How our website works.
Our product scans the ingredients from the photo uploaded by the user and prints a list of them on the website, using OCR. OCR, or Optical Character Recognition, is a software used to read text from a digital image. From a list of ingredients that we inputted into our code and the numbers we assigned them, we trained our website using functions to calculate an average score based off the ingredients it can read. The number presented in front of you is the average score of the ingredients our website was able to read. The higher the average score, the more we recommend a product.
List of ingredients.
We compiled a list of commonly found ingredients in candies and assigned them a number from -1 to 2. We uploaded the list into our code, so the website would be able to recognize ingredients in the product. Individual Scoring Key For Each Ingredient -1: Not recommended for consumption 0: Neutral 1: Recommended 2: Irrelevant After we assigned each ingredient a number, we coded our website to calculate an average score of the ingredients based off of the ones it could recognize from our list and their individual, assigned scores.
What model did you use??
We used this deep learning model called BERT, where it had a good representation of our language. Then we used a two layers fully connected network for classification.
AI Camp teaches middle and high school students to make impressive AI products like this one.