-separated lists
But not directly supported by relational DB,
due to 1NF
12
Challenge 1
Comma-separated lists
Our Solution: Presentation View Layer
Inspired by the presentation data model
concept, defined as a full-fledged layer on top
of the logical models (Jagadish et al., 2007).
13
Challenge 1
Presentation view layer
Logical models
Queries and results are separated
Introducing a usability breakdown between
users’ actions and their results.
14
Challenge 2
Query
Results
Our Solution: Direct Manipulation
Users directly interact with query results
to refine them (e.g., click)
15
Challenge 2
Figure from https://www.microsoft.com/en-us/research/video/holograph-3-d-spatiotemporal-interactive-data-visualization/
Research Goals
16
Duplications in
join query results
A presentation-level view
with easy-to-interpret format
Easy-to-use interactions
based on direct manipulation
Separation of
queries and results
1.
2.
Develop usable database interfaces
1.
2.
Problems: Our Solutions:
We propose ETable
For interactively browsing and navigating
databases at the entity-relationship level
ETable tightly integrates
• A novel hybrid representation
(between relational and nested-relational)
• A novel set of interactions that closely work
with the representation
17
1NF (duplications) may need large screen space
Design of ETable
Each column represents either a base attribute
or a set of entity references (Non-1NF)
Count
Other tables
Entity-reference columnsBase attributes
A set of entities
How to determine additional columns?
Direct neighbors in entity-relationship graphs
created during preprocessing
19
E-R Schema GraphRelational Schema
Users explore data at conceptual level
Once translated, ETable works with E-R graph,
which frees users from logical schema
20
Presentation
Conceptual
Logical
How do users specify queries?
Users directly interact with the table, then those
actions are translated into internal operators.
21
Pivot = Join + Aggregation
SELECT Papers.*, ..
FROM Conferences c, Papers p
WHERE c.id = p.conference_id
AND c.acronym = ‘SIGMOD’ ..
GROUP BY Papers.id
SELECT Authors.*, ..
FROM Conferences c, Papers p,
Paper_Authors pa, Authors a
WHERE c.id = p.conference_id
AND .. AND pa.author_id = a.id
AND c.acronym = ‘SIGMOD’ ..
GROUP BY Authors.id
ORDER BY COUNT(*) DESC
Pivot on authors => shift focus to authors
PapersConferences
acr.. = ‘SIGMOD’ keyword like ‘user’
Equivalent SQL
Internal Query Representation
Add(‘Authors’)
Shift(‘Authors’)
AuthorsPapersConferences
acr.. = ‘SIGMOD’ keyword like ‘user’
Recap
Open Conferences table1
Click paper count2
Apply filters3
Click pivot button4
Find researchers who have published the most papers in SIGMOD after 2005
User Study Design
12 participants performed 6 querying tasks
• Measured task completion times
• ETable vs Navicat Graphical Query Builder
(Within-subject design)
30
Task Category #relations
2. Find all the keywords of the paper titled
`Collaborative filtering with temporal dynamics'.
Attribute 2
4. Find all the papers written by researchers at
`Carnegie Mellon' and published at KDD.
Filter 5
6. Find the top 3 researchers who have published
the most papers in the SIGMOD conference.
Aggregate 4
User Feedback
Pivot made it easy to specify complex queries:
“I also loved... having multiple pivots throughout the
course of forming a query. I messed up a query, but
could still find the right answer by doing an
appropriate pivot.”
Interface is intuitive:
“Visually, I was able to see... the effects of the SQL
operations, which made it easier to use and verify
intermediate results.”
32
33
1. Novel representation for conceptual level exploration
2. Novel operations for interactive query construction
ETable:
Interactive Browsing & Navigation
in Relational Databases
Minsuk (Brian) Kahng
Georgia Tech CS PhD student
http://minsuk.com