Collage of images

A year in review-Deep learning (self)

Prologue

Time is a vague concept designed by us to make sense of our lives. Every day is new and every moment is different. We would not generally sit back and reflect upon a whole chunk of period in time every single day or during every little moment. However the predominately accepted faith is to do it as we approach the end of an year. Looking at the scale of this widely accepted tradition, I feel like participating in this cult

Introduction

I have spent a good amount of time in my career analyzing other people’s behavior on the internet. This made me wonder how I would look if I applied those principles on myself. At first glance I learnt that  I haven’t used too many words in my posts. This has been a year of expression through pictures and videos. It reminds me of the post that I had shared about 2 years back on how the world was shifting towards visual forms of expressions on the internet.

This was conceivably the year when I could fully realize that images/videos let me express better. If we think about it, it makes more sense. One picture is indeed worth a thousand words!

Looking at my posts over the internet, I could see that I have posted 63 images/videos compared to 57 posts which had text. Out of these 57 posts, about 48 of them had enough text which expressed how I felt at that point in time. For example a post saying- “Taking a good look”  does not really say much with words compared to the photo that it captioned.

This brings us to the following question. Can we learn what a photo has to say? Natural Language Processing techniques have been able to design algorithms that can not only understand what topic a piece of text is talking about, but also understand the sentiments behind them. For example we can apply modern NLP techniques to learn that the following post by me is about car and expresses negative sentiment. Can we do the same with images?

Ahh well, we live in interesting times(I have always wanted to say this!). Computer Vision Technologies are not very hard to get these days. We could use those technologies to learn that the following picture shared by me has a child who is smiling.

Isn’t that magical? Being able to create intelligence which can imitate human intellect  is a good feeling!

Skip the next paragraph if you are not interested in the technicality

And thus, I got the idea to apply this sort of intelligence and see what algorithms have to say about my year. You can learn about some ideas on how to apply NLP techniques to your text or extract features from your images in these ebooks- Natural Language Processing with Python, Vision with Python. The internet is also filled with a lot of API’s which you can utilize to apply NLP techniques. In my case, I have used TextRazor’s API s to classify my texts. Though NLP techniques have been widely researched and applied, Computer Vision still has a lot of ground to cover. In my case, I have used Google’s Vision APIs to interpret my images/videos. Here is an interesting blog post Comparing top 5 Computer Vision APIs

The problem

At this point, it is worth it to look at the problem again. The process of understanding human nature by looking at their internet usage over a year means being able to go through all the content that someone has created and coming up with meaningful summaries. I have tried to do this for myself and kept the scope of this analysis to content created by me only on social media. Below is a snapshot of what I had to analyze.  Due to privacy concerns, I have tried to keep this analysis within content (mostly) created by me and not include things that others have posted about me.

Collage of images
Images/videos uploaded on Facebook and Instagram
My 2016 Facebook timeline
My 2016 Facebook timeline

You could assume a lot of things about me by looking at the above media. And I can bet that you will have a new thing to say every time you to try to look closely. An average human could process all these information and draw meaningful conclusion in about a few minutes(This is my hypothesis!). Want to test? Try to come up with a few words describing my actions and preferences based on the above.

thinking

Once the entire process was defined,  the Artificial Intelligence took only 3 seconds to condense all the information into meaningful words which could qualify my tastes…….

What did I learn ? 
I will not say much about the below cluster of words. I will simply drop it here and let others interpret

Top words summarizing my social media content. Size is relative to its importance
Top words summarizing my social media content. Size is relative to its number of occurrences

Irrespective of the daily insecurities and anxieties of our lives, and of all the good and bad ways the world has shaped in 2016, I seem to have shared exciting moments, and talk about things that truly matter to me.

How was your 2016?

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s