[Most-ai-contest] 2020.03.24 羅上堡 M10615012 PPT 內容回應
范正忠
jjfan於iis.sinica.edu.tw
Thu 3月 26 10:52:45 CST 2020
Dear all,
問題: 根据新闻文本,第二届科技大擂台的测验有哪两大类 ?
答案: [' 语音阅 ', ' 连续对话 '] -> Modify FTC golden answer ' 语音阅读和连续对话'
預測: [' 语音阅读 ', ' 连续对话 '] -> 回答正確
問題 : 这起事故发生地段位于哪里 ?
答案: [' 云林 158 线平 ', ' 桥附近 '] -> FTC golden answer ' 云林 158 线平和 桥附近'
預測: [' 云林 158 线平 ', ' 桥 ']
問題: 这起事故的受害者其姓氏分别为何 ?
答案: [' 张 ', ' 魏 ']
預測: [‘ 张姓 ’, ‘ 魏姓 ’] -> Correct ( 在 answer list)
問題 : 在张骞通西域的两次路程中,西域的音乐和什么传到了中土 ,
促成 中西文化的交流 ?
答案: [' 葡萄 ', ' 音乐 ', ' 美术 '] -> Modify answer ' 葡萄 和 美术'
預測: [' 葡萄 ', ' 美术 ']
問題: 请列举出三项高雄的地标建物 ?
答案: [' 高雄 85 大楼 ', ' 梦时代购物中心的高雄之眼摩天轮 ', ' 高雄巨蛋 ']
預測: [' 高雄之眼摩天轮 ', ' 高雄巨蛋 ', ' 高雄 85 大楼 '] -> Correct ( 在 answer list )
目前台湾有哪些县市有提供双层观光巴士服务 ?
答案: [' 台北 ', ' 高雄 ', ' 屏东县 ']
預測: [‘ 屏 东 县 ’, ‘ 高 雄 县 ’, ‘ 台 北 县 ’] -> ‘ 高 雄 县 ’, ‘ 台 北 县 ’ 不能加 ’ 县 ’, 文章內沒有出現 ‘ 县 ’ (參考: 科技大擂台簡答題之答題規範v2)
(5) 問「空間 」時。 若 問題 包含 作答之 行政單位, 且文本描述也包含行政區,則 請 包括 行政區 單 位 做答 。
有哪些国家是使用简体字中文的 ?
答案: [' 中华人民共和国 ', ' 新加坡 ', ' 马来西亚 ']
預測: [' 中国 ', ' 新加坡 ', ' 马来西亚 '] -> 同上述理由要以 ' 中华人民共和国'回答
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Tuesday, March 24, 2020 9:17:30 AM
Subject: [Most-ai-contest] FGC_release_1.7.9 & last 3 weeks performance
Dear all,
Enclosed please find the current performance of your last modules on FGC 1.7.9 data set and last 3 weeks FGC tests.
Train: 0.729, Dev: 0.648, Test: 0.627
Week1: 0.552, Week2: 0.37, Week3: 0.39
Please take a look and analyze the errors. Thanks.
Best,
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Thursday, March 12, 2020 6:32:34 PM
Subject: [Most-ai-contest] FGC_release_1.7.7 performance
Dear all,
Enclosed please find the performance of each answer module tested on FGC_release_1.7.7
with 羅上堡's MSPE_v9_branchy8 & 廖沛俊's bestRoBERTaFGC_L_full_em
question_train_result => 0.704
question_dev_result => 0.676
question_test_result => 0.591
@謝尊安
Please check there are a lot of FINAL answers without ATEXT_TW field.
Best,
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Thursday, March 12, 2020 8:56:59 AM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear all,
Please use this version FGC_release_1.7.7
https://drive.google.com/drive/folders/1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK
with adjustment of the passage of D097
Best,
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Wednesday, March 11, 2020 6:53:23 AM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear all,
Update FGC_release_1.7.6 & DRCD dataset ( https://drive.google.com/drive/folders/1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK ) with
1. refinement of sentence split algorithm, which solves the embedded quotation issue occurred in D033
"《美国-墨西哥-加拿大协议》United States-Mexico-Canada Agreement, (简称《美墨加协议》USMCA;西班牙语:Tratado entre México, Estados Unidos y Canadá,T-MEC;法语:Accord Canada–États-Unis–Mexique,ACEUM)是加拿大、墨西哥和美国之间的一项自由贸易协定,"
2. refinement of NER character index issue when a sentence begins with space character occurred in D087
" 记者 宋玲 报导 以往在台北高雄的观光重镇才会看得到的双层巴士 今天在屏东亮向 顶层开阔的空间 游客可以边吹风边看风景 下层则备有冷气 车上将由有导览人员提供解说服务。"
Best,
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Friday, March 6, 2020 8:12:52 AM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear all,
Update FGC_release_1.7.5 & DRCD dataset
[ https://drive.google.com/drive/folders/1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK | https://drive.google.com/drive/folders/1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK ]
The modifications:
1. Information Extraction (NER, TOKEN, POS) is moved from DIE or QIE into the corresponding sentences with IE keyword. Since we found that Stanford Core NLP will fail while the length of given paragraph is greater than 653 characters
2. DRCD (including DRCD, ASR, Kaggle, Lee) are all provided with IE keyword.
Best,
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Monday, March 2, 2020 8:20:55 AM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear all,
Update FGC_release_1.7.4 with corrections of wrong answer-type and answer-mode
[ https://drive.google.com/drive/folders/1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK | https://drive.google.com/drive/folders/1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK ]
In order to improve performance we let more answer-modules participate in answering questions.
1. For those questions with answer-type 'date-duration', we add 'date-duration' to their answer-mode
2. For those questions with answer-type 'num-measure', we add 'arithmetic-operations' to their answer-mode
So the distribution of this new released data-set as follows (Please note that the count of answer mode is more than that of answer type, since for each question there will be more than one answer mode assigned)
Train
Answer Type YesNo Num-Measure Kinship Person Date-Duration Location Organization Object Event Misc Total
87 71 59 89 158 119 84 181 14 13 875
9.94% 8.11% 6.74% 10.17% 18.06% 13.60% 9.60% 20.69% 1.60% 1.49% 100.00%
Answer Mode YesNo (是否題) Multi-Spans-Extraction (列舉題型) Kinship Single-Span-Extraction (單一答案) Date-Duration Arithmetic-Operations Counting Comparing-Members Common-Sense
87 87 56 651 158 71 11 9 0 1130
7.70% 7.70% 4.96% 57.61% 13.98% 6.28% 0.97% 0.80% 0.00% 100.00%
Dev
Answer Type YesNo Num-Measure Kinship Person Date-Duration Location Organization Object Event Misc Total
28 27 30 27 26 30 19 48 2 5 242
11.57% 11.16% 12.40% 11.16% 10.74% 12.40% 7.85% 19.83% 0.83% 2.07% 100.00%
Answer Mode YesNo (是否題) Multi-Spans-Extraction (列舉題型) Kinship Single-Span-Extraction (單一答案) Date-Duration Arithmetic-Operations Counting Comparing-Members Common-Sense
28 31 26 179 26 27 2 0 0 319
8.78% 9.72% 8.15% 56.11% 8.15% 8.46% 0.63% 0.00% 0.00% 100.00%
Test
Answer Type YesNo Num-Measure Kinship Person Date-Duration Location Organization Object Event Misc Total
25 19 12 14 31 27 20 34 4 4 190
13.16% 10.00% 6.32% 7.37% 16.32% 14.21% 10.53% 17.89% 2.11% 2.11% 100.00%
Answer Mode YesNo (是否題) Multi-Spans-Extraction (列舉題型) Kinship Single-Span-Extraction (單一答案) Date-Duration Arithmetic-Operations Counting Comparing-Members Common-Sense
25 26 6 134 31 20 3 0 0 245
10.20% 10.61% 2.45% 54.69% 12.65% 8.16% 1.22% 0.00% 0.00% 100.00%
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Wednesday, February 26, 2020 2:00:09 PM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear All,
Update DROP dataset for date-duration & arithmetic-operation mode (DROP.7z)
[ https://drive.google.com/drive/folders/1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK | https://drive.google.com/drive/folders/1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK ]
Best,
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Friday, February 21, 2020 12:45:00 PM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear All,
Enclosed please find FGC_release_1.7.3 version with additional information
1. For each passage & question, (NER, CORF, RELATION, TOKEN, POS) are extracted for your training requirements
2. FGC_release_ss_test.json has 132 questions all have 'Single-Span-Extraction' answer-mode for your benchmark dataset
Best,
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Friday, February 21, 2020 9:46:23 AM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear All,
Updated news (PBS, MOTI, CDC, CNA) dataset
[ https://drive.google.com/drive/folders/1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK | https://drive.google.com/drive/folders/1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK ]
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Friday, February 14, 2020 2:36:03 PM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear All,
Updated FGC_release_1.7.2 with correction of answer_mode & answer_type
[ https://drive.google.com/open?id=1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK | https://drive.google.com/open?id=1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK ]
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Monday, February 10, 2020 12:36:32 PM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear All,
Updated FGC_release_1.7.1 with refinement
1. add SHINT:[] in QTYPE:"申論"
2. json.dump(output_dict, out_fh, indent=4, ensure_ascii=False) for easy viewer
Best,
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Wednesday, February 5, 2020 11:31:21 AM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear all,
Please refer to the following link for FGC_release_1.7 & DuReader dataset
[ https://drive.google.com/open?id=1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK | https://drive.google.com/open?id=1y0UCb-n1YKlKUsQk2GJz6iKJqQFJ25rK ]
jjfan
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Wednesday, February 5, 2020 8:23:10 AM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
20 domains
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Wednesday, February 5, 2020 7:51:36 AM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
更正:
1. DM 19 domains assigned person: 廖沛俊, 謝尊安, 蔡仕竑, 梁朝鈞
1.1 2/7 前提供每個 domian 的'訂單'細項 (依據大會提供的 CSV, 對話腳本, 參考網路相關訂單系統, 個人的經驗), 格式可以參考 ppt 內的 GIST format
1.2 2/14 根據'訂單'細項, 產出 system bot question texts, user bot answer statements (& corresponding regular expression), 細節的部分可以請教 Dr. Lin
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Tuesday, February 4, 2020 5:11:28 PM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
2.2 謝尊安 & Smolka co-work
更正
2.2 郭家銍 & Smolka co-work
From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "Balancy" <balancy at iis.sinica.edu.tw>
Cc: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Tuesday, February 4, 2020 5:07:54 PM
Subject: Re: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear all,
Enclosed please find
1. Dr. Lin's DM presentation slides and
2. The last Dialog dataset (with CVS, output order items, and dialog examples)
3. 相關時程如下:
謝謝大家的參與, 今日會議結論如下:
1. DM 19 domains assigned person: 廖沛俊, 謝尊安, 蔡仕竑, 梁朝鈞
1.1 2/7 前提供每個 domian 的'訂單'細項 (依據大會提供的 CSV, 對話腳本, 參考網路相關訂單系統, 個人的經驗), 格式可以參考 ppt 內的 GIST format
1.2 2/4 根據'訂單'細項, 產出 system bot question texts, user bot answer statements (& corresponding regular expression), 細節的部分可以請教 Dr. Lin
2. 申論題:
2.1 Smolka 會先 implement 'Query-focused' model on Chinese language
2.2 謝尊安 & Smolka co-work
3. 1.7 ver. Train/Dev/Test dataset 調整相同的 passage 不會同時出現在 Train/Dev/Test dataset, 完成後再 release
4. 各個 module 如果完成改善功能, 請提供給我進行系統 e2e 的 performance test
5. NN Library 暫定一 Pytorch 為主, 版本請相關的 owners s 協商一 common version
jjfan
From: "Balancy" <balancy at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Tuesday, February 4, 2020 8:04:40 AM
Subject: [Most-ai-contest] 科技大擂台討論會(Today, 12:30-15:00 pm)
Dear all,
今天有科技大擂台討論會,時間及地點如下,請大家預留時間參加。
時間:2/4 (二), 12:30-15:00PM
地點:新館101會議室
注意事項
*請今天報告人員於開會前,將您的報告檔案存到會議室電腦的桌面上。
感謝大家配合:)
----
Best,
吳佩瑾 Peggy, Pei-Jin Wu
蘇克毅老師實驗室助理
Administrative Assistant
Institute of Information Science, Academia Sinica
Tel: 02-2788-3799 ext . 1453
Email: balancy at iis.sinica.edu.tw
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.iis.sinica.edu.tw/pipermail/most-ai-contest/attachments/20200326/0b423ef5/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 12291.040000040084
Type: image/png
Size: 14935 bytes
Desc: not available
URL: <http://www.iis.sinica.edu.tw/pipermail/most-ai-contest/attachments/20200326/0b423ef5/attachment-0001.png>
More information about the Most-ai-contest
mailing list