This is a final project for Data Focused Python (95-888) at Carnegie Mellon Univeristy. This course is taught by John Ostlund.

Contributors:

• Harish Karumuri
• Dhruvi Khankoje
• Jiyoon Jung
• Scott McDonald
• Saloni Priyani

Make sure secrets.json is addded to the main directory.

This should be formatted as a JSON file, with an API Key from Spoonacular, which can be obtained here: https://spoonacular.com/food-api. The JSON should be formatted as below.

{
    "apiKey": YOUR KEY HERE
}

(Note to graders -- this should be directly included in the zip file provided in the formal canvas submission).

In order to to run this project, run the main.py file. For example, this can be run directly from the terminal as

$ python3 main.py

This will install pip if it is not yet installed, create a virtual environment, install all dependencies from the requirements.txt file, and then finally open the streamlit app. This may take a few minutes.

Your preferred browser should open the streamlit application automatically, but in the event it does not, go to http://localhost:8501

Once the running indicator stops in the top right corner, type how you are feeling into the textbox. Here are a few examples you can test:

• I'm scared for my exam tomorrow
• I got broken up by my boyfriend
• I had a great day today!

If you would like another food reccomendaiton, hit submit again to get a new reccomendation.

We used the following sources of data:

1. GoEmotions: https://ai.googleblog.com/2021/10/goemotions-dataset-for-fine-grained.html -- we used this data set to train our ML Model
2. Flickr: https://www.flickr.com/search/ -- The Flickr Website was scraped in order to provide a base image for the picture
3. Spoonacular API: https://spoonacular.com/food-api -- The webiste provided us recipe instructions and image.
4. Google Search: https://www.google.com/ -- A query was provided as "FOOD_TYPE_HERE restaurants near me" and the first result was provided to the user.
5. Food preference survey data: https://github.com/priyanshrastogi/mood-based-food-recommender/blob/master/food_choices.csv -- survery data that was parsed

raw mood text -> mood -> food --> recipe and restaurants

• main.py Main code runner

• app.py Runs streamlit app and holds front end app logic

• predictmapping.py Uses pretrained model (model_jlib) in order to condense and classify emotions based on raw text input. Model trained using dfp.py. GoEmotions dataset saved as emotion_dataset_raw

• mood_to_food.py Uses mood_to_food.csv and sets up a dictionary, with mappings. This parsed data was cleaned up using Preprocessing-mood-to-food-csv.ipynb

• food_to_recipe.py Connects to spoonacular api and produces recipe text and an image.

• food_to_restaurant.py Scrape first result google given a query.

• Provide a larger amount of food options
• Improve emotion detecting algorithm
• Get more survey data for food preference to provide more tailored food
• Connect to other services to allow easy purchase of ingredients or food order
• Collect wider range of recipes for reccomending to user

• For the Mood Detection ML model: https://github.com/Jcharis/end2end-nlp-project/
• For processing mood to food reccomendations: https://github.com/priyanshrastogi/mood-based-food-recommender