CS6350
Big data Management Analytics and Management
Fall 2021
Homework 3
Submission Deadline: November 11th, 11:59 p.m.

Spark Streaming and Visualization

Q1.
You are required to implement the following framework using Apache Spark
Streaming, Kafka (optional), Elastic, and Kibana. The framework performs
SENTIMENT analysis of particular hash tags in twitter data in real-time. For
example, we want to do the sentiment analysis for all the tweets for #trump,
#coronavirus. Note that if you implement this framework with Scala, there is no
need for Kafka and you can connect to twitter via the internal API. But if you want
to implement it with Python, Kafka is required. Be careful about the Scala version
compatibility.
Figure: Sentiment analysis framework

The above framework has the following components:

1. Scrapper (for python, but Scala needs to produce same result)
The scrapper will collect all tweets and sends them to Kafka for analytics. The
scraper will be a standalone program written in PYTHON and should perform the
followings:
a. Collecting tweets in real-time with particular hash tags. For example, we
will collect all tweets with #blacklivesmatter.
b. After filtering, we will send them to Kafka in case if you use Python.
c. You should use Kafka API (producer) in your program
(https://kafka.apache.org/090/documentation.html#producerapi)
d. Your scrapper program will run infinitely and should take hash tag as input
parameter while running.

2. Kafka (for Python)
You need to install Kafka and run Kafka Server with Zookeeper. You should create
a dedicated channel/topic for data transport

3. Spark Streaming
In Spark Streaming, you need to create a Kafka consumer (for python, shown in
the class for streaming) and periodically collect filtered tweets (required for both
Scala and python) from scrapper. For each hash tag, perform sentiment analysis
using Sentiment Analyzing tool (discussed below).

3. Sentiment Analyzer
Sentiment Analysis is the process of determining whether a piece of writing is
positive, negative or neutral. It's also known as opinion mining, deriving the
opinion or attitude of a speaker. For example, the following tweets taken from
Twitter are shown along with their sentiment.

“RT @jeremycorbyn: It is shameful the UK Government won’t condemn Trump.
Now is the time to speak up for justice and equality. #BlackLives”- has positive
sentiment.

“RT @larryelder: How many unarmed blacks were killed by cops last year? 9. How
many unarmed whites were killed by cops last year? 19. More” - has negative
sentiment.

You can use any third-party sentiment analyzer like Stanford CoreNLP (Scala),
NLTK(python) for sentiment analyzing. For example, you can add Stanford
CoreNLP as an external library using SBT/Maven in your Scala project. In python
you can import NLTK by installing it using pip.

4. Elasticsearch
You need to install the Elasticsearch and run it to store the tweets and their
sentiment information for further visualization purpose.
You can point http://localhost:9200 to check if it’s running.
For further information, you can refer:
https://www.elastic.co/guide/en/elasticsearch/reference/current/getting-
started.html

5. Kibana
Kibana is a visualization tool that can explore the data stored in Elasticsearch. In
this assignment, instead of directly output the result, you are supposed to use the
visualization tool to show your tweets sentiment classification result in a real-time
manner.
Please see the documentation for more information:
https://www.elastic.co/guide/en/kibana/current/getting-started.html

Q2
Clustering
This is a two-part question where you are expected to perform incremental K-
means clustering on Twitter data by collecting relevant tweets after every 30
second interval. You are also expected to show the changes in the clusters via a
scatter plot that updates in real-time.

a) For the first part use the scrapper from the first question, extract tweets
having the hashtag #BLM and perform K-Means clustering with K=3. Then
plot the points and show which cluster they belong to.
b) Then after every 30 second interval, extract tweets for the same hashtag
for an indefinite amount of time. For each interval, incrementally update
the K-Means clusters and show how they change due to the addition of the
new set of clusters.

What to submit:
1. Python/Scala code
2. Screenshots of your visualization charts

欢迎咨询51作业君