“You are obsessed with WhatsApp!”

It was the last Sunday of 2017 and also the last day of the year.

“You should stop wasting all your time looking at that silly phone. You are obsessed with WhatsApp”, said my wife.

Her remarks stung me; but there was an element of truth in the same. I had been happily contributing my 2 bits to the 100 million hours spend daily by Indians on WhatsApp (link).

I did not want to start 2018 on a wrong footing with my life partner; so perhaps, it was time to work on a New Year resolution to analyse my WhatsApp activities.

My problem: How could I convince her on what was being discussed on WhatsApp and my contribution?

Maybe, it was time to ask Doctor Google AI to inspect all my 23 WhatsApp groups. These groups were broadly categorised into Social, Business & Politics as shown below.

Table 1: Broad categorisation of WhatsApp groups

Within each of the 3 main categories (Social, Business & Politics), other sub-categories are also shown. In addition, the following 5 groups were created after extracting the specific topical content from the above 23 groups.

Table 2: Additional 5 Specific Topic groups created from above 23 Groups

Whatsapp Text Content Meets Google’s AI Natural Language Analysis Tool

WhatsApp messages consist mainly of 5 types of content — Text messages, Static Images, Dynamic animated GIF’s, videos and document attachments. Since in most WhatsApp chat messages, text messages form the bulk of the content, I used Google’s Artificial Intelligence Machine Learning Natural Language Analysis Software (link) to analyse the same.

As the above Google’s link suggests, it’s software can analyse any text data in diverse ways such as classifying the content as per these categories, determining the entities (people, places, etc.), finding out the sentiments (whether positive or negative as regards these entities) and also the overall sentiment of the content.

Over weekends and nights, I asked Google’s AI to do the hard work of analysing more than seven hundred thousand lines of chat messages.

The results? The following were the major text categories discovered by Google’s AI software in the various Whatsapp Groups — Social (Table 3), Business (Table 4) & Political (Table 5) :

Table 3 — Text Categories discovered by Google’s AI Software for Social Whatsapp Groups
Table 4— Text Categories discovered by Google’s AI Software for Business Whatsapp Groups
Table 5— Text Categories discovered by Google’s AI Software for Political Whatsapp Groups

Google’s AI Natural Language software analyses the chat patterns and predicts a text category (or categories). It also determines the confidence percentage (%) associated with that prediction (indicating how sure or unsure it is).

As can be seen from the above tables, except for the Artificial Intelligence (AI) group (marked in Red), Google’s AI was pretty accurate in determining the types of content categories being discussed for all the remaining focussed topic groups (Healthcare, Entrepreneurship, Information Technology, Cryptocurrencies & Politics).

Perhaps, since the content of the AI group consisted mainly of recent contemporary links on AI, Machine Learning (ML) and Deep Learning (DL) that I had shared with my group members, it is likely that Google’s AI software has not been trained sufficiently enough in this topic area to correctly categorize the same.

Another odd result (also shown in Red color above) was that for the School batchmate’s family member cancer treatment funding group, Google’s AI Analysis reported ‘Arts & Entertainment/Movies’ as a category with a confidence figure of 59%. This category seems unusual and would need to be checked further.

Further Investigation

To further verify that Google AI was indeed predicting the results of focussed groups correctly, I also asked it to analyse the following 5 WhatsApp specific topic groups. The results are in Table 6.

Table 6: Further analysis of the five focussed groups

As can be seen once again from the reported text categories and confidence % figures for the above corresponding WhatsApp groups, Google AI is quite accurate in making the correct predictions. The results are aligned with the topic matter of the groups.

“What has been your contribution to the chats?”

A very valid question asked by my life partner. Was I only playing the role of WhatsApp ‘havaldar’ (Indian traffic police constable)? Forwarding messages from one group to other without adding any value?

It was time to once again ask Google’s AI to analyse only my content across the various WhatsApp groups. It discovered the following text categories and % confidence averaged across all my content shared across all the groups :

Table 7: Google AI Text categorisation for self-contributions for all my Whatsapp groups

“Please have a look at the data above.” I told my wife. As you can see, I have been interacting on a wide range of useful topics with others and not just wasting my time. I hope you are finally happy”.

She saw the above data and looked at me suspiciously. “What is sensitive subjects? Does it mean Sunny Leone?”

“Oh no!” I sighed. “Yes, Google AI has also found Sunny Leone in the chats. But that’s a story for another day.”