WET: Word embedding-topic distribution vectors for MOOC video lectures dataset

Area of Science:

• Natural Language Processing
• Machine Learning
• Educational Technology

Background:

• Massive Open Online Courses (MOOCs) generate vast amounts of textual data from video lectures.
• Analyzing this unstructured text data is crucial for understanding course content and improving educational tools.
• Existing methods often struggle to capture the semantic nuances and thematic structures within lecture transcripts.

Purpose of the Study:

• To present a comprehensive dataset of word embeddings and document topic vectors derived from MOOCs video lecture transcripts.
• To make this dataset publicly available for researchers in Natural Language Processing and machine learning.
• To facilitate advancements in AI-driven video classification and educational data mining.

Main Methods:

• Collected transcripts from 12,032 video lectures across 200 courses on the Coursera platform.
• Applied Word2Vec (a word embedding technique) to generate vector representations of words.
• Utilized Latent Dirichlet Allocation (LDA) to determine topic distribution vectors for documents (lectures).
• Implemented Word2Vec and LDA using the Gensim package in Python.

Main Results:

• Generated a large-scale dataset of word embeddings capturing semantic relationships within lecture content.
• Produced document topic vectors that represent the thematic structure of each video lecture.
• The dataset is linked to a research article on integrating these features into a video classification framework.
• The dataset is accessible via the Mendeley Data repository.

Conclusions:

• The presented dataset offers valuable resources for NLP and machine learning research, particularly in educational contexts.
• This data can significantly aid the development of sophisticated video classification models.
• Enables further exploration of deep learning applications for analyzing educational video content.