Hugging Face's logo

Spaces:
galcomis
/
Bite_Wise_Final
Sleeping

App Files Community
Bite_Wise_Final / README.md
galcomis's picture
galcomis
Update README.md
476f798 verified
preview code
|
raw
history blame contribute delete
2.83 kB

A newer version of the Gradio SDK is available: 6.9.0

Upgrade
metadata 
title: Bite Wise Final
emoji: 📚
colorFrom: purple
colorTo: green
sdk: gradio
sdk_version: 6.3.0
app_file: app.py
pinned: false
license: apache-2.0

🥗 BiteWise: Discover Your Culinary Twin

BiteWise is a boutique AI recommendation engine. It bridges the gap between personal lifestyle and culinary cravings, utilizing a massive dataset of 10,000 curated text records and 1,000 high-quality images.

📊 Exploratory Data Analysis (EDA)

Our EDA process ensured data integrity and revealed the behavioral patterns of our community.

1. Quality Control: Rating Distribution

![Ratings Distribution]

image

We performed outlier removal to ensure recommendations are based on authentic, high-quality reviews. The distribution shows a reliable baseline for our ranking algorithm.

2. Community Demographics: Age Range

![Age Distribution]

image

The "Culinary Twin" logic accounts for age diversity. By analyzing this distribution, we ensure that vibes and restaurant contexts match the user's life stage.

3. Personal Aesthetics: Fashion Styles

![Fashion Styles]

image

This visualization highlights the diversity of our users' styles. Identifying these clusters is key to our 35% persona-based weighting.

4. The Core Logic: Style vs. Food Preference

![Correlation Heatmap]

image

This Behavioral Heatmap is the "Brain" of BiteWise. It proves the strong correlation between lifestyle (Fashion) and taste (Food Vibe), enabling high-precision hybrid matching.

🧠 The AI Engine: Semantic Hybrid Search

BiteWise doesn't just look for keywords; it understands culinary context through vector embeddings.

The Pipeline:

  1. Model: We utilize the all-MiniLM-L6-v2 Sentence Transformer from Hugging Face.
  2. Embeddings: Text descriptions and user personas are converted into 384-dimensional vectors.
  3. Hybrid Scoring: - 65% Weight: Semantic match between user craving and dish description.
    • 35% Weight: Persona similarity (Age, City, Style) between the user and the "Culinary Twin."
  4. Distance Calculation: We use Cosine Similarity to find the top 3 closest matches in the vector space.

📂 Dataset & Tech Stack

  • Data: 10,000 synthetic records & 1,000 indexed images.
  • Framework: Gradio
  • Libraries: Sentence-Transformers, Scikit-learn, Pandas, Seaborn.