Airbnb’s process of system learning uses algorithms to show desirable housing to potential guests. However, it can be problematic in matching guests that are potentially too similar, and therefore, the algorithm should perhaps only take into account availabilities and space preferences.
The old YouTube algorithm appears familiar-it uses the same types of filters that amazon currently employs and pioneered in 1998. The new algorithm, though, uses neural networks and other forms of machine learning, which is a much more highly sophisticated method. The post itself shows the problems with this advanced method, including the bucket problem, recommendation loop, and frequency bias. Desirability here is therefore used in a context of attaining an appropriate video recommendation for any given user.
Political Preference Algorithm (http://designsurvivor.tumblr.com/post/157178738560/the-political-preference-algorithm)
The political preference algorithm takes psychographic principles and profiles people based on their backgrounds. The Trump campaign used this to create their own definitions of desirability and win the election. Obviously, these preferences can be tailor-made for any potential advertiser, so this is a genius method. It’s based in actual psychology, and has proven results, so I would not want to change this algorithm.
Redfin Hot Homes (http://designsurvivor.tumblr.com/post/157174878930/redfin-hot-homes-algorithm)
Redfin’s algorithm gives space to very attractive houses, likely to be sold within two weeks of going on sale. It supposedly tabulates over 500 characteristics of the house and interactions with the house on Redfin itself considered desirable to determine which houses are “hot.”However, as some of these are self-reported, and they don’t guarantee that the house will sell quickly, so improvements could be made to the algorithm-perhaps with more inputs from people, house experts and consumers alike, to more accurately predict these hot houses.
Pinterest caters to the just-under-40, female crowd that does the most amount of online shopping, and appears to value handmade, vintage, or otherwise “quirky” products. The algorithm seems rather self-explanatory, being based on your interests, people you follow, and otherwise popular/high-rated pins found elsewhere. These pins aren’t as temporal or popularity-based as Instagram or Facebook posts, so it’s more justified to show higher quality posts, as opposed to newer ones. For the audience Pinterest targets, this is justified and appears fine.
Instagram recently changed to showing users photos in order of perceived relevance, which self-perpetuates the notion that certain popular peoples’ posts are more desirable than other people you may follow. They also used Snapchat’s “story” feature to bring updates from people who are very active on Instagram to the front of your page. This may work for some, but for me and others whose posts are less likely to be discovered due to either small amount of posts or less friends than celebrities, this is problematic. I would revert the algorithm to the original one.
Amazon (http://designsurvivor.tumblr.com/post/157173765210/amazon-preference-ranking-algorithm, http://designsurvivor.tumblr.com/post/157162389620/amazons-preference-system)
Amazon’s Preference System is aligned to not only consumer preference, but to increasing company profits. This is obviously beneficial to the company, and creates a definition of desirability that emphasizes the needs of the company over the needs of the customer, which can be potentially harmful. In other aspects, this algorithm counters Spotify’s, as it works on an item-per-item basis, and appears to consider recently viewed items over long-term trends of purchases. While I believe this is a poor algorithm, as it never seems to work for me, the advent of adding subscription options and the ability to circulate packages through the delivery network preemptively so they get to customers first, certainly works in Amazon’s favor. Addendum for first post: A9 manages the majority of queries, and the overall algorithm appears to work more in the short-term, which works well for impulse buyers.
Spotify aims to create a totally curated and personalized, seemingly-human experience, meant for those who love new music and streaming music wherever they are. Doing this as successfully as they do is a feat in itself. Their algorithm is very impressive, and is multifaceted, taking into account “collaborative filtering,” the “ensemble method,” and “outlier detection.” These all borrow directly from other users’ libraries to tailor music, rather than relying on a song-by-song basis. Spotify’s sophisticated algorithm is based on work done by Matt Ogle at Echo Nest and Drinkify. While this algorithm is incredibly sophisticated, there are still people out there who like to own their music, and like to be reminded of favorites in their libraries, rather than to just discover new music, as important and exciting as that process definitely is.
Desirability appears to be flexible and ever-evolving for Hulu. They seek out the content most relevant to us at one specific time rather than over the long term, acknowledging that our tastes change often. I would argue that younger people without access to cable television are the likely audience, and minds of these younger people tend to be more malleable and change rapidly, anyway. I would say this is problematic when designing for an older audience, though, but they are also less likely to use this service. I would, however, consider more than just content (I would consider production quality, plot quality, etc.), and therefore weight explicit ratings more.
Desirability is attractiveness, closeness, personality (this is secondary, communicated through the biography), and mutual interest. This is for primarily young people looking for hookups or relationships, but is stereotypically used for hookups. They are not designing well for people looking for a serious connection, nor are they designing well for older people or people who find it easier to meet people in person. The app’s constraints are that it is virtual, and formatted casually, without providing things like in-app access to OpenTable or other services that could f
oster dates. Its algorithm, although clever and thorough in determining scores for each user, could be improved in several key ways. For instance, people who swipe at certain times of day or night could be paired together better, as it could reveal what their aims are from the app.
Google News certainly appears to take a “best-of-both-worlds” approach, similar to Spotify in its new curation, with its algorithm consisting of automated, highly relevant stories, handpicked stories, and hyperlocal content. All in all, using Google’s vast knowledge of each individual user allows them to tailor this content well, and therefore appears to be functionally acceptable.