Algorithmic Play
Education and Digital Culture 2020
University of Edinburgh-Digital Education
s1891551
The Amazon™ A9 Algorithm
I set about testing Amazon's™ A9 algorithm which bases its search results and recommendations on data from thousands of queries that people make every day.
The recommendations are generated through a mix of customer preferences and past purchases, together with many other factors which I tried to determine from accessing the following link.
I also tried to determine how recommendations are made by placing myself in the shoes of a seller rather than a customer. I did find some useful information on the site below which allowed me to generate an image of how products gain popularity in Amazon™.
Amazon's™ most popular products are the ones that make it to page 1. These are products that have the highest ranking or the ones that sell most. Rank is as important as selling and sellers will aim to market their product in order to achieve as much rank as possible.
A number of things affect rank. Product reviews, keyword ranking, sales incentives and even the conversion rate might promote (or not) a product to page 1. When promoting a product, sellers must be careful to use the right amount of description, keywords, photos and user-friendly.
Page 1 products are the ones that sell and the chance of selling drops exponentially by 30% from page 1 to other Amazon™ pages. Products that are popular usually sell within the first couple of minutes an online shopper accesses the product. Reviews are also very important as many people rely of reviews to choose a product over another.
How does the A9 algorithm work
My Shopping Test
Timing
Recommendations
I decided to test Amazon's™ algorithm using three diverse products: an Aquarium Heater, an iron and a laptop. I got most recommendations for aquarium stuff, with some interesting items.
Upon testing, it seems that for a product to be included in the Amazon™ browsing history, a user needs to spend at least 4 seconds on the item page once s/he has clicked a particular product. Anything less will not be included.
The screenshots below show some of the recommendations Amazon™ generated for me. Most recommendations were generated on aquarium paraphernalia as I spent more time searching this topic than anything else.
My Search Test
I decided to perform a simple product search in order to determine which products are most popular and appear at the top of the search results. I used a simple 'fruit' search as I was aware that fruit names are used as product brands. I first typed in 'ORANGE' and got this:
I then changed fruit and typed in 'Apple' and 'apple' and was provided with a list of 'Apple' products. The concept of apple as a fruit/food does appear but only twice in a list of ten products.
I finally tried a different fruit altogether...a 'Kiwi' and did not get one search item related to a fruit. It seems that fruit names as product types and brands are very popular. The search also goes to show what is popular and better ranked.
How has the algorithm affected the options I was given or what I can see?
Amazon's™ complex A9 algorithm which is based on the ranking system explained above, presents results of products that have better ranking according to buyer preferences and item popularity. Results are also based on searches from thousands of customers. The fruit example above displayed results according to product popularity and ranking.
Once a particular item is selected for more detail, a list of similar products or items is then added to the recommendations list or 'What other customers bought' list, in an effort to sell me more stuff by showing what others are doing.
Recommendations are selected on the same basis.
While Amazon™ users might use the platform to search specific products, Amazon's™ first page provides a selection of the most popular and ranked items for those buyers who have nothing specific in mind but feel like buying something.
How have other people been involved in shaping results?
How have my actions changed what the algorithm has done?
My searches and actions have been the catalyst for Amazon's™ algorithm by generating 'data points' (Williamson, 2017) that are stored and evaluated. By choosing a particular item in the midst of thousands, I have allowed the algorithm to determine my needs and provide me with alternatives, additions and choices related to my hobby, in this case, aquariums. Similarly, it performs this every time a search query is made by others.
What may seem passive browsing is, in fact, an exercise which triggers the algorithm into providing specific results which have an impact on me and others. By simply spending time on a product, that product was added to my recommendation list and generated other similar products to be associated with it.
Amazon's A9™ algorithm has provided me (and many others) with one of the most vital tools when making decisions. It has shown me what other customers are choosing and what is most popular in terms of price, quality and brand. This has inevitably influenced my choice when presented with the most popular products first. If a product is popular then 'it must be better!'.
Just like other people's choices have influenced mine, so has mine influenced theirs. Other peoples' associated products similar to mine are a 'nudging' attempt by Amazon™ to sell me more or, perhaps, to steer me towards a particular set of products. Not only does Amazon™ recognize my needs and presenting me products to satisfy them but also encouraging me to buy more than I want because other people have done so.
What are the ethical issues at stake with your chosen algorithm? Is there data here that should be private?
Although I was well-aware of the data mining going on while I was browsing through the products on Amazon™, it was still hard to imagine that Amazon™ was 'getting' an impression of me by monitoring my behaviour. Data that might seem insignificant to most of us, such as how much time we spend browsing an item or if we buy anything within the first few minutes of accessing the platform, 'can have broad implications in regards to our beliefs and ideologies' (Giffin, 2019). My tendency's and shopping patterns can help determine my political leanings, social class, level of education and much more without me being aware of it (Giffin, 2019).
While it came almost naturally for me to accept the terms and conditions of Amazon's™ services, I am still unsure when it comes to when and how my data will be used. Privacy Policies tend to be long and take time to read so people tend to skip reading them. This seems strange in a time when GDPR regulations push data controllers to provide information to data subjects of how personal data will be used.
Perhaps some may argue that hovering over an icon or spending more than four seconds on an item is not exactly personal data but if results from this 'weak data' can be extrapolated to produce a digital persona of myself with specific likes and behaviours, then things do get personal.
Do results feel personal or limiting?
I believe that Amazon's™ A9 algorithm fits the three-step model described by Pariser in his 'You Loop'. It initially collected data pertaining to what I wanted to buy, compared it with similar and popular products and provided choices. It took into account my choice and forwarded other options I might be interested in.
What are the implications for digital education implied by your chosen algorithm?
Images obtained and modified from OpenClipart-Vectors and Clker-Free-Vectors at https://pixabay.com
The same strategies used to monitor buyer preferences are being used to 'personalize' the learning experience through a process known as datafication (Williamson, 2017). This is a process that quantifies learner data by evaluating how many times a learner takes a particular test, what marks are achieved and how much time was taken on a particular module. Modern technologies have allowed this data to be collected in 'real-time' via digitized e-learning platforms and stored in huge databases in order to provide tailor-made learning experiences.
What are the ramifications of datafication?
According to Williamson (2017), datafication and digitization are shaping the way curricula are designed and policies made. Startup schools, learning laboratories and charter schools are prototypes of educational scenarios created by Silicon Valley entrepreneurs which are heavily funded to produce those positive results which classic education sometimes falls short of (Williamson et al, 2018). The methodologies and practices in these types of centres are then hailed as innovative and marketed as solutions to educational inertia.
Even curriculum content is being questioned 'as new kinds of 'adaptive' learning software are developed that can semi-automate the allocation and 'personalization' of content (Williamson, 2017). Standardized learning content and testing is being supplanted by online courses and platforms that provide learners with programs that can be taken anywhere and at any time.
Following the experience a couple of weeks ago of participating in a MOOC, I can observe how strategies used by Amazon™ can easily fit the pattern of learning platforms like EdX or Coursera in the following ways:
-
courses are advertised by the amount of people taking them.
-
regular updates are sent via email such as reminders of progression and new posts
-
flexibility of learning is promoted
-
'free' learning is advertised even though a certificate needs to be paid for.
-
feedback is immediate as in the case of quiz marks.
These new educational scenarios ' represent the next step in the 'corporatization of public schools'-not just the 'transformation of the school on the model of the corporation (Saltman, cited in Williamson et al, 2018), but more specifically the transformation of the school on the technical, economic, cultural and scientific model of Silicon Valley itself. (Williamson et al, 2018).