Research On The Learners' Topic Detection And Its Evolutionary Analytics In The SPOC Discussion Forum

SPOC course discussion forum,considered as one of the main channels of communication and collaboration for learners to participate in the course interaction,provides significant basic data to support the evaluation of learners' learning process.Learners have produced a large amount of discourse data in course discussion forum,and the accumulation of these data generally brings the cognitive load to learners and teachers.As a result,it is difficult to rely entirely on the way of manual browsing and annotation.It is urgently necessary to automatically mine the hidden valuable information,such as,learners' topical content,lerners' emotional tendencies and behavioral patterns.In addition,the text-based discourse data generated by learners has a significant life-cycle characteristic.Therefore,the analysis of text data plays a key role in the accurate mining of learner's discourse behavior and content in the context of online learning platform.In order to capture learners'focused topics automatically and these topics' evolutionary trends,how to model learners' discourse semantic information and track them dynamically bring us new research perspectives and challenges.This paper aims to automatically detect and dynamically track multi-dimensional semantic information of learners' focused topics in the course discussion forums of colleges,as well as to explore the inter-relationships between them and the learning outcomes,which can be utilized to the feedback of learning process,the adjustment of teaching methods,the management of platform construction and so on in the SPOC platform.Moreover,the analytical results are beneficial for providing guidance in teaching practices and optimization of learning experience.In this paper,the learners-associated textual data in the discussion area of SPOC is used as the analysis object.According to the single and static characteristics of the topic detection in the course discussion area,and the neglect of learners' discourse content semantics related to their performances.We present the based research route called "topic detection-topic evolution-the application of learning outcomes".In order to mine the hidden discourse content,the discourse behaviors and associated emotion information generated by learners in the forum discussion area is first used to construct a topic model that combines multi-dimensional features.Moreover,the algorithm performance and practical application of the model ability are verified.Second,combined with the time characteristic of the learners' discourse expression,a dynamical topic model is introduced to dynamically track the evolution of the topic.The performance comparison and contextual application of the model are also discussed.Finally,this study explores the relationship between learners' final academic achievements and learners' explicit discourse behavior and implicit discourse content in the course discussion forum.The main research works and innovations of the dissertation are as follows:(1)Considering that the topic detection has a single dimension,this paper introduces a topic model combining learners' discourse behaviors and emotion trends.This model weakens the hypothesis of the original standard topic model by determining the dependence of the learners'discourse behaviors,emotions and semantic content features in the context of the online course discussion forum,and takes the learner's personal document as the model input object.The word is specified as the minimum unit of topic sampling.That is,each word in each discussion post,involves a single topical,emotional,and behavioral category.In this way,we can provide a formalized abstract description of the visualization of the solution and establish the starting point of the quantitative research,thus the mapping spaces of learners-topics probability distributions,topics-semantic concepts,topics-emotional information and topics-behavioral characteristics are able to be established.The experimental results show that compared with the traditional topic models,the proposed method presents stronger model generalization ability and obtains better effect in generating topic diversity and topic aggregation degree.In addition,we find that,in terms of the practical applications of topic mining towards the course and personal level,the model can better mine the topic-related semantic information.(2)Considering the limitation of static feature in topic detection,this paper shows a dynamical topic model into which integrates the time variable of learners' posts.The topic model utilizes the learners' time attribute of the discussion posts to correlate the learners' focused topic content,and constructs the probability distribution matrix of the topics over time.Moreover,each document published by the learner is considered as the model input object.In order to constrain the generation of the topic sampling,the word is treated as the smallest building unit of the dynamic topic model.Thus,we can dynamically track the evolutionary strength and content tendencies of the learners'focused topics in different time units.The results show that compared with the traditional time discretization methods,the proposed dynamic model has better advantages in generating the quality of dynamic topic.In addition,this model has the ability to detect the evolutionary intensities and content of dynamic topics well in mining dynamic topic information for curriculum level and personal level.(3)Considering that the existing researches generally neglect the learners' discourse content related to the learners' final performance,this paper combines the explicit learners' discourse behaviors and the implicit learners' discourse content to comprehensively discuss the relationship between the learners' discourse differences and their learning achievements.To this end,this study,on the one hand,explores discourse behavior and content variables that affect the learning outcomes of learners.This study,on the other hand,investigates the differences in the interactive discourse behaviors,the focused topics and emotional attitudes among the different academic performance groups.The experimental results show that there are significant positive correlations between the number of the learners' common thread posts,the topic posts,the total posts and the academic performance of the participants in the course discussion forums;there is a significant correlation between some focused topic content of learners participating in the course discussion area and their learning outcomes;the positive/negative emotional intensities of participants in the course discussion forum is significantly positively/negatively correlated with their academic performance.Besides,there was a significant difference in the distribution of thematic posts,replies and ordinary posting behaviors between the high and low achievement groups;there was a significant difference in the topics discussed between the high and low performance groups;the hidden emotional attitudes of the learners' interactive discourse between the high and low performance groups do not show any differences.