2

10/25/2024 1
 Project Idea

10/25/2024 2
Brief Description of the Problem:
so what is the problem:
Brief Description of the Problem
"Many individuals struggle to find personalized diet plans and recipes that align with their medical
conditions, dietary goals, and preferences. Traditional approaches often overlook the complexity of
balancing health needs with personal tastes, making it difficult to stay committed to healthy eating
habits. Additionally, healthcare providers need quick and effective tools to recommend personalized
diets for their patients health condition"

So in most cases you can’t make a diet plan with a specific number of calories and in the same time find
a good recipes that match the meals you like and at the same time taking your medical health into
considration

10/25/2024 3
Solution:
Our project addresses these challenges by developing a Recommendation System that
provides:
1. Personalized Diet Plans:
Tailored to individual medical history, fitness goals, and dietary preferences.

2. Recipe Recommendations:
Using collaborative filtering, the system suggests recipes that match the user’s preferences
and align with similar users' patterns.

3. User-friendly Platform:
Built with Streamlit, the application offers an easy-to-use interface where users can input
their data, get personalized diet recommendations, and explore recipes that suit their health
needs.

4. Chatbot
to help the user figure out what he actually need to maintain a healthy lifestyle

10/25/2024 4
Unique Value
What Makes Our Project Stand Out?
•Use of GANs for Dynamic User Profiles:
We employ Generative Adversarial Networks (GANs) to create detailed user profiles, improving the
system’s ability to adapt and recommend based on user behavior.

•Advanced Collaborative Filtering Techniques:

We implemented SVD to predict users’ ratings on recipes, and KNN-based filtering to recommend recipes
based on similarities between users’ preferences.

•Continuous Learning & Feedback Loop:

The system evolves based on user interactions, improving recommendations over time by incorporating
user feedback and new input.

•Healthcare-Focused:
Unlike generic recommendation systems, our platform emphasizes health-oriented solutions, helping
users get a good and healthy diet plan while maintaining variety in meals.

10/25/2024 5
 4

10/25/2024 6
End Users: Who Benefits from Our System?
Our recommendation system is designed to cater to a range of users, with the primary user personas
being:
1. Individuals with specific dietary needs
Users managing medical conditions such as diabetes, hypertension, or allergies.

2. Health-conscious users
People aiming for weight loss, muscle gain, or general wellness through personalized diet plans.

3. Nutritionists and Healthcare Providers

Professionals who can use the system to recommend recipes can be used for a diet plan tailored to
patient needs.

Each user group benefits differently, with the system providing them with highly personalized meal
recommendations based on their input.

10/25/2024 7
Key Features to Address User Needs:
1. Personalized Diet Plans
Users receive customized meal plans based on their medical history and dietary goals.
2. Recipe Recommendations Based on Ratings
The system uses collaborative filtering to suggest recipes that match users' tastes and
preferences.
3. Easy Data Entry via Streamlit Interface
Users can quickly input personal information (e.g., allergies, preferences) through a simple,
interactive web interface.
4. Continuous Learning from User Feedback
As users rate recipes, the system refines its recommendations over time using SVD and KNN-
based collaborative filtering.
5. Diverse Recipe Suggestions
The app ensures variety by recommending recipes not only from the user's preferences but
also based on similar users' choices.

10/25/2024 8
How These Features Solve User
Problems:
1. For Individuals with Health Conditions:
The system ensures that dietary restrictions (e.g., no sugar for diabetic users) are respected,
providing safe and healthy meal options.
2. For Health-Conscious Users:
Offers balanced diet plans aligned with fitness goals, with variety to avoid repetition.
3. For Nutritionists:
Saves time by generating instant diet plans that are specific to a patient’s condition and
preferences.

‘This personalized approach ensures every user gets value, whether they are looking for a quick
recipe suggestion tailored to their medical needs’

10/25/2024 9
 8

10/25/2024 10
 Recommendation System
Architecture
Before diving into the deliverables, let me first explain the architecture
of our recommendation system and the types of collaborative filtering
we’ve used

10/25/2024 11
A recommendation system works by analyzing user behavior like ratings, clicks, or
purchase history and providing personalized suggestions. There are two main types of
recommendation systems:

•Content-based filtering – Recommends items based on the features of the item itself.
Example: If you liked a spicy recipe, it will recommend other spicy recipes.

•Collaborative filtering – Looks at how similar users interact with items and recommends
Items accordingly This is what we’ve focused on in our project.

10/25/2024 12
In our system, we implemented three types of
collaborative filtering techniques:

• Collaborative Filtering the system is expected rating for the user on the recipe
using SVD

• Collaborative Filtering The system will recommend similar recipes to users based on their past
ratings

• user-based collaborative filtering. This method recommends recipes to users based on their
rating patterns in comparison with similar users
using KNNBasic.

10/25/2024 13
Matrix Factorization with SVD (Singular Value
Decomposition):
This method breaks down a large user-recipe rating matrix into smaller pieces,
predicting how a user would rate an unrated recipe.
It helps fill in the gaps in the data by making personalized predictions.

Mathematically:

SVD decomposes a matrix R into three

matrices: R≈U⋅Σ⋅VTR
where U represents users, V represents recipes, and Σ is a diagonal matrix
capturing the importance of the latent features.

10/25/2024 14
Item-based Collaborative Filtering:
This method looks at the recipes a user has rated highly
and finds other similar recipes based on these
interactions.

10/25/2024 15
User-based Collaborative Filtering
using KNNBasic:

•This technique compares a user’s ratings to other users

with similar rating patterns. It finds the closest users using a
mathematical method called K-Nearest Neighbors (KNN)

•.In KNNBasic, we calculate the similarity between two users

based on how similarly they rate items, using formulas like
cosine similarity.

10/25/2024 16
 These techniques work together to create personalized
recommendations for users, predicting which diet plans
and recipes will best match their preferences and medical
history

10/25/2024 17
User Manual:
Explains how to interact with the system:

1. Input medical history and preferences on the homepage.

2. Receive personalized diet plans and recipe recommendations immediately.

3. Rate recipes to improve future recommendations."

10/25/2024 18
Final Products
Our final product deliverables include:
Working WebApp:
The app is built using Streamlit, which allows us to provide a simple and interactive web
interface.
Users can:
1. Enter their medical history and preferences.
2. Receive customized diet and recipe recommendations.
3. Rate recipes to refine recommendations over time.
Source Code Repository on GitHub:
Our codebase includes:
1. The SVD and KNN models used for collaborative filtering.
2. The GAN model that generated user profiles.
3. Streamlit frontend code, along with deployment instructions.

10/25/2024 19
10/25/2024 20
10/25/2024 21
10/25/2024 22
10/25/2024 23