Amazon Metadata Streaming Data Pipeline and Itemset Mining

This repository houses an implementation of finding similar items utilising A-Priori and PCY Algorithms on Apache Kafka.

Using a 12GB .json file as a sample of the 100+GB Amazon_Reviews Dataset, it was developed as part of an assignment for the course Fundamentals of Big Data Analytics (DS2004).

The project leverages:

1. Apache Kafka for robust real-time data streaming.
2. (Optional) Use of Azure VMs and Blobs, providing a scalable solution for large datasets.

Repository Structure:

└── preprocessing.py            # Script for preprocessing data locally
└── sampling.py                 # Script used to randomly sample the original 100+GB into 15 GB.
├── preprocessing_for_azure.py  # Script for preprocessing and loading data to Azure Blob Storage
├── blob_to_kafka_producer.py  # Script for streaming data from Azure Blob to Kafka
├── consumer1.py                 # Kafka consumer implementing the Apriori algorithm
├── consumer2.py                 # Kafka consumer implementing the PCY algorithm
└── consumer3.py                # Kafka consumer for anomaly detection 
└── producer_for_1_2.py          # Kafka producer for Apriori and PCY consumers
└── producer_for_3.py            # Kafka producer for Anomaly detection consumer

Setup Instructions

1. Data Preparation

The first step is to download and preprocess the Amazon Metadata dataset.

Download the dataset from the provided Amazon link. Use EITHER of:
    └── Preprocessing_for_azure.py if using Azure,
    └── Preprocessing.py if not.

Upload the preprocessed data to Azure Blob Storage (set blob and connection string in the script) (If not using Azure, skip this step).

The original dataset's size necessitated sampling for efficient analysis. We ensured a good mix of metadata for our analysis.

2. Streaming Pipeline

Next up is setting up Kafka (and optionally Azure Blob Storage):

Deploy Apache Kafka. Ensure Kafka brokers are accessible.

Modify azure_blob_data.py with your Azure Blob Storage connection details and Kafka bootstrap servers.

Run blob_to_kafka_producer.py to stream data from Azure Blob Storage to Kafka.

3. Consumer Applications

Then deploy the consumer scripts:

consumer1.py: Consumes data for frequent itemset mining using Apriori. Adjust Kafka topic and MongoDB details.

consumer2.py: Similar setup as Apriori, but implements the PCY algorithm.

consumer3.py: Implements anomaly detection. Configure for the relevant Kafka topic.

Technologies and Challenges:

Used Technologies:

Azure Blob Storage: For storing and managing large-scale dataset preprocessing.

Apache Kafka: Utilized for robust real-time data streaming.

Python: Scripting language for data processing and mining algorithms.

MongoDB (optional): Recommended for storing consumer application outputs for persistent analysis

Streaming Challenges and Solutions:

Sliding Window Approach Approximation Techniques

Why This Implementation with Kafka and Sliding Window Approach?

This project leverages Apache Kafka and a sliding window approach for real-time data processing due to several key advantages:

Scalability of Kafka:

Kafka's distributed architecture allows for horizontal scaling by adding more nodes to the cluster. This ensures the system can handle ever-increasing data volumes in e-commerce scenarios without performance degradation.

Real-time Processing with Sliding Window:

Traditional batch processing wouldn't be suitable for real-time analytics. The sliding window approach, implemented within Kafka consumers, enables processing data chunks (windows) as they arrive in the stream. This provides near real-time insights without waiting for the entire dataset.

Low Latency with Kafka:

Kafka's high throughput and low latency are crucial for e-commerce applications. With minimal delays in data processing, businesses can gain quicker insights into customer behavior and product trends, allowing for faster decision-making.

While Azure Blob Storage provides excellent cloud storage for the preprocessed data, and Azure VMs allow for easier clustering, it's Kafka that facilitates the real-time processing aspects crucial for this assignment's goals. The combination of Kafka's streaming capabilities and the sliding window approach within consumers unlocks the power of real-time analytics for e-commerce data.

Team:

• Manal Aamir: GitHub
• Mohammad Malik: GitHub
• Aqsa Fayaz: GitHub