Inspiration

Our inspiration for Comprendly was finding a way to test our understanding of topics we learned in school. Especially when studying for exams, we believed it would be useful to create an app that could test one’s understanding of a topic so they can know what they may need to study and what they’ve mastered.

What it does

Our app teaches a user what they already know about a topic as well as what topics they may not know. By inputting a link to a Wikipedia article or by copy pasting your own document and recording yourself explaining what you already know about the topic, our app compares the two to identify differences in the transcript and link/document, and lets you know what you missed and what you included.

How we built it

We built the app using Java and Android Studio, and creating an original design in Figma.

Challenges we ran into

One of the challenges we ran into had to do with implementing NLP (DillBERT model) and Machine Learning into Comprendly. Our idea was to create an app that would use NLP to compare the two texts for similarities and differences in meaning, but as we did research and a large amount of trial and error in python, we realized that comparing meaning was a difficult task to be accomplished in a short period of time. In our GitHub link, we have attached a Python Script of what we were able to accomplish, including using a keyword extractor (in our case, YAKE) to extract key words and phrases from a Wikipedia article, as well as using cosine similarity to compare those key phrases with the user input.

What we learned

Throughout this process, although we were unable to successfully implement NLP and a python backend in our final product, we learned a lot about NLP, including how different keyword extractors function to rank keywords in a string, as well as how cosine similarity works to compare two strings by converting them into non-zero vectors, and taking the angle between the vectors to calculate similarity. In android studio, we learned about permissions and specific accessor problems, but overall we worked through our problems efficiently, learning a lot along the way. .

What's next for Comprendly

In the future, we plan use machine learning and NLP to improve the comparison between the chosen topic and the audio transcription. We also plan to add an option to take a picture of a textbook page and extract text from there.

Built With

Share this project:

Updates