Peace Analyzer

A webapp to determine if online content is mentally harmful or not by converting a given data into text data with assembly-ai and testing the data through a neural network model.

Comment

Tracks

  • Best use of AssemblyAI api
  • Mental Health Track Best Hacks
  • Best UI/UX Designed Application
  • Most Creative Idea

Inspiration

A recent study shows that online content that contains negative words may affect on our mental health negatively. Furthermore, due to covid pandemic, we have lesser social interaction with others, which worsen our mental condition as well. Indeed, a data shows that the number of cases in depression has increased over 16% comparing with pre-covid.

What it does

To prevent people from having an unhealthy mental, we implemented a webapp, "peace analyzer", to determine if a online content is harmful or not.

How we built it

Peace Analyzer works as followings:

  1. User uploads a video or an article to peace analyzer.
  2. Assembly Ai API convert the audio data into text data.
  3. Neural Network model will test the text data and output the probability. If the probability is less than 50%, then the content is likely to be unhealthy. If the probability is more than 50%, the it is likely to be healthy.

We trained the neural network with Sentiment-140, a large text dataset of tweets. By using Sentiment-140, the model enables to adapt wide range of topics and content of the given data. By testing this model with a splitted testing data, we obtained that 78% accuracy, which is considered as a good model.

Challenges we ran into

At the process of building a neural network model, due to its complexity, it took 2 hours to run the entire model in jupyter notebook. So, we switched to google colab and setup GPU to run the program, which was much faster than the original running time.

Accomplishments that we're proud of

We are very proud of especially two elements:

  1. Fast compiling of Aditya: He implemented front-end as well as API in very short period of time, which enabled us to review our product and yield time to improve.

  2. Implementation of Neural Network on NLP I, Manami Kanemura, has some experience in machine learning, but not in neural network. It may correspond to the next section, but it was a very decent 24 hours to understand the logic behind of neural network and libraries (i.e., keras) and implement them into code. And we could eventually reach high accuracy with the model.

What we learned

Throughout this hackathon, we learned the following:

  1. Time management We are the group of two people, so we had to allocate our work and proceed the individual work while having a discussion on the product. Also, Aditya lives in India, we had to manage with time differences too. It all went well and it was great that we could submit our product on time.

  2. Neural network with NLP NLP is such a complicated topic in deep learning, and neural network is also a complicated algorithm. Since I was working on it for entire 24 hours, my brain got exhausted but I could learn the background and implementation of Neural network on NLP. I would not be able to do so without such a deep concentration.

What's next for Peace Analyzer

There are plenty of applications for peace analyzer. By changing a training data for neural network model, we may apply to

  • parents' control for children's safe usage on online content -> determine if a content is clear from any harmful contents.

  • political opinion checker -> determine which political opinion an author has based on his/her articles.

In terms of the peace analyzer itself, we could improve UI/UX so that the user do not need to lurk inspect pages, where no body knows except computer science people. User-friendly product will gather many users, which may promote our peace analyzer.

Built With

Share this project:

Updates