Is it a catchy title?

是根據吸睛的標題？

Maybe an interesting synopsis?

是看到感興趣的影片簡介？

Perhaps it's the unwillingness to watch another episode of The Office...

還是逼不得已再繼續觀賞一集《我們的辦公室》(美國情境喜劇)？

“Save Bandit!”

「顧好我家 Bandit！」

Or just maybe, a particular piece of cover art speaks to your spirit?

也可能是某部影片的主視覺設計讓你一見鍾情？

But go on your friend's Netflix account — if they aren't still using your subscription,

至於你朋友的 Netflix 帳號──前提是他們沒有和你共用帳號

and you may find the cover art there doesn't interest you.

──你或許會發現那些影篇首圖並不吸引你。

In fact, what you see might be completely different.

你甚至可能看到完全不一樣的首頁內容。

It's no accident — Netflix's thumbnails are all tailor made.

這並不是偶然發生的，因為 Netflix 上的影片縮圖都是客製的。

For you.

為你一個人打造。

At its heart, Netflix is all about creating personalized experiences—or, rather, calculating them.

Netflix 的首要任務就是要創造出，更精確地說，是用電腦算出專屬個人的觀賞體驗。

They'll draw you in with curated trailers of upcoming releases, point out new episodes from previously watched videos,

他們用即將上映的電影預告吸引點擊，或者為你指出新的集數，

and even gauge your interest in content via match scores.

甚至計算媒合分數，猜測你會對哪種內容有興趣。

And while streaming services are notoriously tight-lipped about sharing their viewership data,

雖然影片串流業者一向對用戶分析資料守口如瓶，

over the years, Netflix has shared glimpses into how their technology works.

這幾年來，Netflix 讓我們一窺他們的運作模式。

According to internal studies, a typical viewer spends 1.8 seconds considering each title,

根據內部研究，通常一位觀眾會花上 1.8 秒的時間瀏覽標題，決定要不要看，

and Netflix believes it only has 90 seconds to get your attention before you move onto another activity.

在你分心做其他事情之前，Netflix 相信自己必須有 90 秒可以抓住你的注意力，

And among all the things that could catch your attention and make you watch a show, or several,

而其中最能吸引你觀賞一部影片，或者好多部影片。

Netflix found that the biggest influence were the thumbnails.

關鍵在於網站上的影片縮圖。

Humans are intensely visual creatures.

人類是非常仰賴視覺的生物，

Our eyes move three to four times a second to process new information.

為了吸收新資訊，我們的雙眼在一秒鐘內會移動三到四次。

And because Netflix's goal is to get your attention and hold it,

而既然 Netflix 的主要目的是要攫取並維持你的注意力，

the company puts a lot of work into choosing every thumbnail you'll see.

頁面上的影片縮圖可都是他們精挑細選的結果。

But before they can decide what image will show up on your account,

在他們決定你將會看到哪些圖片之前，

they have to sift through a ton of data.

他們必須過濾一堆數據資料。

An hour-long Stranger Things episode has almost 86,000 video frames.

舉例來說，片長一小時的《怪奇物語》每集都有大概 86000 幀影格，

To figure out which ones will make the best thumbnails,

為了找出哪些影格最適合做成影片縮圖，

Netflix uses a pretty scientific selection process called Aesthetic Visual Analysis, or AVA.

Netflix 採用了一種挺科學的方法：美學質量評估，簡稱 AVA。

AVA is a set of tools and algorithms that search Netflix videos for the best images, and pull them out to create thumbnails.

AVA 結合各種工具與演算法，搜尋所有站內影片、找到最適切的圖片，再揀選出來做成影片縮圖。

The process can be broken down into two basic steps.

這個過程大概可以分成兩個步驟：

First is frame annotation.

首先是影格註解。

A program analyzes every static video frame of a video,

這個程式能夠分析一部影片中的所有靜態影格，

and image recognition algorithms use information gleaned from each shot to create metadata.

再來圖像辨識演算法會利用蒐集來的資訊，構成所謂的後設資料。

The metadata is like an electronic fingerprint identifying characteristics unique to each video frame.

後設資料就像是電子化的指紋，能夠辨識出專屬於每幀影格的特色。

All of this info helps build a database of information that makes it easier to pick out the best images for thumbnails later.

這些資訊都會被投入數據庫的建構，讓稍後挑選適合圖片的工作更加容易。

And to sort all this data, the company groups it into roughly three categories that are key to identifying good images:

至於這些資料的分類約略可以分成三組，且這三項標準都是區別好圖片的關鍵。

Visual — focusing on brightness, color, contrast, and motion blur.

視覺效果──涵蓋亮度、顏色、對比、動態模糊。

Contextual, which documents face and object detection, motion, and shot angles.

情境脈絡── 記載著臉部與物體偵測、動作和拍攝角度。

And compositional — which focuses on visual principles in cinematography, photography, and design.

畫面構成──針對不同的視覺原則，涵蓋電影美學、攝影、設計領域。

The second step is a process called image ranking.

第二個步驟是圖像排名。

An algorithm uses the metadata to pick out specific shots that Netflix has determined are the most attractive and clickable.

一套運算法會運用後設資料來揀選片段，那些片段經由 Netflix 判定，最能吸引點擊。

Ones that aren't blurry, have varied imagery, feature major characters,

他們必須清晰不模糊、圖像元素多樣、包含主要角色，

and don't contain sensitive or unauthorized branded content.

且沒有敏感或者未經授權的品牌訊息。

Then, finally, a creative team steps in to use the best images to design the thumbnail artwork.

最後創意團隊才會介入，把這些最適合的圖片設計成影片縮圖。

But the process doesn't stop there — Netflix still has to figure out which ones work best for each user.

但這還不是最後一步，Netflix 還得決定不同圖片與使用者的媒合。

A/B testing is executed — again, again, and again.

因此會有好多輪的 A/B 測試。

You'll regularly see changes in your thumbnails — based on your engagement with previous titles.

你看到的縮圖會定期更換，根據你對前一類標題有多感興趣。

On the most basic level, let's say you're a fan of comedy,

舉個最簡單的例子來說，假設你熱愛喜劇，

and you've watched a bunch of stand-up specials.

且你看過許多不同的獨角喜劇，

When you search for Good Will Hunting, you may get a thumbnail with Robin Williams,

當你搜尋《心靈捕手 》這部電影，你可能會看到影片縮圖中出現 Robin Williams。

a famous comedian, and one of the movie's main characters.

他是位家喻戶曉的喜劇演員，也是這部電影的主要角色之一。

But people more into romance titles could be shown the cover art with the two leads kissing.

但如果你比較常被愛情片標題吸引，你看到的圖片可能會是兩位主角在親吻。

There are also regional differences.

地域之間也存在著差異。

From glancing at a show's thumbnails across different countries, you could infer Germany is more into abstract images,

當我們觀察同一部影片在不同國家顯示的縮圖，概括而言德國偏好的是抽象圖片，

and US viewers may prefer clearly defined characters and story plots.

美國觀眾則比較喜歡確切的角色以及故事情節。

Though there are a considerable number of thumbnails to choose from,

雖然有很多影片縮圖可供選擇，

and lots of science that goes into each decision, it doesn't mean Netflix gets it right all the time.

同時決策過程中也有不少科學介入，Netflix 也有犯錯的時候。

One Twitter user found the cover art for, "Like Father" a movie starring Kristen Bell,

有一位推特用戶發現電影 Like Father，主要演員有 Kristen Bell、

Kelsey Grammer, and Seth Rogen, had cover art that didn't exactly match who the lead actors in the film were.

Kelsey Grammer、Seth Rogen，可是影片首圖卻並不相符。

It was also kinda odd to see on my own account, that the Catwoman thumb has an image of a barely recognizable supporting actress Sharon Stone,

我也曾在自己的頁面上看到《貓女》的縮圖上出現一位不太重要的小角色，由 Sharon Stone 飾演；

and Blade II had someone other than its star, Wesley Snipes, in the thumbnail.

《刀鋒戰士2》的縮圖上不是主角 Wesley Snipes 而是別人。

Netflix is obsessed with A/B testing new features like video promos, intro skipping, and auto-playing trailers, just as obsessively as they are with testing thumbnails.

每當有新功能推出，Netflix 都堅持進行 A/B 測試，例如影片廣告、略過導覽、自動播放預告等，這些都會經過和影片縮圖類似的測試過程。

Netflix wants everyone to watch more, so it's unlikely they'll stop doing these tests.

Netflix 希望用戶看更多影片，所以這些測試大概會持續下去。

And if you don't like being a guinea pig, there is the option to opt out.

倘若你不喜歡當 Netflix 的白老鼠，你可以更動這項設定。

But with binge-watching quickly becoming a national pastime, it's likely that,

但有鑑於追劇已經漸漸演變成全國性的消遣，最可能的情況是，

when Netflix asks you,

當下一次 Netflix 再問你：

"Are you still watching?"

「你還想接著看下去嗎？」

You will be.

答案應該會是肯定的。

Hey! Thanks for watching The Goods.

感謝觀賞 The Goods 系列影片。