图 书 馆
通过比较实体摘要促进人类干预共指消解Facilitating Human Intervention in Coreference Resolution with Comparative Entity Summaries |
|
| 课程网址: | http://videolectures.net/eswc2014_cheng_entity_summaries/ |
| 主讲教师: | Gong Cheng |
| 开课单位: | 南京大学 |
| 开课时间: | 2012-07-30 |
| 课程语种: | 英语 |
| 中文简介: |  Web数据集成的主要挑战是共引用解析，即从引用同一真实世界实体的不同数据源中识别实体描述。共同引用解决方案的解决方案越来越多地涉及人员。例如，许多主动学习，众包和现收现付方法趋于合理，因此，合法用户反馈会验证由自动方法计算出的候选核心实体。尽管减少验证任务的数量是这些方法的主要考虑因素，但很少关注执行每一个验证任务的效率。为了解决这个问题，在本文中，我们没有提取两个可能很冗长的验证实体的完整描述，而是建议提取并给出它们的紧凑摘要，并期望这种长度有限的比较实体摘要可以帮助人类用户进行更有效的验证，而不会显着影响其验证的准确性。我们的方法利用了两个实体的共同和不同特征，它们可以最好地指示（非）共指关系，还考虑了有关其身份的各种信息。实验结果表明，使用我们的比较实体摘要进行验证时，验证速度提高了2.7-2.9倍，并且准确性没有受到明显影响。 |
| 课程简介: | A primary challenge to Web data integration is coreference resolution, namely identifying entity descriptions from different data sources that refer to the same real-world entity. Increasingly, solutions to coreference resolution have humans in the loop. For instance, many active learning, crowdsourcing, and pay-as-you-go approaches so licit user feedback for verifying candidate coreferent entities computed by automatic methods. Whereas reducing the number of verification tasks is a major consideration for these approaches, very little attention has been paid to the efficiency of performing each single verification task. To address this issue, in this paper, instead of showing the entire descriptions of two entities for verification which are possibly lengthy, we propose to extract and present a compact summary of them, and expect that such length-limited comparative entity summaries can help human users verify more efficiently without significantly hurting the accuracy of the their verification. Our approach exploits the common and different features of two entities that best help indicate (non-)coreference, and also considers the diverse information on their identities. Experimental results show that verification is 2.7–2.9 times faster when using our comparative entity summaries, and its accuracy is not notably affected. |
| 关 键 词: | Web数据集成; 数据源; 共引用解析; 任务验证; 实体摘要 |
| 课程来源: | 视频讲座网 |
| 数据采集: | 2021-05-12：zyk |
| 最后编审: | 2021-05-26：zyk |
| 阅读次数: | 44 |