L

4 Messages

 • 

100 Points

Tue, Jun 1, 2021 10:56 AM

0

Regional or search powered lists - Favourite people (ios app)

Favourite people list is all white folks. Zero representation for anyone outside the mainstream. Also not possible to search and add folks out of the top 100.

Please be more inclusive when building product features.

Based on this feature the data will be inherently skewed saying 100% people only add Hollywood folks as favorite so no need to expand. The feature is inherently skewed against inclusion, diversity or internationalization.

Responses

7.9K Messages

 • 

181.3K Points

19 d ago

Curious. What is the handle of the author of the list?

4 Messages

 • 

100 Points

It is IMDb curated list based on usage.

4 Messages

 • 

100 Points

Leading to a list like this which is naturally skewed towards Hollywood.

Employee

 • 

1.7K Messages

 • 

33.3K Points

Hi @LastPlot

Thank you for your feedback. Please note the list you've shared is not manually contrived but rather, automatically generated and sorted based on Popularity. STARmeter tracks the popularity of an individual's Name page on IMDb. 

 

You can find more information on STARmeter and our ranking systems on the STARmeter, MOVIEmeter, and COMPANYmeter FAQ.

4 Messages

 • 

100 Points

@Taylor I understand that the above list is automatically generated and it exemplifies the user base, which is originally being fed skewed content, likely targeting money-making markets (ie: Hollywood). However, AI or data is not an excuse for lack of representation, and automatically generated (a human-written piece of code) is a sorry excuse.

This is similar to "the racial bias in Twitter's image-cropping AI" which was pre-dominantly fed skewed data and no measures were taken to correct it. The STARmeter list that you shared has 1-2 POC and one non-western person (who however happens to be affiliated with Hollywood). 

I have personally rated 1.2k+ movies and 10k+ search or interaction points on the platform, with that much data, I would expect IMDb (and Amazon) to understand or recognize my regional/language/genre preferences and generate a weighted list that reflects the same.

The title rating system has a disclaimer about "weighted mechanism" at play, would expect some of that for "STARmeter" as well. Because truth be told, the STARmeter is disappointing to anyone who consumes any content outside Hollywood.

(edited)

7.9K Messages

 • 

181.3K Points

The composition of the top five, top ten, top twenty, top fifty, top hundred or so changes from day to day, meaning that sometimes celebrities not of Caucasian descent do indeed appear in the list or even in the number one slot. The rankings reflect the whims of IMDb's visitor base, which will invariably be influenced by the news, talk shows, cinema journals, commercial advertisements, movie teasers, movie trailers, movie posters, TV spots, "mass media", so on and so forth. There isn't actually a "fix" for the phenomenon itself, but certainly a way to filter for specific ethnicities or colors, the latter (like face types) being easier for a machine learning algorithm to figure out, could be introduced at some point. Maybe there could be and ought to be more than one MOVIEmeter metric, more than one STARmeter metric and more than one COMPANYmeter metric, such as a unique metric for each "region" of the world, each a mere component of the composite that the general Xmeter represents.

I've observed concerns raised about this overarching subject matter before, and I saw fit to provide some bit of explanation: https://community-imdb.sprinklr.com/conversations/data-issues-policy-discussions/whats-this-about-imdb/5f4a79d38815453dba8c5130?commentId=5f4a7b198815453dbac9da8c.