Source-linked AI summary
Text mining in education
R. Ferreira-Mello, M. Andre, A. Pinheiro, E. Costa, C. Romero
TL;DR
Online education produces large volumes of unstructured text, but no survey had synthesized its educational text-mining applications. This paper systematically reviews studies from 2006 to July 2018 to characterize techniques, resources, and goals, finding that text classification and natural language processing are the main methods.
Problem
Online education generates large volumes of unstructured data, while the growing educational text-mining literature lacked a survey of its techniques, resources, and applications.
Method
The paper conducts a literature review of educational text-mining applications published between 2006 and July 2018.
Results
Text classification and natural language processing are the main text-mining methods reported in the educational literature.
Takeaways & Limitations
The review organizes the field’s principal techniques, educational resources, goals, and future trends.
Takeaways & Limitations
The reviewed literature is more focused on output than process, producing systems with good accuracy but limited interpretability.
Abstract
from arXiv · showhide
The explosive growth of online education environments is generating a massive volume of data, specially in text format from forums, chats, social networks, assessments, essays, among others. It produces exciting challenges on how to mine text data in order to find useful knowledge for educational stakeholders. Despite the increasing number of educational applications of text mining published recently, we have not found any paper surveying them. In this line, this work presents a systematic overview of the current status of the Educational Text Mining field. Our final goal is to answer three main research questions: Which are the text mining techniques most used in educational environments? Which are the most used educational resources? And which are the main applications or educational goals? Finally, we outline the conclusions and the more interesting future trends.
1 | INTRODUCTION
Online education generates extensive unstructured text, while existing educational data methods do not fully exploit resources such as forums, chats, social networks, open questions, and essays. This paper surveys Educational Text Mining to identify commonly used techniques, resources, applications, and future trends.
- Online learning platforms generate large volumes of structured and unstructured data from forums, chats, wikis, blogs, open questions, and essays.
- Educational Data Mining and Learning Analytics improve learning and support teachers, but do not fully explore available educational resources.
- Text mining extracts high-quality information from unstructured text and can be applied to educational platforms.
- Educational text mining has produced significant results in essay, forum, academic-text, and open-question analysis.
- The review examines text-mining techniques, educational resources, and applications by addressing three research questions.
2 | METHODOLOGY
The review searched major academic databases for educational text-mining studies published from January 2006 to July 2018, applying explicit relevance and scope criteria. It retained 343 papers and classified them by methods, resources, goals, citations, publication year, and publication type.
- The search used IEEE Xplore, Springer Link, ScienceDirect, ACM, and Google Scholar with education-related text-mining keywords.
- Three exclusion criteria removed papers lacking relevant keywords, educational goals or applications, or publication dates within January 2006–July 2018.
- 343 papers remained after the exclusion process, with IEEE Xplore, Springer, and ACL identified as the three most important sources.
- Publications were categorized by text-mining methods, educational resources, goals, citations, publication year, and publication type.
- Only seven papers appeared in the first study year, the average exceeded 26 papers per year, and 56 appeared in the first seven months of 2018.
- Conference proceedings accounted for 59% of publications, followed by journal papers at 25%, workshop papers at 15%, and other publications at 1%.
3 | TEXT MINING METHODS AND TECHNIQUES
Educational text-mining research most often uses text classification and natural language processing, with additional applications spanning clustering, information retrieval, summarization, sentiment analysis, assessment, feedback, and recommendation. These methods are applied across essays, open questions, forums, chats, and other educational texts.
- Most-used methods: Text classification accounted for 32% of studies and natural language processing for 31%, followed by theoretical work at 12%.
- Most-used methods: Information retrieval and text clustering each represented 5% of studies, while text summarization represented 4%.
- Classification and clustering: Classification and clustering support goals including discourse categorization, student grouping, curriculum adaptation, engagement measurement, learning-pattern identification, and recommendation.
- Natural language processing: Natural language processing is used for essay cohesion and argumentation, semantic evaluation and question generation, feedback, collaborative chats, and interaction or project-performance prediction.
- Information retrieval: Information retrieval helps students find relevant books, organize and navigate texts, visualize essay topics, enhance discussion, and improve recommendation systems.
- Text summarization: Automatic text summarization is applied to forum evaluation, academic writing support, collaborative learning, feedback evaluation, content summarization, keyphrase extraction, and visual modeling.
4 | EDUCATIONAL SOURCES AND RESOURCES
The survey organizes educational text-mining research by the resource analyzed, with essays, forums, and online assignments forming the most prominent resource categories. It also covers question creation and evaluation, essay assessment, and forum-based communication and participation.
- Essays, forums, and online assignments account for more than 50% of the retrieved works.
- Online assignments: The survey includes question creation and evaluation, including automatically generating questions and multiple-choice answers with natural language processing and template-based algorithms.
- Essays: Essay-related text mining evaluates students through discourse analysis, cohesion measures, linguistic features, and classifier-based scoring.
- Essays: Essay analysis also supports writing development through automatic feedback and collaborative-learning approaches based on text mining, maps, and probabilistic topic models.
- Forums: Forums provide asynchronous communication but create information overload, motivating automatic extraction, feedback, recommendation, classification, and topic-relevance methods.
- Forums: Forum research includes topic-driven search, similarity-based answer recommendation, participation assessment, post classification, and semantic-statistical modeling of discussion context.
5 | EDUCATIONAL GOALS AND APPLICATIONS
Educational text mining is applied mainly to evaluation, followed by student support and analytics. The surveyed applications span assessment, writing support, interaction analysis, question generation, feedback, recommendation, and content organization.
- Evaluation is the most popular educational application of text mining, followed by student support and analytics.
- Evaluation: Evaluation applications assess essays, online assignments, student performance, questions, interactions, and formative-learning activities.
- Evaluation: Essay evaluation has progressed from shallow features and word counts toward semantic, writing-style, cohesion, and argumentation analysis.
- Student support: Student-support applications assist writing, provide feedback, encourage collaboration, analyze sentiment, and help maintain motivation or prevent dropout.
- Analytics: Analytics applications extract information from writing, forums, chats, and email to analyze student performance, behavior, characteristics, and learning interactions.
- Question-related applications: Other applications generate, rank, scaffold, and evaluate questions or provide teachers with content and course-context support.
- Feedback: Feedback systems deliver recommendations or dashboards directly to students, support collaborative environments, or help instructors formulate feedback.
- Recommendation and content organization: Recommendation systems help students find relevant information by using writings, access history, books, or web content, while other systems organize learning objects semantically.
6 | CONCLUSIONS AND FUTURE RESEARCH
This systematic review synthesizes Educational Text Mining research across techniques, resources, applications, and educational goals, based on 353 relevant papers. It identifies dominant methods and resources while outlining future directions including deeper text interpretation, writing analytics, multilingual mining, generation, and sentiment analysis.
- Review scope: The review analyzed 353 relevant papers to answer questions about educational text-mining techniques, resources, and goals.Its contribution is a literature review of text-mining applications in online education.
- Main techniques: Text classification and natural language processing are the main text-mining methods used in educational literature.The review notes that this pattern follows general text-mining applications.
- Future research: Future research should address deeper interpretation through text extraction, summarization, feature engineering, and writing analytics for text resources.The review also identifies limited analytics for text resources and anticipates improvements in teacher feedback, student monitoring, and evaluation.
- Resources and goals: Online assignments and essays are heavily explored because automatic evaluation and feedback are major educational goals.Student support also includes self-reflection, instructor analytics, recommendation systems, collaborative writing, and answer suggestions.
- Resources and goals: Social networks and documents also receive substantial attention, supporting community applications and document recommendation in digital libraries.Forums and chats are additionally identified as relevant textual resources in the literature.
- Future research: Other proposed directions include educational content generation, multilingual text mining, broader sentiment analysis, and improved attention to online discussion participation and collaboration.Suggested applications include generating questions, interacting in forums and chats, and proposing essay themes; sentiment analysis should extend beyond forums, chats, and social networks to essays and assignments.