基于众智的学习资源知识点关联标注方法研究

The correlation annotation between learning resource and knowledge element is an important and foundational job, manual labelling is a time-consuming and laborious job, and automatic annotation based on machine learning is rather difficult for the resource correlated with tacit knowledge, and it relies on the large scale data set. Finishing the job of correlation annotation between learning resource and knowledge element based on abound of heuristic knowledge of collective wisdom is an efficient and feasible solution. From the perspective of the assignment requirements and user attributes, this research analysis the factors affecting the credibility of the crowdsourcing assignment, put forward the assignment credibility computational model based on knowledge ability evaluation and users’ participate willingness, and gives the quantitative calculation method. For the assignment allocation, in order to ensure the quality of the crowdsourcing assignment, and taking into account the opportunity for new user to get crowdsourcing assignments at the same time, so the fresh uses can be trained into expert users, this research focuses on assignment allocation mechanism for the double target of assignment quality and user training. Finally, this research establish the deep learning model based on the features of collective wisdom, and use the text resource in the form of test as research object. Combination the collective wisdom features and text features, this research proposes a combinatorial multi classifier, and focus on the correlation knowledge element fusion calculation based on combinatorial multi classifier. The method of correlation annotation between learning resource and knowledge element based on collective wisdom will have a far-reaching impact on the application of adaptive online learning platform, educational big data learning and analysis.

学习资源的知识点关联标注是一项基础而又重要的工作，手动标注费时费力，基于机器学习的自动标注，对于存在隐性知识关联的资源较为困难，且依赖大规模样本集。因此利用众人众智的先验知识完成学习资源知识点关联标注工作，成为一种高效、可行的解决方案。本项目从任务需求和用户属性出发，分析影响众智任务可信度的因素，提出基于知识能力评价和参与意愿的众智任务可信度计算模型，并给出定量化计算方法。对于众智任务分配，在保证众智任务质量的同时，兼顾新用户也能获得任务分配机会，训练成长为专家用户，重点研究面向任务质量和用户训练双目标的众智任务分配机制。最后，建立基于众智特征构建深度学习模型，同时以试题文本资源作为研究对象，融合资源文本向量特征，构建组合多分类器，研究基于组合多分类器的关联知识点融合计算。基于众智的学习资源的知识点关联标注方法的研究，将对自适应在线学习平台、教育大数据学习与分析等方面应用产生深远的影响。

课题针对如何将深度学习模型与众人众智的先验知识相结合，应用于学习资源与知识点关联标注的研究。本项目在分析影响众智任务因素的基础上，提出众智任务可信度定量化计算方法。在众智任务的分配方法上，项目研究不同目标任务下的众智任务分配机制。同时利用机器自动标和众智标注各自优势，实现对标注结果的融合。最后基于学习资源的知识点关联标注展开典型的教育应用。.本课题主要研究成果包括：1）从工作者维度和任务两方面，研究影响众智任务分配影响因素，并提出众智任务工作者属性的向量化表示模型，实现众智任务分配的可定量化计算。2）在研究众智任务分配模型的基础上，提出众智任务分配策略，包括任务完成质量最佳、效率最佳和兼顾任务完成质量及效率三种典型分配策略。3）在众智标注结果融合方面，主要研究了基于用户置信度的标注结果融合方法。4）在机器自动标注方面，重点研究了基于Bi-LSTM和集成学习器的学习资源知识点自动标注方法。5）结合利用机器标注和众智标注各自优势，提出自动标注与众智标注结果的融合方法。6）开展基于知识关联的教育资源教学应用研究，包括基于知识关联的学习资源推荐及学习成绩预期与预测。7）研发了基于众智的学习资源关联标注系统，实现自动标注、众智标注、任务广场、众智任务结算等功能模块。.经过3年研究，课题组完成了预期任务，获得软件著作权2项，申请发明专利2项，开发一套基于众智的学习资源关联标注系统，在国内外重要学术期刊及会议发表学术论文8篇，其中SCI/SSCI收录论文3篇，EI/CSSCI收录论文2篇；培养博士研究生3名，硕士研究生5名。