Suggested approaches to Curate content, resources, and tools for the user
Approach with Navigator
Video Segmentation: Videos are an increasingly popular method of delivering lectures online. A lot of learning videos on popular learning platforms have lengths up to 120 minutes, in which the lecturer discusses several sub-topics under the ambit of a broader topic. This creates the need for the video to be searchable so that the student can switch to the topics they are interested in learning. We seek to automate the process of generating timestamps where the transitions in topics take place in the videos since manual segmentation is expensive and slow. There is little visual information which can be made use of to detect scene transitions in learning videos. We hence make use of the audio information encoded in the video by generating the video transcript using Automatic Speech Recognition. We propose two models that automatically detect time stamps where topic transitions take place using Natural language Processing.
Efficacy: Measures how effective a learning resource is in making an observable difference in the competency obtained by the learner
Engagement: Measures how engaging a learning resource is to learners (in terms of visits, likes, shares) while they try to acquire a particular competency
Relevance: Measures how relevant a learning resource is to acquiring a particular competency
Title, Description: Extracted from the standard metadata of the resource
Transcript: A transcript service deploys a specific library to extract the text based on type of resource – video, static and dynamic web pages, audio, interactive, pdf etc.
Summary: Generates both extractive and abstractive summary of a resource from the transcript for that resource
Phrase Cloud: Rake NLTK library used to derive the phrase cloud in a body of text by analyzing the frequency of word appearance and its co-occurrence with other words in the text
Classification: Auto classify content based on competency, lexile level, audience and other attributes using ML/Deep Learning techniques.
Learn: Indicates the opportunity to learn a competency and is computed based on the number of learning activities tagged to the competency and their REEf values
Assess: Indicates the opportunity to assess a competency and is computed based on the number of assessments tagged to the competency and their REEf values
Interest: Indicates the interest in a competency and is computed based on the number of learners pursuing the competency, factoring in the progress – activities completed, assessments attempted
Talent: Indicates the supply of talent for a competency and is computed based on the number of learners who have mastery in the competency, factoring in confidence, decay, DoK and other measures for the learner
Image2Data: Photos of students’ essays, proofs and other assignments are converted using OCR and NLU techniques both for record keeping and evaluation. Preprocessing is done to identify written vs printed documents before using CNN and RNN trained models for chunking and then applying OCR to extract text from the chunks.
ImageScoring: Use NLU techniques to score, using rubrics, the student answers in the converted text from Image2Data.
AI Techniques and Tools
Video Segmentation
Automatic Speech Recognition to generate video transcript
Pre-trained sentence embeddings (InferSent) to represent sentences
Pre-trained word embeddings (Word2Vec) to represent words
0 Comments