Re: [ACAT] How Netflix Reverse Engineered Hollywood – Atlantic Mobile

Posting to Autocat

On 1/2/2014 7:18 PM, Bryan Campbell wrote:

<snip>
This is a fascinating article about how Netflix genres are created. How Netflix Reverse Engineered Hollywood – Atlantic Mobile http://www.theatlantic.com/technology/archive/2014/01/how-netflix-reverse-engineered-hollywood/282679/ 

In case that breaks, try http://theatln.tc/1cKP4gx
</snip>

That is really interesting. Thanks so much for sharing this. I thought I would share my own, short analysis.

Netflix is not in Italy, so I have never seen it, but the spreadsheet the author gives of the terms he found used by Netflix https://docs.google.com/spreadsheet/ccc?key=0AlC_pAJFqGnHdGxFNGlLdlVpcmc0OTBOeWNiamROMVE&usp=sharing is certainly intriguing. Although some of those terms are fairly clear in their meaning, “Time travel” “Martial arts” even “Zombie”, others are laden with personal opinion, e.g. “Goofy” vs “Inspiring” vs. “For Hopeless Romantics” vs “Hidden Gem” (four people could see the same movie and use any of those terms to describe it, and maybe all four for the same movie), and others are totally unclear, “Understated” “Gritty” “Irreverent”. For instance, what does “Mind-bending” mean?

Others are clearly errors, e.g. “Based on Books” vs. “Based on a Book”

These terms are quite puzzling, and I assume there are no cross-references of any sort, so that if I would search for “disrespectful” it would not reply: “See: Irreverent”.

An answer may be in one of the comments to the article by Jeff A. Taylor:
“Ah. Explains why Netflix has steadily made its search function harder and harder to use. It really does not want to empower end-users, it wants to effectively program content for you.”

He may very well be right, and this is an example of an “intelligent agent”. (Sorry that I cannot resist. I discussed Tim Berners-Lee’s dream of intelligent agents–which is my nightmare–in my “Musings on the Linked Data Diagram” http://blog.jweinheimer.net/2012/03/cataloging-matters-podcast-no-14-musings-on-the-linked-data-diagram.html)

I can’t imagine anybody actively searching with these terms found in Netflix. The author of the article generates some new ones, e.g. “Fight-the-System Political Love Triangle Mysteries”. Who will search for that?

So, it seems that these terms cannot be primarily for searching,  and they may not even be there primarily for grouping, as Mr. Taylor suggests for “programming content for us”. This would involve creating groups of similar films, e.g. the set of all films about “Scary Cult Movies from the 1980s”. Of course, providing the searcher with such groupings are part of the functions that library headings are supposed to do: to provide the user with “the set of all items about the Roman empire during Trajan” or “the set of all items about “Zydeco music”. Yes, there are subtleties with achieving that: 20% coverage, and above all, the terms must be assigned consistently–but the Netflix terms are so vague that achieving any level of consistency for assigning a term such as “Mind-bending” or “Nostalgic”, let alone distinguishing between similar terms such as “Fight-the-System” vs. “Underdog” would be next to hopeless for human experts, let alone a machine. So what is the purpose of these terms?

The article also mentions the surprising fact that Raymond Burr and Barbara Hale are rated very high at Netflix (strangely, Raymond Burr is the most highly rated actor!). Of course, I doubt if very many people would rate Raymond Burr as the greatest actor of all time, so they are measuring something else. The author concludes, and I think it is obvious, that all of Netflix’s work is aimed toward getting people to watch these films, or in other words, to buy Netflix’s goods. In a similar way, I think the terms we see are chosen not so much for search and retrieval, but far more to pique the person’s curiosity so that maybe, they might look at something that otherwise they would never have even considered. I mean, who would not at least look at “Steamy Mind Game” movies, even though it could mean almost anything!

I think the people in the old burlesque world would understand this attitude. They had people watching the audience and when they saw people getting bored, they would send out a pretty girl to wriggle around on the stage, or maybe just hit somebody with a rubber fish so that everybody would sit up and laugh. I think Netflix is doing something similar: adding strange terms just to get people’s attention to movies that they would otherwise ignore.

This isn’t saying anything against Netflix. I wish we had it here in Italy and I would probably subscribe.

One other point I discovered. While I was looking for related articles, I found this: “Netflix’s New TV Interface Should Ease Search for Movies, TV Shows” (ABC News, Nov. 13, 2013 http://abcnews.go.com/Technology/netflixs-tv-interface-ease-search-movies-tv-shows/story?id=20847719), which talks about a new interface Netflix decided to implement. One sentence stood out at me: “It’s about Netflix getting out of the way and letting the user connect with the movie or TV show,” Jaffe said. “We’ve been testing with 200,000 users and the result is that they watch more.”

A test with 200,000 users?! Wow! That shows they are taking matters seriously.

Is the library cataloging community doing any testing with the users?

-488

Share