oDCM - Opening Lecture

We're about to start with the first lecture of this class.

If you haven't done so, please

• today's slides are available here: https://odcm.hannesdatta.com/docs/modules/week1
• explore other sections of the course page at https://odcm.hannesdatta.com.
• complete the software installation (https://odcm.hannesdatta.com/docs/modules/prep/)

• Part 1 (10.45 to about 11.45)
  • Getting to know each other
  • Motivation for the course
  • Course framework and learning goals
  • Agenda and practical arrangements
• Break
• Part 2: Python Bootcamp on your laptops (about 12.00 - 13.30/13.45)

• You will learn how to write code that automatically downloads and structures information from the internet for the purpose of (scientific) analysis.
• We call these programs “web scrapers” (for any internet pages) and “APIs” (for official data access)
• Web scraping are the foundation of Google Search (“web spiders”) and ChatGPT (e.g., for training); APIs are at the core of many business models (e.g., Twitter API - back in the days; OpenAI API)
• I also almost got sued doing scraping (more about it later…)

• This is predominantly about web scraping and APIs - while I teach a bit of Python, becoming an expert requires years of practice
• Mix of students at various levels (e.g., beginners, advanced Python users)
• You can also extract web data using other software packages (e.g., R)
• I will record sessions and post content immediately after class (but, attendance is strongly encouraged)
• Consider me your coach, not your distant professor (yes, I respond to WhatsApp messages at +31134668938)
• Slow me down if you need to

• scraping nerd — learned it in 2008 using Visual Basic in Excel
• started doing my own research with scraped and API-extracted data in 2012 (so, 10+ years experience)
• left Germany around your age, now 15+ years in NL
• Associate Professor at Tilburg University

• Substantive interests

  • streaming business models (e.g., music, movies)
  • marketing-mix modeling and optimization
  • open science

• Methodological interests

  • online data collection via APIs and web scraping
  • causal effects with observational data

• MSc Marketing Analytics
  • Data preparation and workflow management (https://dprep.hannesdatta.com)
  • Online data collection and management (https://odcm.hannesdatta.com)
  • MSc Thesis supervision (https://thesis.hannesdatta.com)
• Other initiatives
  • Tilburg Science Hub (https://tilburgsciencehub.com)
  • Tilburg.ai (https://tilburg.ai)
  • Music-to-scrape.org (https://music-to-scrape.org)
  • YouTube channel (https://youtube.com/c/hannesdatta)
  • My code (https://github.com/hannesdatta)
  • More about my work (https://hannesdatta.com)

• What's your background - previous education (e.g., program)?
• Any experience in Python (or other programming languages)?
• What are your passions & talents? (+ why I am asking you this…)

• started out as a PhD student without data
• was interested in music, and found website with data (https://last.fm)
• no best practices in scraping; learnt all by myself and made many mistakes
• scraping was undervalued in academic job market, but, key role in shaping relevance and rigor of your work
• now scraping and APIs are a large part of what defines my research

• scraped reviews at Amazon.com
• investigated (how music consumption changed with Spotify)
• monitored Spotify's user interface and make lists of new releases
• found a way to estimate “power imbalances” in the music industry
• share data on the playlist ecosystem with APIs
• downloaded 100k+ images from Amazon, ran Image and Text Analyses using Google Cloud NLP and Vision APIs
• investigating “video streaming wars” (e.g., Netflix versus Disney+, when do consumers use multiple streaming services?), see trakt.tv
• share code on GitHub
• developed a methodological framework on scraping/APIs (which we use in this class)
• faced legal battles…

With web scraping, you can capture anything you can view in a web browser

• pricing data at bol.com
• reviews at Amazon.com
• movie data at imdb.com or reviews at Rotten Tomatoes

With APIs, you obtain official data from a firm in a programmatic way

• e.g., as a developer, interact with Instagram, Twitter/X, ChatGPT / OpenAI, AWS, …
• as a researcher, construct data set from analytics firms

• for businesses
  • “stitch together” different services (e.g., augment functionality of ChatGPT)
  • do market research (e.g., pricing data; see Zyte)
  • initialize recommendation systems (e.g., music metrics)
• for research
  • discover/document novel phenomena (e.g., new platforms/technologies)
  • improving methods (e.g., get data to try out new methods, such as review data & text analysis, images, videos)
  • improved inferences (i.e., get more accurate results)
  • collect metrics managers care about

• What’s the last app/website that made you say ‘wow’?

• If you could only use three apps/websites for the next year, which ones would you pick and why?

• What's a niche online community you're part of that most people don’t know about?

• What’s the last thing you saw on TikTok that made you stop scrolling?

• Imagine an AI tool that could make you internet-famous overnight. What would it do?

Let's talk about it right now…

Why do we need a course on this? Isn't this how research is always done?

Yes, but collecting web data is different from other datasets!

• data source selection
  • finding the right data source may be difficult as many potential alternatives exist
  • differ in delivery formats (website versus API, compared to CSV/databases)
  • access to data that is not available commercially (or that a firm would not like to share)

• extraction design
  • which information to select (which is available?!)
  • type of variables (sales is rare; review scores are abundant)
  • different stakeholders that could potentially be addressed
  • need to tackle legal and ethical issues

• collecting data at scale!
  • unprecedent: it's totally automatic (but prone to errors)
  • need to put monitoring procedures in place
  • data is not documented, and there is no direct way to ask questions about the data (sampling?! generalizability?)

Each project is totally unique - that's why there is no universal “best way” to approach things…

• Explain how to use web data for creating marketing insight
• Select web data sources and evaluate their value to inform a specific research context or business problem
• Design the web data collection while balancing validity, technical feasibility, and exposure to legal/ethical risks
• Collect data via web scraping and Application Programming Interfaces (APIs) by mixing, extending, and repurposing code snippets
• Document and archive collected data and make it available for public (re)use

• Weekly modules, structured along the methodological framework

  • On-campus lectures and tutorials (Hannes)
  • On-campus (sometimes online) coaching sessions (Roshini)
    • ~45m check-in with all teams
    • ~45m “free floating” questions
  • Self-study (own tutorials (!), Datacamp, readings)

• Project in which you put into practice your skills (40% of your grade)

• On-campus computer exam (60% of your grade)

• Specifics

  • scrape data via web sites or use APIs
  • work through the method framework of Boegershausen et al. 2022
  • receive feedback by students and Roshini (your coach)
  • inspiration for the setup: past projects
  • should address a broad research context

• Evaluation

  • grading rubric available [show website!]
  • self- and peer assessment used
  • 5 group members (4 is the exception!)

Visit https://odcm.hannesdatta.com!

• It's open education, so spread the word!

• Course website is your #1 resource, Canvas used for

  • posting announcements,
  • signing up for teams,
  • submitting (weekly) deliverables,
  • receiving grades.

• Do all students have access to Canvas?

• I encourage the use of AI to help you learn Python and web scraping/APIs efficiently
• Note though that, on the exam, you've got to be 'good enough' to do it without AI
• Make use of…
  • Course Chatbot at https://odcm.tilburgai.nl (old); https://chatbot2.testing.tilburgai.nl/ (new)
  • MANY tutorials and extra resources at https://tilburg.ai
  • GitHub Copilot for Education (“AI programming guide”)

• Take time to become acquainted with the course (e.g., not using Canvas)
• Can be tough at first, but you will gain experience rapidly!
• It can be overwhelming to follow dPrep and oDCM simultaneously, but the skills you learn are important for your course work, Master Thesis & future career!
• TWO exam moments per academic year - choose what suits you best (but: oDCM will be given annually after the summer)
• Software
  • Open (but geeky) (usable right away, no admin rights required)
  • Computer exam on Windows computers; all students are encouraged to explore using the software on the computers on campus (e.g., library, computer rooms)

• Easing your start-up pain
  • Start preparing early on (the first weeks will be the most challenging!)
  • Have the same group members across dPrep and oDCM
  • Collaborate with each other and try to help one another, also across teams!
  • Be in touch via WhatsApp for Business (add +31 13 466 8938)

• Coding can be extremely frustrating if you're starting out
• I tend to become semi-“brain-dead” after a day of coding
• Take breaks! Stop coding. Go for a run. Start again.
• You will learn from your mistakes
• Use ChatGPT, cheat sheets and our support section

→ quick feedback loops in first few weeks

When you run into trouble, this is your way out!

1. Use the course's Chatbot (https://odcm.tilburgai.nl)
2. Try to find the info on the course website or course material
3. Ask ChatGPT / Google / Stackoverflow
4. Ask friend/classmate (form learning groups)
5. Can it wait? Defer to lecturer or feedback sessions.
6. If it can't wait: be in touch with me (–> next slide!)

• Investment in research skills

  • e.g., collect own data & shape its creation to create relevant and rigorous research

• Essential skills for entrepreneurs

  • e.g., have your own company on the basis of web data and APIs
  • be the linking pin (interface between marketing and analytics)

• Showcast expertise in coding

• We will continue with our Python bootcamp after a break
• Haven't followed the installation guide? Check it out on the course page now! (modules –> preparation)