http://hlt.suda.edu.cn/index.php?action=history&feed=atom&title=CCL2021-discussion 2026-09-21T03:55:48Z 本wiki的该页面的版本历史 MediaWiki 1.35.2 http://hlt.suda.edu.cn/index.php?title=CCL2021-discussion&diff=3642&oldid=prev 2021-03-09T07:41:43Z

<p></p> <table class="diff diff-contentalign-left diff-editfont-monospace" data-mw="interface"> <col class="diff-marker" /> <col class="diff-content" /> <col class="diff-marker" /> <col class="diff-content" /> <tr class="diff-title" lang="zh-Hans-CN"> <td colspan="2" style="background-color: #fff; color: #202122; text-align: center;">←上一版本</td> <td colspan="2" style="background-color: #fff; color: #202122; text-align: center;">2021年3月9日 (二) 07:41的版本</td> </tr><tr><td colspan="2" class="diff-lineno" id="mw-diff-left-l1" >第1行：</td> <td colspan="2" class="diff-lineno">第1行：</td></tr> <tr><td class='diff-marker'>−</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div>== <del class="diffchange diffchange-inline">2020</del>.2.22 ==</div></td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div>== <ins class="diffchange diffchange-inline">2021</ins>.2.22 ==</div></td></tr> <tr><td class='diff-marker'> </td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"></td><td class='diff-marker'> </td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"></td></tr> <tr><td class='diff-marker'> </td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>* Bert-fine-tune</div></td><td class='diff-marker'> </td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>* Bert-fine-tune</div></td></tr> <tr><td colspan="2" class="diff-lineno" id="mw-diff-left-l22" >第22行：</td> <td colspan="2" class="diff-lineno">第22行：</td></tr> <tr><td class='diff-marker'> </td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>* 给一些基础的结果（帅克）</div></td><td class='diff-marker'> </td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>* 给一些基础的结果（帅克）</div></td></tr> <tr><td class='diff-marker'> </td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>   concat：baseline, +bert, +bert-ft-3, +bert-ft-10 (supar)</div></td><td class='diff-marker'> </td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>   concat：baseline, +bert, +bert-ft-3, +bert-ft-10 (supar)</div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"></ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">== 2021.1.25 ==</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">* CCL-2021句法领域移植评测，发布CODT 2.0</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">* 增加两个领域的数据：法律、金融（无标注数据分别给15万句，分好词）；规范通用领域的训练集增大为3万句</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">* 所有领域的训练集，都由BC-train-partial上训练的SuPar-BERT-CRF-1o补全。【明月指导帅克做】</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">* 苏大跑一些基准实验，主要就用SuPar。帅克来跑。包括self-training tri-training</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">* 数据处理过程:[http://hlt.suda.edu.cn/index.php/CCL2021Data 网址]</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">* 2021-2-22讨论：[http://hlt.suda.edu.cn/index.php/Discussion2021-2-22 网址]</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"></ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"> 如何验证补全的正确性？如何验证每一件事的正确性</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"> 最终补全，要用bert-finetune增强的模型做补全</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"> 不可以用codt1.0的数据</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"> 要做哪些基础实验，来帮助我们了解数据，了解codt1.0和2.0的差别；不同领域的差别</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"> 基础实验做完之后，就以参赛者的标准，来打比赛，主要的方法就是self-training（数据增强？知识蒸馏？异构数据？）</ins></div></td></tr> <tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"> open（只能用大家都可以公开获取的工具、数据资源等）和closed track（只能用我们提供的embedding/bert/unlabeled data）；不确定可以咨询我们，有新的规定会通知大家，并写到评测主页上。</ins></div></td></tr> </table>

Myzhou http://hlt.suda.edu.cn/index.php?title=CCL2021-discussion&diff=3605&oldid=prev 2021-02-22T09:34:24Z

<p>创建页面，内容为“== 2020.2.22 == * Bert-fine-tune 在base基础上，在所有的unlabeled/labeled data上只用LM-loss去finetune3轮/10轮（给两个fine-tune的bert model，py...”</p> <p><b>新页面</b></p><div>== 2020.2.22 ==<br /> <br /> * Bert-fine-tune<br /> 在base基础上，在所有的unlabeled/labeled data上只用LM-loss去finetune3轮/10轮（给两个fine-tune的bert model，pytorch）。将来我们给基准结果时，也可以给两个有bert-finetune的结果。<br /> <br /> * 预训练word embedding，我们也要提供<br /> 借鉴去年的做法，也重新训练下，帅克<br /> <br /> * CODT-2.0的数据处理<br /> 全角转半角<br /> 标点符号的处理<br /> proj/non-proj不用管<br /> <br /> * train数据的补全<br /> 在所有train（6个领域，包括BC）合并的train-all数据集上，训练一个supar-crf1o-with-finetune-bert的模型，开发集用dev-all（6个领域）。然后对train-all进行补全。补全的head/label的概率相乘，也放到某一列；如果是人工标注的head，概率写成2.0<br /> dev和test不补全，没有人工标注的，父亲节点为-1<br /> <br /> * 分成开放和封闭两个track？待定<br /> 封闭：只能用我们提供的有标注和无标注数据、我们提供的word embedding<br /> 开放：用任何数据、资源、工具都可以，只要大家能够通过某种方式公开获取（包括购买）到就可以。<br /> <br /> * 给一些基础的结果（帅克）<br /> concat：baseline, +bert, +bert-ft-3, +bert-ft-10 (supar)</div>

Zhli