Spotify is the home of audio.

Spotify 是音樂之家。

It's known for its personalized playlists, made with its recommendation algorithm.

它以個人化播放清單聞名，而清單是由其推薦演算法製成。

Think about users as this raw material, and then, on top of the data layer, we're able to build shared models.

把使用者想做原物料，然後在數據層之上，我們得以建構共享模型。

But relying so much on artificial intelligence has also drawn criticism from some industry experts worried about algorithmic bias.

但這麼大程度地仰賴人工智慧也引起了業界專家的批評，他們擔心會發生演算法偏差。

Here's how Spotify uses AI to personalize users' experiences on the platform.

以下是 Spotify 如何利用 AI 在平台上個人化用戶體驗。

This is the tech behind Spotify.

這是 Spotify 背後的科技。

In the early 2000s, many people found music recommendations through top charts and early streaming platforms like Pandora and Last.fm.

在 2000 年早期，許多人都透過排行榜或像是 Pandora 或 Last.fm 等早期串流平台發現推薦音樂。

With the Last.fm app from the App Store, you can listen to great bands...

利用 App Store 的 Last.fm 應用程式，你可以聆聽優秀樂團⋯⋯

So, when Spotify entered the scene in 2008...

所以說，當 Spotify 在 2008 年出現在市場上時⋯⋯

It's not so much that they were the first people to start using analytics to recommend music,

並不是說他們是第一群利用分析法推薦音樂的人，

but it was the way in which they combined various computational techniques in order to make their recommendations feel more lifelike.

但是他們透過結合數種電腦技術的方式讓推薦清單更生活化。

Thomas Hodgson studies algorithms and artificial intelligence, with a focus on how new technology from music streaming companies impact artists.

Thomas Hodgson 研究演算法和人工智慧，重點研究音樂串流公司新科技對於藝術家的影響。

Fans who listen to discover weekly and daily mix.

粉絲會聆聽每週、每日探索清單。

The way that they talk about them is in very human-like terms.

他們討論它們的方式是非常人性化的。

Discover weekly, you magnificent ****; you've done it again.

每週探索，你這了不起的 ****，你又做到了。

In 2014, Spotify acquired music analytics firm, the Echo Nest,

2014 年時，Spotify 收購了 Echo Nest 這間音樂分析公司，

which blended machine learning and natural language processing to build a database of songs and artists.

他們結合了機氣學習和自然語言處理以建構歌曲和藝術家的資料庫。

Spotify says this technology marked an important step in the evolution of its recommendation system.

Spotify 表示，這項刻記象徵著其推薦系統進化重要的一步。

So, how does that system work?

那麼，這個系統是如何運作的呢？

It starts with a process called "collaborative filtering".

它從一個稱為「協同過濾」的過程開始。

Collaborative filtering looks at the pattern across all of this data and tries to understand: when do tracks happen to be playlisted together very often?

協同過濾會參考這所有數據的模式，並試圖了解：歌曲何時會恰好很經常地被放在同一個清單。

You can think of it as building a map of music and podcast.

你可以把它看作建構一張音樂和播客的地圖。

That map looks something like this.

那張地圖看起來有點像這樣。

Each point represents a different track in Spotify's catalog,

每一個點都代表 Spotify 音樂庫中的不同歌曲，

and the location of each point is determined by collaborative filtering.

而每一點的位置取決於協同過濾。

Which means that these tracks go together according to the way users have playlisted them and listened to them.

這表示這些歌曲會根據使用者將它們列入同一播放清單並聆聽而被放在一起。

So, if these two songs are frequently playlisted together, they will be close to each other in this map.

所以說，如果這兩首歌很經常地被放在同一播放清單，它們在地圖上就會更靠近彼此。

Whereas if these songs are never playlisted together, they will be farther apart in the map.

而如果這些歌曲從未被放在同一播放清單，它們在地圖就會離得更遠。

But recommendations based purely on collaborative filtering aren't perfect.

但純粹基於協同過濾的推薦並不完美。

For example, during the holidays, Mariah Carey's "All I Want for Christmas Is You" might get playlisted more frequently with "Silent Night",

舉例而言，在假期時，瑪麗亞·凱莉的《All I Want for Christmas Is You》可能會更常地跟《Silent Night》放在同一清單上，

even though this sounds like a pop song, and this sounds like a Christmas carol.

雖然這首歌聽起來像是流行音樂，而這聽起來像是聖誕歌謠。

If Spotify only generated recommendations based on proximity,

如果 Spotify 只根據距離產生推薦清單，

then users who like Mariah Carey might get recommended "Silent Night" when they aren't interested in Christmas carols.

那麼，喜歡瑪麗亞·凱莉的使用者可能會在對聖誕歌謠沒興趣的情況下，被推薦《Silent Night》一曲。

To prevent this, Spotify adds another layer of analysis called "content-based filtering".

為了防止這種情況，Spotify 增加了一層稱為「內容基礎過濾」分析。

This algorithm gathers metadata, like the release date and label, and executes a raw audio analysis.

這種演算法會搜集元資料，例如發行日期和唱片公司，並執行原始音檔分析。

It uses metrics like danceability and loudness to describe the sonic characteristics of the track.

它會利用適合跳舞程度和音量等指標描述一首曲子的聲音特徵。

These are the results for "Uptown Funk", which sounds like this...

這些是《Uptown Funk》的結果，聽起來是這樣的⋯⋯

and has a danceability score of 0.856 on a scale of 0 to 1.

然後在 0 到 1 分之間，獲得 0.856 分的適合跳舞程度。

The algorithm also dissects each track's temporal structure.

該演算法也會分析每一首歌的節奏架構。

Here's a visual representation of that for "Anti-Hero" by Taylor Swift.

以下是泰勒絲《Anti-Hero》這首歌的視覺呈現。

These are the beats, the bars, and the sections.

這些是節拍、小節和分區。

Content-based filtering also takes into account a track's cultural context,

內容基礎過濾也考慮到曲目的文化背景，

which means studying the lyrics and analyzing the adjectives used to describe the track in articles and blogs.

這意味著研究歌詞並分析文章和部落格中用於描述該曲目的形容詞。

These filtering techniques are not unique to Spotify,

這些過濾技術並非 Spotify 獨有的，

but industry experts say what sets the platform apart is the amount of user data it has and the products it creates from it.

但業界專家說，讓這個平台與眾不同的是它擁有的使用者數據以及它利用數據創造的產品。

Spotify says its content-based filtering technology has evolved over the years, and now includes more advanced proprietary-facing features.

Spotify 表示，其內容基礎過濾技術已經在這些年間演化，現在包含更先進的專有面向功能。

But Hodgson says the danger with algorithms is that they could reinforce existing biases.

但 Hodgson 說，演算法的危險之處在於它們可能會加強現有的偏差。

This could mean that a particular catalog of music has more male artists than female artists.

這可能意味著某個特定音樂分類中，男性藝術家會多於女性藝術家。

One of the dangers with machine learning is that,

機器學習的危險之一是，

as listeners start to engage with that catalog, those biases become magnified, and that this creates what's called a, kind of, "feedback loop".

雖著聽眾開始使用該分類，這些偏差會被加大，而那會產生所謂的「回饋循環」。

Spotify says its research teams evaluate and mitigate against potential algorithmic inequities and harms, and strive for transparency about its impact.

Spotify 表示，其研究團隊會評估和減輕潛在的演算法不公和傷害，並努力實現其影響的透明化。

Another criticism is that the algorithm isn't optimized for new artists because there's no user data.

另一個批評是，由於缺乏用戶數據，該演算法並不會針對新藝術家進行優化。

This is known as the "cold-start problem".

這就是所謂的「冷啓動問題」。

Sultan says this is where human editors play a significant role in delivering recommendations.

Sultan 說，這就是人類編輯在提供建議方面發揮重要作用的地方。

They're possibly some of the best people in the world that's trying to understand new releases and culture and what's relevant.

他們可能是全世界最適合試圖了解新發行音樂和文化以及相關事物的人。

But Hodgson says the bigger concern is that certain metrics used in the platform's audio analysis might be culturally biased

但 Hodgson 說，更大的擔憂是，該平台音樂分析使用的某些指標可能有文化偏差。

In other parts of the world, they have musical systems and musical cultures that are entirely different.

在世界其它地區，他們擁有完全不同的音樂系統和文化。

Like this North Indian classical track, for example.

舉例而言，就像這首北印度的古典歌曲。

Spotify's algorithm labels its key signature as E minor, which Hodgson says is inappropriate for this musical tradition.

Spotify 的演算法將其調號標記為 E 小調，而 Hodgson 認為這對此種音樂傳統是不合適的。

However, it's still the case that

然而，現況仍然是，

the music that is emerging from South Asia is being understood algorithmically under, you know, the Western equal temperament scale.

發跡於南亞音樂的演算是在西方十二平均律音階之下被理解。

Spotify says the audio analysis is one small part of the overall system, which takes into account many factors before making a recommendation.

Spotify 表示，音樂分析是整個系統的一小部分，該系統在做出推薦前會考量許多因素。

Some industry experts also point to issues with how the system understands metadata for classical music.

一些業界專家還指出了該系統如何理解古典音樂元數據的問題。

For example, the metadata for a Tchaikovsky track can include not just the name of the work and the artist, but also the movement, opus number, and conductor.

例如，一首柴可夫斯基曲子元數據不僅可以包括作品的名稱和藝術家，還可以包括樂章序號和指揮家。

Spotify's algorithm isn't optimized for that.

Spotify 的演算法並未為此進行優化。

Apple Music, which has emerged in recent years as a competitor to Spotify,

近年來，作為 Spotify 競爭對手而出現的蘋果音樂

released a new app in March that the company says is designed to solve this problem.

在 3 月份發佈了一款新的應用程式，該公司表示其目的在於解決這個問題。

Spotify says it doesn't comment on a competitor's marketing campaigns.

Spotify 表示，它不會對競爭者的行銷活動發表評論。

In February, the streaming service joined the recent buzz around generative AI.

2 月時，該串流服務加入了最近火熱的生成式人工智慧。

I'm X, and from this moment on, I'm gonna be your own personal AI DJ on Spotify.

我是 X，從此刻開始，我會擔任你在 Spotify 上的專屬人工智慧 DJ。

The DJ gives the algorithm a human voice and offers listeners additional context around a recommendation.

這個 DJ 讓演算法變得人性化，並提供聽眾推薦曲目之外的額外資訊。

Up next, I know you've been on a summer song kick lately.

接下來的歌曲，我知道你最近很喜歡夏日曲目。

Sultan says the company is also exploring reinforcement learning,

Sultan 表示，該公司也在探討強化學習，

a technique that would allow the recommendation system to learn automatically based on feedback.

這個技術讓推薦系統能夠根據回饋自動學習。

It will help with the diversity of their recommendation, it will help with the longer-term retention.

它會幫助推薦的多樣化，並幫助較長期的留客率。

And we're trying to push the state-of-the-art in each of those,

我們正努力在每個領域推動最新技術，

introducing new technologies, new capabilities, and bringing new experiences.

介紹新科技、新能力，並帶來嶄新體驗。