Spaces:
Sleeping
Sleeping
File size: 2,830 Bytes
5a89edd 6bba77f 476f798 6bba77f 476f798 6bba77f 476f798 6bba77f 476f798 6bba77f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 | ---
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]
> 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]
> 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]
> 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]
> 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. |