Code for applying natural language processing methods on whatsapp conversations

SoAn (Social Analysis) can be used to extract word frequency, word clouds, TF-IDF, sentiment analysis, and more from whatsapp conversations. The main application was initially used to analyze the messages between my wife and me, but I extended so that it can be used for your own messages.

1. Instructions

2. Output

   a. General Plots

   b. TF-IDF

   c. Emoji

   d. Sentiment

   e. Word Clouds

   f. Topic Modeling

Back to ToC

There are several steps for using this repository:

• Download or fork this repository
• Install the requirements with pip install -r requirements.txt
• Save your whatsapp.txt file in the data folder
  • To download your whatsapp messages simply go open your whatsapp, go to a conversation, click the three vertical dots and export the file
• Finally, from the commandline, run the following:
  • python soan.py --file whatsapp.txt --language english
• The results will be saved as images and text files in the results folder

In the notebooks folder, you will also find the soan.ipynb where you can run individual pieces of the code.

Back to ToC

There are 4 types of plots to be generated:

• Messages over time

• Active days of each user

  • Spider
  • Histogram

• Active hours of each user

• Calendar plot

• There are 2 types of stats that are generated:

  • General statistics (text frequency, etc.)
  • Timing

Below are some examples of the plots above:

Below are some examples of the text generated:

##########################
Number of Messages
##########################

4444 Her
3266 Me

#########################
Messages per hour
#########################

Her: 0.1259887165820883
Me: 0.09259206758710628

Using a class-based TF-IDF, I extract the most important words per person and plot them using a horizontal barchart with a mask as image. I created a horizontal bar chart with two bars stacked on top of each other both plotted on a background image. I started with a background image and plotted the actual values on the left and made it fully transparent with a white border to separate the bars. Then, on top of that I plotted which bars so that the right part of the image would get removed.

NOTE: In the notebook, you will see more instructions on how to use your own image.

These analysis are based on the Emojis used in each message. Below you can find the following:

• Unique Emoji per user
• Commonly used Emoji per user

The sentiment from each sentence in the messages is extract per user using Vader and visualized as follows:

For each user, a word cloud will be made based on frequent and important words. Stopwords are removed if you have supplied the language:

For each user, the most frequent topics using LDA and NMF are modeled and saved a .txt file:

Me

Topics in nmf model:
Topic #0: ga boodschappen nodig lieverd halen uurtje half
Topic #1: thuis wel goed haha lekker we morgen
Topic #2: lieverd dank hey fijn allerliefste plezier verwacht
Topic #3: gezellig jeey super jeeeey erg hartstikke samen
Topic #4: love you most more schattie much very

Visualizations Wife
Below, you will find an overview of the visualizations I made for my wife, in part using this package: