Skip to content
/ twitter-for-bigquery Public
forked from xdevplatform/twitter-for-bigquery

Simplest way to get Tweets into BigQuery. Uses Google Cloud & App Engine, as well as Python and D3.

twitter-for-bigquery.appspot.com

License

Apache-2.0 license
0 stars 36 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

ChrisStevens/twitter-for-bigquery

BranchesTags
 
 

Repository files navigation

Twitter for BigQuery

This sample code will help you streaming Twitter data into BigQuery, and running simple visualizations. This sample also generates the queries you can run directly in the BigQuery interface, or extend for your applications.

Additionally, you can use other public or private datasets in BigQuery to do additional joins and develop other insights/correlations.

Requirements

Setup & Configuration

To work with Google Cloud and BigQuery, follow the below instructions to create a new project, service account and get your PEM file.

  • Go to http://console.developers.google.com

  • Click on "Create Project"

  • Open the project dashboard by clicking on the new project

  • Open "APIs & auth->Credentials"

  • Click on "Create new Client ID", "Service account" and "Create Client ID"

  • Note your Service Account email (Under "EMAIL ADDRESS")

  • Generate and store your P12 key (Or save from auto-download)

  • Convert the P12 key to a PEM file with the following:

    cat key.p12 | openssl pkcs12 -nodes -nocerts -passin pass:notasecret | openssl rsa > key.pem

  • Copy config.template to config

  • Fill out the following fields:

  • Run setup.py to generate appropriate yaml and config files in the image_gnip and image_twitter

Loading Twitter data into BigQuery from your local machine

As a pre-requisite for setting up BigQuery, you need to first set up a billing account. To do so:

  • Go to https://console.developers.google.com/billing and add a credit card
  • Back in your project view, click on the gear icon in the top-right and then "Project billing settings"
  • Ensure your project is associated with a billing account

The enclosed sample includes a simple load.py file to stream Tweets directly into BigQuery.

  • Go to http://console.developers.google.com
  • Go to your project
  • In the left-hand side, click on "Big Data->BigQuery" to open the BigQuery console
  • Click on the down arrow by the project, select "Create new dataset" and enter "twitter"
  • Run python load.py to begin loading data from your local machine

When developing on top of the Twitter platform, you must abide by the Developer Agreement & Policy.

Most notably, you must respect the section entitled "Maintain the Integrity of Twitter's Products", including removing all relevant Content with regard to unfavorites, deletes and other user actions.

Loading Twitter data into BigQuery from Google Compute Engine

To help simplify your setup, this project is designed to use:

The Dockerfile describes the required libraries and packaging for the container. The below runs through the steps to create your own container and deploy it to Google Compute Engine.

# start docker locally
boot2docker start
$(boot2docker shellinit)

# build and run docker image locally
docker build -t gcr.io/twitter_for_bigquery/image .
docker run -i -t gcr.io/twitter_for_bigquery/image

# push to Google Cloud container registry
gcloud preview docker push gcr.io/twitter_for_bigquery/image

# create and instance with docker container
gcloud compute instances create examplecontainervm01 \
    --image container-vm \
    --metadata-from-file google-container-manifest=./container.yaml \
    --zone us-central1-b \
    --machine-type n1-highcpu-2
    
# log into the new instance
gcloud compute instances list
gcloud compute --project "twitter-for-bigquery" ssh --zone "us-central1-b" "examplecontainervm01" 

# pull the container and run it in docker 
sudo docker pull gcr.io/twitter_for_bigquery/image
sudo docker run -d gcr.io/twitter_for_bigquery/image

# view logs to confirm its running
sudo -s
sudo docker ps
sudo docker logs --follow=true 5d

More notes for Docker + Google Cloud:

Running the app

Running locally

From the command line, you can use dev_appserver.py to run your local server. You'll need to specify your service account and private key file on the command line, as such:

dev_appserver.py . --appidentity_email_address="[email protected]" --appidentity_private_key_path=/PATH/TO/key.pem

Once this is complete, open your browser to http://localhost:8080.

Deploying on Google App Engine

To run in Google App Engine, do the following:

  • Update app.yaml with the project name pointing to your project.
  • Open the GAE Launcher.
  • Click on "File->New Application".
  • Specify the application ID (twitter-for-bigquery) and application directory (path where twitter-for-bigquery project exists).
  • Click "Save".
  • Select the Application in the list and click on "Edit->Application Settings".

  • In the "Extra Flags" section, add the command line flags, as above:

    --appidentity_email_address="[email protected]" --appidentity_private_key_path=/PATH_TO/key.pem

To confirm the deploy worked, you can do the following to view the logs:

Querying and loading large sets of tweets onto BigQuery

If you need large amounts of past tweets loaded onto BigQuery, you will need to use Gnip's Historical Power Track. The best way to load large amounts of tweets is:

  • Use the Gnip Python Historical Utilities library to run an async PowerTrack job and download the data.
  • Run the included batch.py file to process each gzip file and load onto BigQuery

When running the above processing, choose an environment that is optimized for network performance, as you may be downloading multiple GB of files onto your server and then onto BigQuery.

The schema

Schema

The load.py file takes tweets and loads them one-by-one into BigQuery. Some basic scrubbing of the data is done to simplify the dataset. (For more information, view the Utils.scrub() function.) Additionally, JSON files are provided in /schema as samples of the data formats from Gnip/Twitter and stored into BigQuery.

Sample queries

To help you get started, below are some sample queries.

Text search

Querying for tweets contain a specific word or phrase.

SELECT text FROM [twitter.tweets] WHERE text CONTAINS ' something ' LIMIT 10
#Hashtag search

Searching for specific hashtags.

SELECT entities.hashtags.text, HOUR(TIMESTAMP(created_at)) AS create_hour, count(*) as count FROM [twitter.tweets] WHERE LOWER(entities.hashtags.text) in ('John', 'Paul', 'George', 'Ringo') GROUP by create_hour, entities.hashtags.text ORDER BY entities.hashtags.text ASC, create_hour ASC
Tweet source

Listing the most popular Twitter applications.

SELECT source, count(*) as count FROM [twitter.tweets] GROUP by source ORDER BY count DESC LIMIT 1000
Media/URLs shared

Finding the most popular content shared on Twitter.

SELECT text, entities.urls.url FROM [twitter.tweets] WHERE entities.urls.url IS NOT NULL LIMIT 10
User activity

Users that tweet the most.

SELECT user.screen_name, count(*) as count FROM [twitter.tweets] GROUP BY user.screen_name ORDER BY count DESC LIMIT 10

To learn more about querying, go to [https://cloud.google.com/bigquery/query-reference]https://cloud.google.com/bigquery/query-reference)

Going further

Using BigQuery allows you to combine Twitter data with other public sources of information. Here are some ideas to inspire your next project:

  • Perform and store sentiment analysis on tweet text for worldwide sentiment
  • Cross reference Twitter data to other public data sets

You can also visit http://demo.redash.io/ to perform queries and visualizations against publicly available data sources.

FAQ

When deploying to AppEngine, I'm getting the error "This application does not exist (app_id=u'twitter-for-bigquery')"

You will want to create your own app_id in app.yaml. If that does not work, then per this thread (http://stackoverflow.com/questions/10407955/google-app-engine-this-application-does-not-exist), try the following:

`rm .appcfg_oauth2_tokens`

My TaskQueue entries die unexpectedly/only run for 10 minutes/get a DeadlineExceededError.

The default Google AppEngine TaskQueue (named 'default') has a limit of 10 minutes for any task. To run a task for longer, you need to set up a custom task queue and a backend server. The instructions are above, but the basics include:

  • Ensure the queues.xml file (which defines a new queue named 'backfill') is uploaded to AppEngine.
  • Ensure a background app is created using the appcfg.py update app.yaml backfill.yaml command to start both the main app and the background app.

I am getting 'Process terminated due to exceeding quotas.' errors in my log console/'This application is temporarily over its serving quota. Please try again later.' when accessing my backend server.

Google AppEngine has usage quotas to regulate billing and usage. You can read about the quotas for various products here:

https://cloud.google.com/appengine/docs/quotas#When_a_Resource_is_Depleted

To increase quota limits, you can go into Compute->App Engine->Settings and edit your daily budget to allow for increased usage.

https://console.developers.google.com/project/YOUR_PROJECT_NAME/appengine/settings

Additional reading

The following documents serve as additional information on streaming data from Twitter and working with BigQuery.

Credits

The following developers and bloggers have aided greatly in the development of this source. I'm appreciative of contributions and knowledge sharing.

TODO

  • One Pager
  • FAQ
  • Easier to deploy full stack
    • environment settings
    • container deploy script
  • Figure out location, specifically don't use Utils.scrub()
  • Admin save/config page + deploy of service?

About

Simplest way to get Tweets into BigQuery. Uses Google Cloud & App Engine, as well as Python and D3.

twitter-for-bigquery.appspot.com

Resources

Readme

License

Apache-2.0 license
Activity

Stars

0 stars

Watchers

2 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Languages

  • Python 77.3%
  • JavaScript 20.9%
  • HTML 1.3%
  • CSS 0.5%