Modelling Data on the Web

Tim Morris & Uli Sattler

Week 1 Introduction, Data Models, Tables, and SQL

Topic Overview

  • What is a (core) data model? E.g.,
    • Flat: flat files
    • Table based: relational
    • Tree based: XML and a bit of JSON
    • Graph based: RDF
  • Trade offs (esp. representational) between them
    • Discussing pain points & sweet spots, distinguishing
      • principled ones from
      • DM-based ones from
      • those caused by your usage of DM

Course Goals:

Knowledge & Understanding

  • This course unit aims to give you a
    • good understanding of core concepts of data modelling
    • some familiarity with formalisms, APIs, and languages
      • for modelling data on the web
      • design/representation issues that arise

Course Goals:


  • This course unit aims to give you the ability/skill to
    • compare different data modelling formalisms,
    • design or analyse a data management system,
      • does it make good use of the formalism's features?
      • does it fit its purpose?

Course Structure

Our Expectations

  • Lectures:
    • active listening & participation
  • Lab Mondays afternoon:
    • make sure you understand the coursework!
  • Lab during week:
    • work on your coursework
    • make use of TAs: 14:00-15:00
  • Coursework:
    • submit on time
  • Read!


  • Coursework (50%, ≈200 marks)
    • Each week, a mixture
      1. MCQ quizzes (≈10 marks)
      2. Short essays (≈5 marks)
      3. A modelling assignment (≈10 marks)
      4. A programming assignment (≈15 marks)
    • Precise mark breakdown varies
  • Exam (50%)
    • Taken online
    • Very like 1 & 2

Materials & Blackboard

  • All course materials are available online on the materials page
  • We use Blackboard for
    • Coursework
    • Online forums
      • Subscribe to each forum
      • Ask questions there
      • Answer questions there
      • Share examples, test cases there
    • Exam

Variant Circumstances

...feel free to ask us: we're happy to advise!

Assistance & Help

  • Early intervention is more effective
    • If you are having challenges of any sort
      • the sooner they are identified and
      • communicated to us
      • the more likely we can find a good resolution
  • This is very true for mitigating circumstances
    • If something is interfering, document it!
    • Fill out the form when things are happening
    • There is a "too late" here!

...when in doubt, ask us and SSO for MitCircs

Expected conduct

  • We expect of you (and ourselves) to
    • be fair minded
    • treat each other well & with respect
    • avoid academic malpractice
    • take responsibility for course duties
    • be engaged, curious, and active
  • If you have a problem or issue
    • please raise it with us
    • if that doesn't help, contact your programme director


Baby penguins keep warm in their parents' feet

We all have to start somewhere

Data Management (1)

  • Almost every program must do some data management
    • If only config files!
    • Many are information heavy
      • and must deal with that information over time
  • Database Management Systems (DBMSs)
    • Separate (or separable) component
    • Specialised for variables purposed
      • secondary storage, scaling, complexity, etc.

Data Management:


  • Some data is (typically) transient or ephemeral
    • Position of the cursor on the screen
  • Some data is (typically) persistent
    • Bank records, addresses, health data, library entries
    • Cursor position can be!
      • (If you are recording the screen...)

We're focused on data that leans toward persistent

Data Management:


  • Some data is (more or less) informationally opaque
    • e.g., images, video, text, audio
    • its information/content isn't (easily) available
      • You typically must do some extraction
    • this is called unstructured data
  • Some data is informationally transparent
    • its information/content is programmatically explicit
    • this is called (semi-)structured data

Out of Scope

  • There is lots of DM that's outside our scope
    1. Performance & Scaling: see COMP62421
    2. Concurrency
      • Thus transactions
        • (You should read up on ACIDity)
    3. Tuning, indeed most physical level stuff
    4. Cleansing
    5. Integration
      • Except for a tiny bit, around merging

These considerations do affect modelling!

Data and the Web

  • The Web is a collaborative information structure
    • Largely decentralised
    • Immense
    • Growing rapidly
    • Changing rapidly
  • The Web produces new data challenges
    • Scale of data
    • Kind of data
    • Shape of data
    • Use of data

Data on, from, behind the Web

  • On the Web
    •,, ...
  • From the Web
    • Log files
  • Behind the Web
    • Data(base) backed Websites
      • The filesystem is a kind of database
    • Content Management Systems
      • Wordpress
    • Sites as Database Front Ends
      • See Amazon

What is a Data Model?

  • Three Key Aspects
    1. Underlying Data Structure, "Core Data Model"
    2. Data Integrity
    3. Data Manipulation
    4. (Plus a fourth!) Data Sharing
      • More important on the Web *

"Data Model" is Ambiguous:

  1. a complete data representation and manipulation approach (we do this!)
  2. just the core data model
  3. a particular data representation for a domain or application, also called the domain model
    • "Does your calendar data model include leap years?"

Generally, you can tell from context, (2) is rare.

Kinds of Data

  • Data can lend itself to different shapes
    • Array-like
    • Tree-like
    • Graph-like
    • Document-like
  • Data can have different volumes
    • Small to "big" data
  • Data can have different velocities
    • Static/offline to streaming
  • Data can have different use patterns
    • Many readers/few writers or the reverse or other!

Data Does Not Grow on Trees

  • Data may lend itself to one shape
    • e.g., tree-shape or graph-shape
  • but this does not mean that
    • we have to persist it in this form
    • we know exactly how to cast it in this form
    • ...consider pain-points and sweet spots
    • others share it in this form

Polyglot Persistence

...we are gearing up for a shift to polyglot persistence — where any decent sized enterprise will have a variety of different data storage technologies for different kinds of data. There will still be large amounts of it managed in relational stores, but increasingly we'll be first asking how we want to manipulate the data and only then figuring out what technology is the best bet for it.

Martin Fowler

Polyglot Persistence (2)

This polyglot [e]ffect will be apparent even within a single application. A complex enterprise application uses different kinds of data, and already usually integrates information from different sources. Increasingly we'll see such applications manage their own data using different technologies depending on how the data is used.
Martin Fowler

Poly -glot/-system Persistence

  • Even a single core data model can result in
    • multiple systems with different characteristics
    • multiple, overlapping, domain models
    • multiple, overlapping owners, versions, variants

This is particularly true in on the Web!

"Flat Files" -- A Simple Model

Basel 2012-10-06 Batch Part 3 (31)

A Sample Domain

  • We start with a classic example: The Address Book
    • People and information about them
    • Names and contact information
  • We can do a first cut as a diagram


For example

  • Bijan!
    • Name: Bijan Parsia
    • Company: University of Manchester
    • Email:
    • ...
  • Uli!
    • Name: Uli Sattler
    • Company: University of Manchester
    • Email:


  • Slides are not a good storage place for data
  • We have an array like structure so...
    • How about a spreadsheet!
      • 1 entity/record/person per row
      • Each field/attribute is a column
  • We have software that works well with this!

Interacting with the data

To the demo!

Pain points

  • Around "name"
    • Sorting is on columns
      • Cannot sort by surname
    • Filtering: can filter by names beginning with Z
      • Cannot filter by surnames beginning with Z
  • Around "address"
    • Can't sort or filter by postcode
    • Can't sort or filter by city
    • Can't sort or filter by county

Problems with spreadsheets or our format?

Format 2

  • This should fix our pain points!


Demo encore!

New Pain Points

  • Variable numbers of the "same" attribute
    • Phone number
    • Email address
    • Web page
    • Inserting columns is painful
      • Lots of partial columns
      • Sheer number sucks
  • Companies have addresses!
    • More than one!
    • And phone numbers, etc.

More problems with our format

NOT a New Format

  • Not a fix to our format:

Fixing the Format Again

  • We want adding a (similar) column to be easy!
    • Easy as adding a row!
    • Make a new table just for phone numbers
    • Index numbers with person rows

Format 3

  • Now this should fix our pain points!

Still Pain Points

  • Sorting destroys the relationship
    • We used row numbers to connect
    • Sorting changes the row number!
  • Hard to see the record
  • No longer a simple flat file
    • CSV format makes assumptions

These are (mostly) implementation problems!

Analyse Format Failure

  • Did we
    • get the domain wrong (addresses)?
    • fit it wrong into our core DM (tables)?
    • pick the wrong core DM to model it in?
  • Is our format
    • unworkable?
    • workable but requires a lot of application code?
    • reasonable with some workarounds?
  • How much technical debt are we piling up?
  • What's the cost of switching?

Unsuitable Core Data Model

  • If you are
    • always "fighting" the system
    • use lots of application code to hack things
    • live in an error rich environment
    • have increasing amounts of workaround support in your data

Your core data model might not be a good fit for your domain and application!

The Rest of the DBMS

  • Even if your core DM isn't a good fit, you might
    • be stuck with the system
      • You paid good money for that Oracle database!
    • need features of the implementation
      • is there an XML database with transactions?
      • what's the support contract?
    • be stuck with the model (critical legacy apps)
  • Just because the model is broken doesn't mean that the system is
    • Or is broken enough to justify a switch

Flat File Programming

Sharing Our Databases

  • Spreadsheets?
    • Propriatory-ish (Excel, Google Doc, OpenOffice)
  • Lingua franca: CSV
    • Comma (or Tab) Delimited Values
    • Exactly the (pure) flat file model
    • Format: text file
      • 1 record per line
      • First line can be special (column names)
      • Each column separated by a ","
        • We may need to quote cells (with commas)

CSV Example

Programmatic Manipulation

  • If we store our databases as CSV
    • We can load and parse them into structures
    • Manipulate our data from our programs
  • E.g., using Python
import csv
with open("../Adresses/mod2-uk-500.csv") as csvfile:
    line_count = 0
    myreader = csv.reader(csvfile, delimiter=',', quotechar='t')
    for row in myreader:
        if line_count == 0:
             line_count += 1
            print(f' Candidate {line_count}: Firstname {row[0]} Lastname {row[1]} City {row[4]}')
        line_count += 1
print(f'Processed {line_count -1} Candidates.')

Solving problems

  • This solves some problems!
    • Inserting/removing columns a "small matter of programming"
      • Or we could use multiple arrays with pointers
    • We can split/combine fields at will
      • Well, with a bit of programming
    • We can control sorting well enough
      • Use pointers to connect
  • Lots of work!

Against Bespoke Programming

  • This is all at the wrong level
    • Flat files and flat file++ are ubiquitous
    • We shouldn't be coding complex functions
      • Over and over again!
  • Even if we can program our way around problems
    • Doesn't eliminate the problems
    • Some solutions (pointers) effectively change the core model: no longer flat files!

A Relational Model

Relational model concepts


  • A core DM where table (or relation) is the core data structure
    • A table is a set of tuples
    • A tuple is
      • an n-ary sequence
      • a set of key-value pairs
  • Flat file had one table
    • We allow many!
    • Named tables
    • Aka relations


  • (We use table and relation interchangeably)
  • Relations are like First Order Logic (FOL) predicates
    • Relation name = Predicate name
    • Number of columns = Arity of predicate
      • Person(bijan, u_o_manchester, ...)
    • Predicate is true (or false!) of its arguments
      • Relation is "true" of tuples which occur in it
    • Predicates can have
      • definitions (intensional!)
      • facts (extensional!)

Order and Identity

Records/Rows/Entities need identity

  • In Excel, we had the row label
    • the order or position of a record was significant
  • In our model, we need distinguishing attributes
    • we push identity into the data: a key
      • either a "naturally" unique set of attributes
      • or a made up one: an ID
  • Order is always a property of the
    • data values
    • implementation

Multiple tables

  • Actions on multiple tables:
    • Splitting at
      • design time: try to normalize your DB
      • run time: dropping bits
    • Combining
      • Take two tables and produce a new table
  • The key to relational domain modelling
    • Decompose your problem into "base" tables
    • Derive new tables for specific needs

A Relational Formalism


What is a formalism?

  • A formal system (or formalism):
    • syntax: what can we write?
    • semantics: what does our writing mean?
    • with precise (mathematical) definitions
    • designed to capture a coherent set of operations
    • ("syntax" is loose, e.g., we might just have a collection of operators)

Key goals of a formalism

  1. to be clear about what we mean
    • In our spreadsheet is "1" a number, a string, either, both, something else?
  2. to allow the determination of key properties
    • e.g., complexity of query answering
  3. to abstract away from particular implementions
    • e.g., allow us to determine when wildly different implementations are correct thus can interoperate

Formalism vs. Language

  • Formalisms are often abstract
    • This can be an advantage!
    • Can be hard to use if only abstract
    • Concrete instances typically involve compromise
  • We focus on concrete languages
    • Formalisms are the theory
    • Languages are the practice
    • Other Quotes On Theory vs Practice
      • Well, it may be all right in practice, but it will never work in theory.
      • In theory, there is no difference between theory and practice. But, in practice, there is.

SQL: A Language For Tables

  • Schema
    • CREATE TABLE table_name
  • Update
    • INSERT INTO table_name
    • DELETE FROM table_name
    • UPDATE table_name
    • ...
  • Query
    • SELECT ... FROM table_name

SQL operations (largely) are closed over tables

An Infelicity

There is a lot of lingo with slight different meanings. Concepts get divided up in slightly different ways.

Our talk Common Learning SQL p.10
Core Data Model
Data Integrity Data Definition SQL schema statements "CREATE"
Data Manipulation Query/Update
SQL Data statements

A Sample SQL Program

    name varchar(255),
    company varchar(255),
    address varchar(255),
    phone varchar(255),
    email varchar(255),
    home_page varchar(255));

    VALUES ('Aleshia Tomkiewicz', 'Alan D Rosenburg Cpa Pc', 
            '14 Taylor St, St. Stephens Ward, Kent CT2 7PP', 
SELECT name FROM People
  • You must Define before Update before Query
    • I.e., CREATE before INSERT before SELECT

Modelling with SQL

  • SQL lets us express models at the logical to (some of the) physical level
    • Specifying indices is a bit physical
    • Knowledge about implementation may inform modelling choices
  • SQL has no mechanisms for conceptual level

Format 1 in SQL

Format 1 in SQL

    name varchar(255),
    company varchar(255),
    address varchar(255),
    phone varchar(255),
    email varchar(255),
    home_page varchar(255));

    VALUES ('Aleshia Tomkiewicz', 'Alan D Rosenburg Cpa Pc', 
            '14 Taylor St, St. Stephens Ward, Kent CT2 7PP', 

Can we do all that we did in the spreadsheet?

SQL Manipulation of Format 1

  • Count records in your People table:
  • Search for items:
    • SELECT *  FROM People
      WHERE name like 'Aleshia%'
    • SELECT *  FROM People
      WHERE name like '%Tomkiewicz'
  • Sort the table!
    • SELECT *  FROM People
      ORDER BY name asc

Format 2 in SQL

Format 2 in SQL

    first_name varchar(255),
    surname varchar(255),
    company varchar(255),
    street_address varchar(255),
    city varchar(255),
    county varchar(255),
    post_code varchar(255),
    phone varchar(255),
    email varchar(255),
    home_page varchar(255));

    VALUES ('Aleshia', 'Tomkiewicz', 'Alan D Rosenburg Cpa Pc', 
            '14 Taylor St', 'St. Stephens Ward', 'Kent', 'CT2 7PP', 

SQL Manipulation of Format 2

  • The old queries work, but we can improve them
    • Search for items:
      • SELECT *  FROM People
        WHERE first_name = 'Aleshia'
      • SELECT *  FROM People
        WHERE surname =  'Tomkiewicz'
  • We can recreate Format 1!
    • SELECT first_name || " " ||surname as name, 
      street_address || ", " ||city  ||", "|| county ||" " || post_code as address,
      FROM People

Format 3 in SQL

Format 3 in SQL

     person_id SMALLINT UNSIGNED,
    first_name varchar(255),
    surname varchar(255),
    company varchar(255),
    street_address varchar(255),
    city varchar(255),
    county varchar(255),
    post_code varchar(255),
    email varchar(255),
    home_page varchar(255),
    CONSTRAINT pk_person PRIMARY KEY (person_id));

     person_id varchar(255),
     number varchar (255),
     CONSTRAINT pk_phone_number PRIMARY KEY (number));

    VALUES ('1','Aleshia', 'Tomkiewicz', 'Alan D Rosenburg Cpa Pc', 
            '14 Taylor St', 'St. Stephens Ward', 'Kent', 'CT2 7PP', 
    Values ('1', '01835-703597')
    Values ('1', '01944-369967')

SQL Manipulation of Format 3

  • Recreate Format 1 and Format 2: easy
  • Find everyone with same phone number
  • Can we have unassigned phone numbers?

How did our formats do?

  • Core DM/Data structure: Tables seem to work!
  • SQL and Relational Model
    • We can do everything!
      • All queries in all models
      • Format 3 has 2 tables/requires joins
  • Format 3
    • Neater inserting and deleting
      • Can have as many phones as you want!
    • Every other domain model can be derived
      • Just write the query!

Expressive Power

  • SQL is expressive
    • The core data model is rich
      • Composing and filtering tables does a lot!
      • Operators and functions helpful
        • Without concat(...), there'd be trouble!
    • The language is powerful
      • Reasonably composable
      • Lots of features
      • Extended & extensible in many implementations
        • Interop problems!

Querying with SQL

Schemas vs. Queries

  • CREATE statements
    • "create" empty tables
    • out of nothing at all
    • with certain constraints
    • with some expectation of permanence
  • SELECT statements
    • "generate" new tables (possibly with data)
    • out of existing tables
    • according to some constraints
    • with no expectation of permanence

Closed Over Tables

  • SQL is (mostly) closed over tables
    • Most SQL constructs take & produce tables
    • Clear exception: Functions!
  • Manipulation is manipulation of tables
    • Not rows, columns, or cells directly
    • Rows, columns, and cells are "degenerate tables"...


  • Key operation SELECT: ignoring some parts
    • Basically "find"
    • Can filter rows or columns or both
    • Requires "testing" functions on values

Filtering Columns

  • aka "Projection", specified in SELECT clause
    • Keep all columns:
        SELECT * FROM People
    • Just a single column:
      SELECT county FROM People
    • Multiple columns:
      SELECT name, county FROM People
    • Rename columns:
      SELECT street_address AS address  FROM People

Filtering rows

  • Selecting specific tuples
  • Specified in the WHERE clause of your query:
    • Equality:
        SELECT * FROM People
        WHERE surname = "Smith"
    • Range:
      SELECT * FROM People
      WHERE heartrate > 95
    • Compound criteria:
      SELECT * FROM People
      WHERE heartrate > 95 AND county="Kent"

Building Tables with Cross Join

  • The fundamental operation is Cartesian product
    • T1 x T2
    • for example People x Phone
  • Makes a new row for every pair of rows from T1 & T2
    • What's the size of the result?
  • Not really a user-oriented feature
    • "Incidentally" cross joins are dangerous!

Building Tables with Inner Join

  • An inner join is a join filtered on common columns
    • Useful for our phone records!
      SELECT * FROM People, Phone
      INNER JOIN ON People.person_id = Phone.person_id
  • The above is special case, called "natural" join
    • can be written as follows:

Building Tables with Outer Join

  • An outer join is like an inner join but it returns also rows that do not have a match in the other table
    • left outer different from right outer
SELECT * FROM People, Phone
RIGHT OUTER JOIN ON People.person_id = Phone.person_id
  • will return also people who have no phone!

Building and Filtering

  • Once we've built a table we can filter things we need:
SELECT * FROM People, Phone
RIGHT OUTER JOIN ON People.person_id = Phone.person_id
WHERE People.surname = "Smith"
  • knew that already!?

The Cost

  • A key issue with joins
    • Worst case for their computation is a CROSS
    • Even if you don't generate the CROSS
      • You might have to consider all the pairs
      • (If you aren't careful)
  • Good optimisers avoid both
    • Considering lots of matches (think indexes)
    • Generating large intermediate tables

Incomplete Data

Puzzle black-white missing

Multiple Phone Columns

  • Some people have none or one
  • Or no email or web page

No Surname

  • Even if we normalised that away
    • Some people don't have a surname!

Madonna 3 by David Shankbone-2


  • null is a distinguished value which can mean:
    • "Value not yet known"
    • "Not applicable to this entity"
    • "Value undefined"
    • check out LSQL
  • Key property: Unequal to everything
    • null = null is never true
    • Match on not null, rather than null

Strange value!

Outer Joins

  • If you have no nulls in your base tables
    • you can't get them in tables derived by inner join
  • However, the 2 phone column table is derivable
    • We use the outer join
    • Outer joins take a table T
      • for each row in T
        • extend it with the (projected) columns from another table
        • If there's a match, add the matched values
        • *else, add nulls
  • See Learning SQL Chapter 10 for examples

Null proliferation

  • null never matches
    • So iterated outer joins proliferate nulls
      • As you get wider, you get sparser
        • If you are matching on a sparse attribute
  • nulls pose challenge for relational theory
    • And somewhat for practice
    • Starts moving from the sweet spot

SQL and the Web

A brief tour

Psalter World Map, c.1265

SQL driven Websites

  • Many websites are backed by a database
    • PHP makes it easy
    • Consider WordPress and other CMSs
  • Lots of unstructured content
    • Stuff in blobs and text fields
  • Key properties
    • Scaling
    • ACID: Atomicity, Consistency, Isolation, Durability
      • Transactions
    • Concurrent access

There is a key historical text that is still good reading,
esp chps 11-12

CSV & SQL programs on the Web

Google Query Viz Language

  • A SQL like language
    • Used in Google Docs Spreadsheet
    • QUERY function takes queries as argument


The WhatWG and W3C tried to standardize WebSQL

This specification introduces a set of APIs to manipulate client-side databases using SQL.

function prepareDatabase(ready, error) {
  return openDatabase('documents', '1.0', 'Offline document storage', 5*1024*1024, function (db) {
    db.changeVersion('', '1.0', function (t) {
      t.executeSql('CREATE TABLE docids (id, name)');
    }, error);

Local database backed web apps

  • For offline use
  • Just increased capabilities

What is this data?

  • A recurring issue: what is in this shared document?
    • csv
    • table
    • JSON snippet
    • ...
  • What does it mean?
  • How to parse?
  • How to share? So that it's good to use?
  • Self-Describing and Meaning will be discussed at length

Next Steps

Impossible staircase


There is a key historical text that is still good reading,
esp chps 11-12

Any questions so far?

Labs & Coursework

  • Next, we go to the Labs
  • You look in BB at Week 1 coursework:
    • Quiz Q1
    • Short Essay SE1
    • Small Modelling exercise M1
    • Some querying CW1
  • Read, think, ask us!