我国已全面踏入教育现代化与教育信息化建设的新征程，在线教育是其中重要的组成部分，利于促进教育公平、推进个性化教育。随着在线学习的发展，教师教育的形式发生巨大转变，在线教师培训已成为教师专业发展的一个重要途径。教师在线培训不受时间和空间限制，但在一定程度上会阻碍学习者之间的情感交流，有时会严重影响师生关系和教育绩效。诸多研究表明，情感对学习者具有重要意义。当学习者有积极的态度体验时，就会加强对所学内容的情感联系，进一步提高学习投入度。然而，目前对于在线学习情感研究中仍存在一些问题。例如，对于在线学习情感分析的数据来源主要通过分析学习者在线交互文本，然而，并不是所有的学习者在学习过程中都会产生文本评论，学习者实时发生的情感会围绕学习事件产生特定的学习行为，两者具有显著的相关性。因此，学习行为表现也能反映出学习者的情感状态。另外，在情感分析上，简单的统计学习者的情感表达频次等无法深入刻画学习者情感变化，情感状态具有一定的情景特征、学习者个性特征，以及随时序变化的动态变化特征。因此，在在线学习中，不仅需要关注学习者每一个学习阶段的学习行为，挖掘和分析学习者发布的实时评论中的情感状态，同时描述学习者在线学习过程中的情感变化趋势，对学习者的情感状态变化进行可视化呈现也尤为重要。

本研究聚焦于基于时间序列分析的在线情感分析，以数据建模法为核心研究方法，通过德尔菲法和层次分析法确定可以映射为情感的学习行为评价指标。并采用九维情感分类框架对第三期教师在线学习中产生的5720条文本进行人工标注。然后通过Bert算法进行构建与情感分析模型训练。并采用正交多项式回归分析统计方法构建正交多项式回归分析模型进行时间维度的时序性情感数据分析，最终构建出基于时间序列分析的在线学习情感分析模型。为了验证模型的有效性及其对教师学习者在线学习效果的影响，本研究采用准实验研究法基于在线学习平台开展主题为《混合式理念下的项目式学习》的教师在线培训课程，将基于时间序列分析的在线情感分析模型嵌入到在线学习平台进行应用实践，为教师学习者提供基于时间序列分析的在线学习情感分析反馈。同时，通过问卷调查法和访谈法对教师学习者进行调查和分析，以深入了解模型产生效果的原因。

结合模型的实践应用、数据分析和对学习者的调查访谈，得出本研究结论：基于时间序列分析的在线学习情感分析模型及其应用可有效提升教师学习者的讨论积极性，减少学习者流失，降低辍课率，帮助教师学习者促进学习投入，积极进行学习行为调节和认知情感调节；积极调节学习者自身在在线学习过程中的情感状态，激发“愉悦”等积极情感的产生。本研究构建的基于时间序列分析的情感模型在在线学习中对于学习者和教师都具有较高的应用价值，一方面能促进学习者进行自我积极情感和学习行为的调节，另一方面能促进教师根据反馈状态及时进行课程内容、难度、活动等方面的调节，有利于提高教师在线培训效果，提升在线学习质量。

Our country has embarked on a new journey of education modernization and education information construction in a comprehensive way, online education is an important part of which is conducive to promoting educational equity and promoting personalized education. With the development of online learning, the form of teacher education has undergone great changes, and online teacher training has become an important way of teacher professional development. Online teacher training is not limited by time and space, but to some extent, it will hinder the emotional communication between learners, and sometimes seriously affect the teacher-student relationship and educational performance. Many studies have shown that emotion is of great significance to learners. When learners have positive attitude experience, they will strengthen the emotional connection to what they have learned and further improve their involvement in learning. However, there are still some problems in the research of online learning emotion. For example, the data source of online learning emotion analysis is mainly through the analysis of learners' online interactive text. However, not all learners will generate text comments in the learning process, and the real-time emotion of learners will generate specific learning behaviors around learning events, and the two have significant correlation. Therefore, learning behavior can also reflect the emotional state of learners. In addition, in the aspect of emotion analysis, simple statistics such as the frequency of learners' emotion expression cannot deeply describe learners' emotional changes. Emotional states have certain situational characteristics, learner's personality characteristics, as well as dynamic change characteristics of changes at any time order. Therefore, in online learning, it is not only necessary to pay attention to the learning behavior of learners at each learning stage, dig and analyze the emotional state in the real-time comments published by learners, but also to describe the emotional change trend of learners in the process of online learning. It is also particularly important to visually present the changes of learners' emotional state.

This study focuses on online emotion analysis based on time series analysis, takes data modeling as the core research method, and determines the evaluation index of learning behavior that can be mapped as emotion by Delphi method and analytic hierarchy process. In addition, 5720 texts produced in the online learning of teachers in the third period were manually labeled by using the nine-dimensional emotion classification framework. Then, Bert algorithm was used to construct and train the sentiment analysis model. And the orthogonal polynomial regression analysis statistical method is used to construct the orthogonal polynomial regression analysis model to analyze the temporal emotion data of the time dimension. Finally, the online learning emotion analysis model based on the time series analysis is constructed. In order to verify the effectiveness of the model and its impact on the online learning effect of teacher learners, this study adopts the quasi-experimental research method to carry out the online teacher training course themed "project-based Learning under Hybrid Concept" based on the online learning platform. The online sentiment analysis model based on time series analysis is embedded into the online learning platform for application practice. To provide teacher learners with feedback of online learning emotion analysis based on time series analysis. At the same time, questionnaires and interviews were used to investigate and analyze the teacher-learners, so as to deeply understand the reasons for the effect of the model.

Combined with the practical application of the model, data analysis and survey and interview of learners, the conclusion of this study is drawn: the online learning emotion analysis model based on time series analysis and its application can effectively improve the discussion enthusiasm of teacher learners, reduce learner loss, reduce the dropout rate, help teacher learners to promote learning involvement, and actively regulate learning behavior and cognitive emotion. Actively regulate the emotional state of learners in the process of online learning, and stimulate the production of positive emotions such as "pleasure". The emotion model based on time series analysis constructed in this study has high application value for both learners and teachers in online learning. On the one hand, it can promote learners to adjust their self-positive emotions and learning behaviors; on the other hand, it can promote teachers to adjust course content, difficulty, activities and other aspects timely according to the feedback state, which is conducive to improving the effect of online teacher training. Improve the quality of online learning.