Detect Distraction

[gaurav0535]

Problem Statement: I think the big problem that any EDTECH startup is facing is how to engage people in learning because people are getting distracted while learning complex topics for me it was backtracking. And I think scaler or anyone else should provide a more personalized experience for this.

The big problem is a scaler doesn't know if the person is struggling to learn that topic if the person doesn't bring this to the notice of scaler people . What can be done about this? If we know the context of the web page then we can get some information like when are users getting distracted how soon are users getting distracted what could be the reason?

Solution: Identify the relevance of the learning topic with the current web page-

1. Track the user activity by using chrome extensions
2. Store the metadata information of the web page that person is crawling while studying
3. Get the context of the web page we can achieve this by below ways-

• This method is already used in the current approach wherein we can use spacy to do the POS(part of speech) tagging to
  get the nouns and we can compare them with the learning topic , if it is matching for more than a certain threshold then it is
  relevant . In this approach, we are not sure about the context of the text for example we won’t be able to understand if
  the user is trying to learn a python programming language and landed to python biology one. In this case according to
  current approach there is a big chance that the page will be marked as relevant

• If we have a fixed number of topics that user want to study then we can use sequence to sequence models (deep learning
  techniques like LSTM, transformers ) to train the model to predict the learning topic, this approach is the most accurate
  one but the problem is sequence models need a large amount of data to get trained although we can use transfer learning
  but it will not give that good results and another problem is fixing a number of learning topic will limit the functionality of the extension

• We can use topic modeling for this, it is a technique to extract the topic from the text behind the scene it uses a technique called LDA (Latent Dirichlet Allocation http://blog.echen.me/2011/08/22/introduction-to-latent-dirichlet-allocation/) . It will discover the topic and further, we can compare this with our learning topic and check how much it is matching. If the content of the web page is very big then we can use the google pre-trained T5 model to summarize the topic it is abstractive summarization that is it will extract the summary from the main idea(context based ) rather than extracting and concatenating the main sentences of the corpus. We can use the GENSIM library to achieve topic modeling. Using summarization will definitely add overhead to the AI engine

Further we can store these things in DB and we can analyze how to improve the services

PS:I have mentioned Scaler please don't mind