【轉載】自動原創高質量文章電腦算法

Associated Press（以下簡稱AP），是一家美國非盈利的新聞機構，上週五他們對外宣佈在過去的2014年的最後三個月，他們一共發佈了 3000 篇文章，而在過去他們可能僅僅只能夠發佈 300 篇。 10 倍的發文數量增長的背後並沒有任何人力成本的增長，所有的源頭來自於一個算法。
4 s( ?6 }9 Q, E- c3 {2 a. t2 `. O
9 c; W; U- U4 T* v  H$ ?在去年夏天，Associated Press 與 一家提供實時內容自動化生成的公司 Automated Insights 合作，通過算法將過去積累的數據轉化成文章、可視化圖片等等。就過往發的3000篇的文章，由算法生成的文章幾乎沒有錯誤，你甚至無法將這些文章與人類所自己撰寫的文章分開。所有的文章都是以美聯社的風格作為標桿，而內容豐富多樣的遠遠超出了一個正常作者的能力。
1 {2 H: K  [3 g7 M
- O/ Q  S; J+ p# _不僅僅是高產和高質量，更重要的是他們在新聞時效性的優勢。去年，在洛杉磯地震僅僅過了3分鐘之後，AP 就已經將地震的新聞報道給公眾，而在近日，蘋果剛剛發佈了他們去年的財報幾分鐘之後，AP 就撰寫了關於蘋果 2015 年 Q1 季度財報的預測。這意味著機器已經開始學習，甚至於提供分析、洞察力這些加入更多人腦思考的領域。
& Q3 E$ @$ E/ ]3 j: y( f
* l$ K1 F" v/ h8 QNarrative Science 是另外一家提供類似服務的公司，他們的 CTO Kristian Hammond 認為：到了2030年，可能有90%的新聞都由機器寫作完成。目前，福布斯還有一些新聞媒體，已經使用Narrative Science 公司的軟禁寫作電子郵件等等的文件了。可以想像，未來機器寫作代替的場景會越來越多，同時它們的寫作新聞會越來越智能。
% f( |+ ?$ ?4 L8 m- b2 M
9 H* p8 {) e6 iAP's 'robot journalists' are writing their own stories now
Minutes after Apple released its record-breaking quarterly earnings this week, the Associated Press published (by way of CNBC, Yahoo, and others) "Apple tops Street 1Q forecasts." It's a story without a byline, or rather, without a human byline — a financial story written and published by an automated system well-versed in the AP Style Guide. The AP implemented the system six months ago and now publishes 3,000 such stories every quarter — and that number is poised to grow.
- g% Y6 s2 R" b+ l1 {# m5 G4 [
Quarterly earnings are a necessity for business reporting — and it can be both monotonous and stressful, demanding a combination of accuracy and speed. That's one of the reasons why last summer the AP partnered with Automated Insights to begin automating quarterly earnings reports using their Wordsmith platform.

; m. R) m; p* \' dYou wouldn't necessarily know it at first blush. Sure, maybe reading it in the context of this story it's apparent, but otherwise it feels like a pretty standard, if a tad dry, AP news item. The obvious tell doesn't come until the end of an article: "This story was generated by Automated Insights." According to AI's public relations manager James Kotecki, the Wordsmith platform generates millions of articles per week; other partners include Allstate, Comcast, and Yahoo, whose fantasy football reports are automated. Kotecki estimates the company's system can produce 2,000 articles per second if need be.

""I wouldn't expect a good journalist to not be skeptical.""

Philana Patterson, an assistant business editor at the AP tasked with implementing the system, tells us there was some skepticism from the staff at first. "I wouldn't expect a good journalist to not be skeptical," she said. Patterson tells us that when the program first began in July, every automated story had a human touch, with errors logged and sent to Automated Insights to make the necessary tweaks. Full automation began in October, when stories "went out to the wire without human intervention." Both the AP and Automated Insights tell us that no jobs have been lost due to the new service. We're also told the automated system is now logging in fewer errors than the human-produced equivalents from years past.
/ X& w  }$ |0 a! v' e/ v" A
. d2 r, {$ M  P7 ]3 t"Ten times as many reports every quarter — and no jobs cut"
" ^' ]8 F6 d& C& k" V1 Y+ \
0 Q$ e- [7 O. O! IBefore this program was implemented, the AP estimates it was doing quarterly earnings coverage for about 300 companies. Now it automates 3,000 such reports each quarter. Of those, 120 will have an added human touch, either by updating the original story or doing a separate follow-up piece. One such company is Apple; as Patterson notes, that automated Apple story freed up reporter Brandon Bailey to focus on this angled, more nuanced report contextualizing the company's earnings along with quotes from Apple executives. Others include Google, Coca-Cola, and American Airlines. 180 more are monitored to see if a follow-up is needed.

Then there are ten companies that aren't automated at all due to the nuance of their reports — companies like Citigroup and Wells Fargo. Patterson says all these lists are re-evaluated and updated every quarter.

$ I" ?2 b/ f  w9 W& k$ d& k# aSince the partnership began, elements like business descriptions and forward-looking guidance has been added to the platform's skill list. The next step is expansion — more than 1,000 Canadian companies plus a few elsewhere around the world. Patterson also told us the AP is starting to look at other uses outside of earnings reports.

, M6 m' G. a. q) @. X; z/ e: ASo no, computers are not taking journalists' jobs — not yet, at any rate. Instead, they're freeing up writers to think more critically about the bigger picture. "One of the things we really wanted reporters to be able to do was when earnings came out to not have to focus on the initial numbers," said Patterson. "That's the goal, to write smarter pieces and more interesting stories."
+ d& _& ^1 d! I4 q0 m+ I0 M0 E; Z
$ w$ X. K8 K/ f, w6 q+ \This story was generated by a Homo sapiens who really wanted to use this Shutterstock photo as the lead image:
8 t; m! |3 Z2 r8 u

原文-中文：
http://j.news.163.com/docs/99/2015020122/AHDCL44R9001L44S.html
3 i; O, G) n7 \+ q英文：
+ F& a0 J5 {9 }6 C5 ahttp://www.theverge.com/2015/1/2 ... financial-reporting
% n' |+ j" X9 ?# Y" D. Thttp://qz.com/228218/the-aps-new ... er-is-an-algorithm/
1 B% B1 P4 ]" f6 E; e


$ A. f  ^2 F+ j' \5 w9 t



$ a+ b& k6 e; e# N2 W
* a5 ?; z2 Y4 s" _- T

然後哩???我們只有看的份嗎.

這一點我絕對信，人的工作會最終被機器替代，

夢溪聽竹 發表於 2015-2-2 19:31
3 m# x+ L; w* e然後哩???我們只有看的份嗎.

說明機器原創高質量文章是有人在做了。誰先做好就能賺錢。

只要是機器做內容 那麼重複性就氾濫了

這個太強大了~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# e8 s7 K. [, E3 R2 q& K% [