Credit to 高學歷廢青@lihkg

兩日前民族黨陳浩天到FCC嘅一場演講帶嚟滿城風雨，外國嘅傳媒亦都將中國殖民式打壓香港言論自由嘅醜態楊威國際，香港民族意識終於被外國關注，可喜可賀。

但係講起傳媒，就一定唔可以唔提呢一兩年最多人講嘅fake news。究竟報導陳浩天嘅外國媒體，有幾多係real news，有幾多係fake news呢？唔使用劍嘅，用machine learning AI就得。

我搵咗個叫Fakebox嘅AI，用Natural Language Processing加machine learning，餵個AI食上至Reuters下至The Onion嘅新聞，直到佢可以自己辨認出乜嘢係公正持平嘅real news，乜嘢係譁衆取寵嘅fake news，仲 >95%準添。將各個外媒嘅報道餵畀個AI，得出一個有趣嘅結果（分數越高越持平，用content排序）：

CNN: Title 0.4272, unsure; [#2a8f1d]Content 0.8976 impartial[/#2a8f1d]
Reuters: Title 0.9069, impartial; [#2a8f1d]Content 0.8883, impartial[/#2a8f1d]
Bloomberg: Title 0.8939, impartial; [#2a8f1d]Content 0.8619, impartial[/#2a8f1d]
New York Times: Title 0.4366, unsure; [#2a8f1d]Content 0.7917, impartial[/#2a8f1d]
CNBC: Title 0.8183, impartial; [#2a8f1d]Content 0.7333, impartial[/#2a8f1d]
Guardian: Title 0.8023, impartial; [#2a8f1d]Content 0.6993, Impartial[/#2a8f1d]
Fox News: Title 0.8005, impartial; [#d9ff42]Content 0.6000, unsure[/#d9ff42]
—————————————
Time: Title 0.3861 bias; [#ff0011]Content 0.3977, bias[/#ff0011]

有人可能讀完資料會諗，CNN唔係出名fake news咩？點解會高過Reuters同Bloomberg？原來佢今次做標題黨，個Title得0.4272，但係內容持平。

但係問題嚟啦：究竟點解Time嘅分數會低過其他傳媒咁多嘅呢？我哋睇吓Time喺七月被人篤爆造假，造fake news，用Trump+喊緊嘅女仔做封面嗰期嘅main article:
2 July 2018, "A Reckoning After Trump’s Border Separation Policy: What Kind of Country Are We?"
Title 0.10, bias; Content 0.319, bias

再比較下環球時報嘅社評：
12 Aug "Protecting peace, stability is top of human rights agenda for Xinjiang"
Title 0.57 unsure; [#ff0011]Content 0.20 bias
[/#ff0011]
14 Aug "US defense splurge won't intimidate China"
Title 0.78 impartial;[#ff0011] Content 0.26 bias[/#ff0011]

或者SCMP嘅社評：
13 Aug Mike Rowse "Hong Kong independence talk controversy is evidence that silly season is in full swing"
Title 0.295 bias; [#ff0011]Content 0.267 bias[/#ff0011]

再比較下SCMP平時正正常常嘅real news報導：
16 Aug "Andy Chan’s pro-independence speech at FCC shows Hong Kong’s ‘inadequacies’ in upholding China’s national security, says top Beijing official"
Title 0.59 unsure; [#0fa30a]Content 0.829 impartial
[/#0fa30a]
哦，原來係咁！原來越係個人評論，越係情感投放，越係先入爲主，個AI就覺得越係fake news。Time嘅一篇fake news，對比起其他咁多份報章，真係高下立見。可悲嘅係有一啲香港人竟然畀佢嘅加鹽加醋不實報導領導咗。

以下attach嘅係Fakebox分析完之後嘅數據，畀大家參考。

各外媒報道source：
Reuters: https://www.reuters.com/article/us-hongkong-politics-china/pro-china-groups-protest-outside-hk-press-club-over-separatist-speech-idUSKBN1KZ06C
Bloomberg: https://www.bloomberg.com/news/articles/2018-08-13/pro-china-protesters-target-hong-kong-s-top-journalists-club
CNN: https://edition.cnn.com/2018/08/14/asia/hong-kong-independence-intl/index.html
New York Times: https://www.nytimes.com/2018/08/14/world/asia/hong-kong-andy-chan-fcc.html
CNBC: https://www.cnbc.com/2018/08/14/pro-independence-hong-kong-politician-andy-chan-discusses-china-atfcc.html
The Guardian: https://www.theguardian.com/world/2018/aug/14/hong-kong-activist-andy-chan-decries-china-colonial-invasion-defiant-speech
Fox News: http://www.foxnews.com/world/2018/08/14/activist-compares-beijings-rule-in-hong-kong-to-colonialism.html
Time: http://time.com/5364918/andy-chan-ho-tin-hong-kong-independence-hknp/

Fakebox簡介：https://towardsdatascience.com/i-trained-fake-news-detection-ai-with-95-accuracy-and-almost-went-crazy-d10589aa57c
想自己玩下個Fakebox，去呢到：https://machinebox.io/docs/fakebox