Python scripts to generate histograms and shaped word clouds from chat logs. Currently supports Google Hangouts and Facebook Messenger log formats.

Use Google Takeout: https://accounts.google.com/ServiceLogin?service=backup

Request an archive containing your chat logs. Extract the file called Hangouts.json and put it in the raw folder of ChatShape.

Google switched from "Google Talk" to "Google Hangouts" mid-2013. You will only get your Hangouts logs.

1. Go to "Settings" from the top right drop-down menu.
2. Click on "Download a copy of your Facebook data" at the bottom of the General section.
3. Click on "Start My Archive". It will take Facebook a while to generate it.
4. Once it is done download and extract the archive, then move the file messages.htm in the raw folder of ChatShape.

You will need to specify your own name to the parsers. Use the exact same format as you have on Messenger or Hangouts.

• Google Hangouts: python parse_hangouts.py -ownName "John Doe"
• Facebook Messenger: python parse_messenger.py -ownName "John Doe"

This will generate pickle files in the data folder. For more options use the -h argument on the parsers.

Plot all messages with:

python analyse.py -data data/*

You can then filter the messages as needed:

  -filterConversation FILTERCONVERSATION
                        only keep messages sent in a conversation with this
                        sender
  -filterSender FILTERSENDER
                        only keep messages sent by this sender
  -removeSender REMOVESENDER
                        remove messages sent by this sender

Eg to see all the messages sent between you and Jane Doe:

python analyse.py -data data/* -filterConversation "Jane Doe"

To see the messages sent to you by the top 10 people with whom you talk the most:

python analyse.py -data data/* -removeSender "Your Name" -n 10

You can also plot the conversation densities using the -plotDensity flag.

You will need a mask file to render the word cloud.

python cloud.py -data data/* -m img/mask_image.jpg

The white bits of the image will be left empty, the rest will be filled with words using the color of the image. See the WordCloud library documentation for more information.

You can filter which messages to use using the same flags as with histograms.

• Integrate with gtalk_export to handle Google Talk format.
• Parsers for more chat platforms: WhatsApp? Pidgin? ...
• WordCloud has some issues with non-English languages. For example in French it will transform "ouais" (yep) in "ouai". Fixing this would probably require a pull request on the WordCloud project.
• For now if the phrase "don't worry" appears a lot, "don't" will be removed (as a stopword) and only "worry" will appear in the cloud. This is misleading. It would be interesting to implement an n-gram approach to consider groups of words. This example from WordCloud could be a good starting point.
• Figure out OWN_NAME automatically.
• Handle group chats.

Pull requests welcome!

• Word cloud generated using https://github.com/amueller/word_cloud
• Stopwords from https://github.com/6/stopwords-json
• Code under MIT license