Depriving Amazon

Shopping experience or noise machine?

Claudio Agosti, Jeroen de Vos, Alicja Zak, Elise Olthof, Zdzich Heydel, Ola Bonati, Dimitri Koehorst, Aymann Khatib, Alexander Bernevega, Cem Akca, Boy Singmanee, Shivaani Gore, Yonathan Tesfai, Deirdre Murphy, Romane Donadini, Margaux Reynders, Joana Stockmeyer, Hailey Beaman, Tommaso Campagna, Xiao Wang, Aikaterini Mniestri, Matteo Bettini, Beatrice Gobbo AND MORE :-)

Introduction

• Understanding personalisation algorithms
• Facebook, Youtube, now Amazon
• Developing as-we-speak

​

Research Questions

• The epistemology of shopping experiences

​

• How can we understand Amazon’s personalisation algorithm?

​

• What data is taken into consideration?

​

• How are online shopping spaces sensitive to gender and location?

​

Dynamic pricing

• More than 1.000 products changed price in three days
• No clear determination why

• ​

Operating Systems

RQ

• Do users’ operating systems affect Amazon search results?

​

Methodology

• Cleared browser and Amazon search history on Brave.
• Set the default language to English and default currency to USD.
• Created specific search tags for Windows and Mac devices (OS_Win_Clean) & (OS_Mac_Clean)
• Performed queries on two Windows and two Mac devices for the following items:
• laptop
• smartphone
• tablet
• wireless earbuds

Repeated 5 times

Findings

Old Win

New Win

Old Mac

New Mac

Product

Class Warfare

How does perceived purchase power influence search results and prices on Amazon.com?

Price - Finding

Price - Finding

​

Tracking

Research question:

To what extent are Amazon’s search results influenced by user’s external browsing history in regards to two preselected topics?

PHASE 1:

Clean browser

Pollution (1 hour) - browse the web for music/sports.

Amazon data collection: [backpack], [jacket], [shoes], [phone case], [smart watch].

PHASE 2:

Clean browser

Pollution (15 minutes) - SUN, IMDB, REDDIT, WIRED (all are tracked by Amazon).

Amazon data collection: [backpack], [jacket], [shoes], [phone case], [smart watch].

Tracking

Original price

Product diversity

• RQ: To what extent is the diversity of products on Amazon.com reflected in different regions?

​

• Clean, logged out account searches using Brave with delivery regions: Northern California, Southern California, East Missouri, and West Missouri

​

• Queries: candle, coffee, lamp, nail polish, mouse, and playstation

​

• Screenshotted Amazon pages and used amTREX tool to collect data in CSV

​

• Imported collected data into both Tableau and Gephi to create visualisations

​

​

Product diversity

Gendered shopping

METHODOLOGY 1.0

RQ: To what extent are Amazon search results gender coded?

PHASE 1: Establish baseline of search results

PHASE 2: Test the effect of personalization

PHASE 3: Repeat Phase 1 with personalization

​

​

​

​

​

​

Methodology

​

Methodology 2.0

RQ: How do results vary based on different Amazon algorithms?

• If one picks gender coded items, does Amazon make gendered recommendations?

PHASE 1: Used Product description to determine whether items in the cart recommendations were explicitly gendered

​

​

​

​

​

​

1. Search results were not personalized or gendered to the extent that the “Recommended Products” section was�
2. Recommended products were personalized and gendered in a way that was consistent with browsing history

3. Certain items (or perhaps search terms) are gendered as shown in the search results, regardless of the personalization process.

Discussion

• Sensory deprivation or the noise machine?
• Threshold to personalisation
• AlgorithmS exposed

​

Conclusion

• Personalisation is playing out on variety of levels
• Cannot be isolated from its ecosystem, such as products typology