重视学习者的科学素养的培养与当前我国的发展战略与人才培养目标高度契合，科学论证作为一种核心的科学实践活动，对于科学素养的培养具有重要意义。而学习者的科学论证能力的培养以论证能力的评价为基础。科学论证图作为一种论证能力培养与评价方式具有其独特的优势，自动化评价的实现能够有力推动科学论证图在教学实践中的应用，助力学习者的科学论证能力的培养。而当前基于科学论证图的科学论证能力自动测评研究仍有待丰富。

　　本研究探索了基于科学论证图对学习者的科学论证能力进行自动化测评的方案。基于对以往研究中科学论证能力评价指标的综合分析，本研究构建了初步的基于科学论证图的十个科学论证能力评价特征，对每个特征进行界定与等级划分，形成评价框架。基于所构建的分析式评价框架，本研究为其中九个特征变量制定了计算规则，采用基于规则匹配实现了自动化提取。而针对关联性这一特征变量，本研究综合采用了自然语言处理与机器学习方法，通过将文本转化为概念图，从中提取出复杂网络特征作为算法输入数据，采用逻辑回归算法构建预测模型实现了对关联性的自动化提取。结合相关分析、基线分类器、信息增益等多种方法对自动化提取出的特征进行综合评估，并参照模型预测准确率、F1的宏平均和F1的微平均多个指标对算法模型进行综合评估与选择，最终采用SMO算法构建出了准确率达到79.2079%的基于科学论证图的科学论证能力自动测评模型。

　　研究发现，结合自然语言处理与机器学习可实现基于文本对科学论证中正当理由的有效性进行自动化评分；基于分析式的评价框架，采用机器学习能够构建出具有较高准确率的基于科学论证图的科学论证能力自动测评模型；学习者在科学论证中的反思与批判能力值得单独并重点评估。在此基础上，本研究提出以下后续发展方向：第一，开展自动测评模型的教学应用研究，从学习过程的视角深入分析探究自动测评模型在提升科学论证教学效果方面的作用。第二，进一步探索更多评价特征的基于文本的自动化提取方案，提升测评模型的智能化水平。第三，针对科学论证中的批判反思展开深入探索，促进学习者的批判性思维的培养。第四，探索如何基于自动测评模型开展有效的反馈，包括提供反馈的维度数量、反馈的提供形式，面向不同类型的学习者群体的反馈方案的设计以及可视化反馈的科学设计等。

　　Attaching importance to the cultivation of learners' scientific literacy is highly consistent with China's current development strategy and talent training objectives. As a core scientific practice, scientific argumentation is of great significance to the cultivation of scientific literacy. Based on the evaluation of argumentation ability, learners' scientific argumentation ability is developed. As a way of cultivating and evaluating argumentation ability, scientific argumentation map has its unique advantages. The realization of automatic evaluation can effectively promote the application of scientific argumentation map in teaching practice and help the cultivation of learners' scientific argumentation ability. At present, the research on the automatic evaluation of scientific argumentation ability based on scientific argumentation map still needs to be enriched.

　　This study explored the realization of accurate automated evaluation of learners' scientific argumentation ability based on the scientific argumentation map. Based on the comprehensive analysis of the evaluation framework of scientific argumentation ability in previous studies, this study constructed ten preliminary evaluation features of scientific argumentation ability based on the scientific argumentation map, then defined and graded each feature to form an evaluation framework. Based on the constructed analytical evaluation framework, this study made calculation rules for nine of ten features, and realized their automatic extraction based on rule matching. Besides, natural language processing and machine learning were used to automatically extract the feature of relevance. By transforming the text into a concept map, extracting the complex network features as algorithm input data, and using the logistic regression algorithm to build a prediction model, the automatic extraction of relevance was realized. Combined with correlation analysis, baseline classifier, information gain and other methods, this study comprehensively evaluated the automatically extracted features. The prediction accuracy, the macro average of F1 and the micro average of F1 were used to evaluate and select the algorithm model. Finally, the SMO algorithm is used to build an automatic evaluation model of scientific argumentation ability based on scientific argumentation map with an accuracy of 79.2079%.

　　It was found that the combination of natural language processing and machine learning can realize the automatic scoring of the effectiveness of warrants in scientific argumentation based on text. Based on the analytical evaluation framework and machine learning, an automatic evaluation model of scientific argumentation ability based on the scientific argumentation map with high accuracy can be constructed. Learners' reflective and critical abilities in scientific argumentation deserve to be evaluated separately and emphatically. On this basis, this study puts forward the following follow-up development directions: first, apply the automatic evaluation model in teaching practice, and deeply analyze and explore the role of automatic evaluation model in improving the effect of scientific argumentation teaching from the perspective of learning process. Second, further explore more automatic text-based extraction schemes of evaluation features to improve the intelligent level of evaluation model. Third, carry out in-depth exploration on the critical reflection in scientific argumentation to promote the cultivation of learners' critical thinking. Fourth, explore how to carry out effective feedback based on the automatic evaluation model, including the number of dimensions and forms of feedback, the design of feedback schemes for different types of learners and the scientific design of visual feedback.