---

Spotify Review Chatbot

This repository provides a Spotify review chatbot, built to analyze and answer user inquiries based on Spotify's Google Play Store reviews.

Dataset: The full dataset for this project is available on Kaggle: 3.4 Million Spotify Google Store Reviews.

Requirements

• Docker
• Docker Compose
• Poetry (Python 3.12.3)

---

Getting Started

Install Dependencies

To install project dependencies, run:

poetry install

---

Knowledge Base

The knowledge base is powered by Qdrant as the vector database, which stores the semantic embeddings of review texts.

• Vector Database:

  Qdrant serves as the vector database for this chatbot. To deploy Qdrant, run:

  docker-compose -f docker-compose-qdrant.yml up

• Embedding Model: We use TinyBERT for semantic representation of review texts. TinyBERT is served with Huggingface’s Optimum ONNX and transformed into vector representations using Transformers' feature-extraction pipeline.

  To deploy TinyBERT:

  1. Build Optimum ONNX Serving Image:

     cd tinybert/serving
     docker build -t optimum-onnx-serving-cpu:0.1.2 .

  2. Update model.onnx Path in Docker Compose Volumes

     Example:

     - /home/miftah/Downloads/job_application/mekari/review_bot/tinybert:/app/models

  3. Start the TinyBERT Docker Compose:

     docker-compose up

Once Qdrant and TinyBERT are deployed successfully, load the dataset:

1. Download the dataset: SPOTIFY_REVIEWS.csv.

2. Update fname in qdrant_scripts/load_data.py to point to the dataset path.

3. Run the data loading script:

   cd qdrant_scripts
   poetry run python load_data.py

---

Chatbot

The chatbot uses LangGraph as the main library, integrating LangChain to manage AzureOpenAI and ChatPromptTemplate for seamless dialog flow.

Run Chatbot Interactively

To run the chatbot in terminal mode:

./scripts/run_rag.sh

Deploy Chatbot API

To deploy the chatbot API:

./scripts/run_api.sh

Deploy Streamlit UI

To deploy the UI (Chatbot API has to be deployed first):

./scripts/run_ui.sh

---

API Endpoints

The following endpoints are available in this chatbot:

1. POST /ask

   Description: Responds to user inquiries related to Spotify reviews.

   Input: JSON object with user_input (text).

   Example Input:

   {
     "user_input": "In comparison to our application, which music streaming platform are users most likely to compare ours with?"
   }

   Example Output:

   {
     "response": "Users are most likely to compare Spotify with Pandora and Google Play Music, as these platforms are frequently mentioned in the reviews.",
     "score": 0.99
    }

2. PUT /review/<review_id>

   Description: Inserts a review item into the Qdrant collection.

   Input: JSON dictionary with review details.

   Example Input:

   {
     "_id": 19206,
     "review_id": "14a011a8-7544-47b4-8480-c502af0ac26f",
     "pseudo_author_id": "152618553977019693742",
     "author_name": "A Google user",
     "review_text": "Use it every day",
     "review_rating": 5,
     "review_likes": 1,
     "author_app_version": "1.1.0.91",
     "review_timestamp": "2014-05-27 14:21:48"
   }

   Example Output:

   {
     "status": "item id 19206 is inserted into spotify_review"
   }

3. DELETE /review/<item_id>

   Description: Deletes a review item from the Qdrant collection based on the item ID.

   Example Output:

   {
     "status": "item id 19206 is deleted from spotify_review"
   }

4. GET /collections

   Description: Lists all available collections in Qdrant.

   Example Output:

   {
     "result": [
       {
         "collection_name": "spotify_review",
         "documents_count": 8443
       }
     ]
   }

5. GET /healthcheck

   Description: Verifies if the API is running.

   Example Output:

   {
     "status": "ok"
   }

To test these endpoints, open Swagger UI: http://localhost:<API_PORT>/docs in your browser.

---