OS: Ubuntu 14.04 Base Image from VCL.

####Project Setup:

• Because of the large number of dependencies, the easiest way to setup this project is to load a fresh Ubuntu 14.04 Base image from on VCL (Ncsu virtual computing Lab) and use setup script provided in the project folder. Note: The setup script is tested with clean Ubuntu enviornment without any packages pre installed. Running the script in your local machine may produce some unexpected results.

• Once the vcl instance is ready, download the tweet_analysis/ folder inside $HOME.

• Run sudo sh setup.sh inside project folder. This will install all the required programs for running the project.

• Now we need to start storm topology and submit the job. Run sparse run -t 600 inside project folder. This will automatically start storm instance and submit the topology to storm. -t option is used for specifying the run time for topology. So the above command will ensure that topology runs for 600 seconds. If the complete data analysis is required, we can set the running time around t >3000 seconds.

• After this, In a separate tab, start the live dashboard by running

  1. cd dashboard/
  2. sudo npm install
  3. node app.js 
  

• The live dashboard can be seen at http://<VCL instance ip>:3000

####List of project dependencies

• JDK 7+

• Streamparse: Project Managment package for Apache storm. It enables to write spouts and bolts in Python. It provides single command setup of storm project. -Spouts are located inside src/spouts/ folder. -Bolts are located inside src/bolts/ folder. -Some Extra helper files are located inside src/helper/ folder. -Topologies are defined in form of DAG by clojure file topoogies/tweetAnalysis.clg

• Lein: Used by streamparse package for commnicating with storm.

• Redis: We have used Redis for storing the results generated by storm Topology. Storm writes the aggregated results to Redis. NodeJs reads the result by polling Redis and displays the real time dashboard.

• MongoDb: We collected 600K tweets during semi final match of cricket world cup 2015. We have stored these tweets in MongoDb. The spout used by topology reads tweets from MongoDB.

• NodeJs: For implementing real time dashboard. NodeJs polls redis data base for any updates and displays result in Real Time.

####Storm Topology details:

• MongoDB Spout: Reads data from MongoDb and emits the data in form of stream.
• Clean Up Bolt: Performs basic text cleanup operations on tweet text.
• Split And Filter Bolt: Split the tweet text into words and remove the stop words.
• TopK Bolt: Maintain a CountMin Sketch in order to calculate Top-k words.
• Time Slot Creation Bolt: Performs Binning on time stamp in order to perform time series analysis.
• Location Filter Bolt: Filter the tweets based on country code. In a large proportion of data, country was not available. We try to extract the location using time zone in such cases.
• Time based Tweet Count Bolt: Count number of tweets received in particular time slot bin.
• Location Based tweet Count: Count number of tweets received in particular country.

• Using hash data structure of redis.
• Hash key will be of format time_slot:12 , time_slot:27 ; where 12 and 27 are bin numbers.
• A hash has multiple "fields" which we will use to store summary for particular bin.
  • slot_no: current time slot number
  • t_count: count of tweets received in this particular slot
  • s_pos : positive sentiment count in this bin
  • s_neg : negative sentiment count in this bin
  • s_neu : neutral sentiment count in this bin
  • start_ts : current time slot start timestamp
  • end_ts : current time slot end timestamp

• Using hash data structure of redis.
• Hash key will be of format country:AUS , country:IND ; where AUS and IND are alpha-3 country codes.
• A hash has multiple "fields" which we will use to store summary for particular country's tweets information.
  • t_count: count of tweets received in this particular slot
  • s_pos : positive sentiment count in this bin
  • s_neg : negative sentiment count in this bin
  • s_neu : neutral sentiment count in this bin
  • c_code : current hash country code

• Using set data structure of redis.
• Set name used to store the top-k words is top-k
• A set have a unique list of most trending words

####Individual Installation instructions(In case you want to manually install all the dependencies)

• JDK 7+, which you can install with apt-get, homebrew, or an installler; and

• lein, which you can install from the project’s page

  a) wget https://raw.githubusercontent.com/technomancy/leiningen/stable/bin/lein
  b) sudo mv lein /bin/
  c) chmod a+x /bin/lein
  d) lein -version
  

  It should print Leiningen 2.5.1 on Java 1.7.0_65 OpenJDK 64-Bit Server VM once successfully installed.

• Python 2.7 along with compatible pip installer. If incompatible run following command to upgrade pip.

    sudo apt-get install python-pip
  

• Install Python-dev and Virtual Env

  sudo pip install -y python-dev virtualenv
  

• Install Redis server according according to instruction here

• Install MongoDB according to instruction here

• Clone the project git clone git@github.com:nirmeshKhandelwal/tweet_analysis.git

• Install dependencies of Python

  $ cd tweet_analysis`
  $ sudo pip install -r requirements.txt
  $python2 -m textblob.download_corpora 
  

• Make sure your mongoDB server and Redis server is up and running.

• Run $ sparse run in order to run the topology on local.

##References:

[1] Stream Parse for easy integration on Python with Storm.

[2] [Virtual Env] (https://virtualenv.pypa.io/en/latest/) tool to create isolate environment for python project.

[3] CountMin Sketch Implementation for Python