Skip to main content
  1. Home
  2. Apps

Repository: search-analytics

Extract data from google analytics and process it for use by site search

GitHub
search-analytics
Ownership
#user-experience-measurement-govuk
Hosting
N/A
Category
Services
Links

README

This code extracts analytics data from google analytics and processes it such that it can be used by the site search on gov.uk for improving search result quality.

Installation

For developing and testing locally, it is recommended that you use the same Python version as the one set in nightly-run.sh currently this is 3.8.13.

To achieve this - if you do not already have this version available - is by using pyenv. pyenv should be familiar to anyone who has used rbenv - in fact it is a clone of rbenv so shares many of the same commands and concepts.

Once pyenv is installed and the above version is installed (pyenv install 3.8.13), you should be able to cd into the root of the project where pyenv will read the .python-version file and load the correct version.

From there you can issue the following commands to load a Python virtual environment and install the dependencies. Run deactivate at anytime to exit the Python virtual environment.

$ python -m venv venv
$ source venv/bin/activate
$ pip install -r requirements.txt

Testing, coverage and linting

$ python -m unittest discover
$ coverage run -m unittest discover
$ coverage report -m
$ pylint --recursive=y ./analytics_fetcher

Authentication with Google Analytics

To make the data-fetch from Google Analytics work, you’ll need to fetch a client_secrets.json file from google containing credentials, and use that to generate a refresh token. This refresh token must then be passed to the script via an environment variable.

Some details on generating these credentials are given in the GA tutorial, but in summary:

  • create (or already have) a google account with access to the google analytics profile for www.gov.uk.

  • create a project in the google developers console

  • For the project, go to the “APIs & auth” section on the dashboard, and ensure that the “Analytics API” is turned on.

  • Go to the “Credentials” section on the dashboard, and click the “Create New Client ID” button, to create a new OAuth 2.0 client ID.

  • Pick the “Installed Application” option, and a type of “other”

  • Download the JSON for the newly created client (using the “Download JSON” button underneath it).

  • Run the following command to generate the refresh token.

    PYTHONPATH=. python scripts/setup_auth.py /path/to/client_secrets.json

    It will display a url which you’ll need to open with a browser that’s signed in to the google account that the client JSON was downloaded from; paste the result into the prompt. The command will output (to stdout) a “GAAUTH” environment variable value which needs to be set when calling the fetching script.

  • Delete the client_secrets.json file after use - it shouldn’t be needed again, and this ensures it doesn’t get leaked (eg, by committing it to git).

  • Run the fetch script (scripts/fetch.py) with environment variables set. See below for details.

  • Don’t commit any of the generated secrets to this git repo! For regular runs from Jenkins, pass the environment variables in from the Jenkins jobs.

Fetching data

Ensure that the virtualenv is activated, and then run:

GAAUTH='...' PYTHONPATH=. python scripts/fetch.py page-traffic.dump 14

(Where GAAUTH is the value obtained from the setup_auth.py script, and the final argument 14 is the number of days to fetch analytics data for.)

This will generate a file called page-traffic.dump, which is in elasticsearch bulk load format, and can be loaded into the search index using the bulk_load script in search-api. This contains information on the amount of traffic each page on GOV.UK got (after some normalisation).

The fetching script fetches data from GA by making requests for each day’s data. It caches the results for each day, so that it doesn’t need to repeat all the requests when run on a subsequent day. By default, the cache is placed in a directory “cache” at the top level of a checkout. The location of the cache can be controlled by passing a path in the CACHE_DIR environment variable. Entries which are older than 30 days will be removed from the cache at the end of each run of the fetch script.

Running popularity update without retrieving GA data

When running the full script on integration and staging it is desirable that we don’t retrieve new data from GA.

This can be achieved with the following command:

./nightly-run.sh SKIP_TRAFFIC_LOAD=true

The dump format

The dump is in Elasticsearch bulk load format. It looks like this:

{"index":{"_type":"page-traffic","_id":"/search/all"}}
{"path_components":["/search","/search/all"],"rank_14":3,"vc_14":3201853,"vf_14":0.029424856324613228}

There are rank_%i, vc_%i, and vf_%i entries for each range of days, though the nightly load script only uses 14-day ranges.

The fields are:

  • path_components: the path and all of its prefixes.
  • rank_%i: the position of that page after sorting by vc_%i descending.
  • vc_%i: the number of page views in the day range.
  • vf_%i: the vc_%i of the page divided by the sum of the vc_%i values for all pages.