Mobile Recommendation System

Overview

This project implements a mobile recommendation system using DistilBERT for semantic understanding of user queries. The system takes a text description as input (e.g., "Oppo phone with 8GB RAM, 128GB ROM, 5000mAh battery") and recommends the most relevant mobile phones based on embeddings and cosine similarity.

Features

• Uses DistilBERT (distilbert-base-uncased) for text embeddings.
• Extracts numerical values like RAM, Storage, Battery, and Price from user queries.
• Filters recommendations based on extracted specifications.
• Computes cosine similarity between the user query and stored mobile embeddings.
• Returns the top K recommended mobile phones.

Dataset

The dataset contains:

• MobileName: Name of the mobile phone.
• Description: Text description including specifications.
• Price: Price of the phone.
• Embedding: DistilBERT-generated vector representation of the description.

Model Used

• Transformer Model: distilbert-base-uncased
• Library: transformers
• Embedding Generation:
  • Tokenization using DistilBertTokenizer.from_pretrained("distilbert-base-uncased")
  • Feature extraction using DistilBertModel.from_pretrained("distilbert-base-uncased")

Installation

To run the project, install the required dependencies: pip install torch transformers pandas scikit-learn numpy

Execution

To run just run the command in VS code terminal streamlit run app.py

Conclusion

This project provides an efficient recommendation system using transformers to understand natural language queries and match them with mobile phone specifications.