Aspire For Her, has more than 500 posts, and 2500 followers.
There are multiple classifier algorithms, the best fit for this particular situation was SVM Classifier, its accuracy rate was around 80%!
The best fit for this particular situation was SVM Classifier, its accuracy rate was around 80%!
non-probabilistic binary linear classifier was used to classify the previously predicted comments as positive and negative.
Now, we can see that, one post seems to be working better than the others. AFH can now recreate similar posts to the best-performing post.
AFH can analyse their posts and understand why these posts received negative comments, and make alterations and improvements
Aspire for Her is a non-profit working to motivate women to enter & persevere in the workforce — turning aspirations to action! And an Organization with such an inspiring and powerful intent should definitely reach the masses. One of the many ways to do that, is through social media. But creating an account and posting content without any insights over how your posts are doing and how many readers its attracting, isn't good enough. But, there is a solution, choose LIKULATOR.
What it does
Likulator is an Machine Learning model which estimates the organization's post performance, calculates its reach and gives you back reviews from the intended demographic. Basically it provides you with the inputs and insights of said content, to help you better judge your next and upcoming posts. If you could easily find the most liked features of your content and positively commented upon posts, to help you better design and manipulate your ongoing projects, wouldn't that be neat? Well, that is exactly what Likulator is all about! Likes and Comments is revolutionizing everything. Its engaging. Its feedback. But most importantly, its the input. But all this data, is not used to its potential. Yes, Likes and Comments are naturally displayed, but what if there were thousands of them, would you be able to analyze then? Aspire For Her, has more than 500 posts, and 2500 followers. Which means it is input from 2500 different sources, leaving their likes and comments on more than 500 different designs. Now, think about the people who saw and didn't like. What if we can use that data. All these different attributes, will give rise to this new entire data set with with the correct machine learning models and algorithms can do wonders in get us the most accurate insights to expand our outreach, which is the goal!
How I built it
LIKULATOR uses a variety of algorithms. The three main algorithms used as of now are, Upper Confidence Bound, Natural Language Processing and Support Vector Machine Classifier. With Upper Confidence Bound, we are trying to find the post with the highest reward rate. Let's say, AFH posted 10 different posts on the same subject matter, whenever a reader or a follower finds the post on his/her feed, they either like it, or scroll through because they are unimpressed. But now we have the all this data, we use Likulator. Likulator's UCB algorithm runs this data, and provides the above seen insight. Now, we can see that, one post seems to be working better than the others. AFH can now recreate similar posts to the best-performing post, increase their engagement rate, and hence get promoted to a variety of demographics. This post trial exemplifies decision-making under uncertainty. For the comments, we use Natural-Language-Processing and SVM training classifier, to see how many positive and negative comments the post received, and project its infographics, without even having to read a single comment! Sentiment Analysis, already a booming talk point in social media analytics, here NLP is being used extensively to determine the “sentiment” behind the comments that the users leave behind on AFH's posts. The Data Set will contain sample comments, categorized as 1 or 0, 1 as in positive comment, 0 as in negative comment. And using this, Likulator would predict the sentiment for the newly added comments. Next, the SVM training algorithm using machine learning tools i.e. scikit-learn compatible with Python, builds a model that assigns new examples to one category or the other, making it a non-probabilistic binary linear classifier. We use this to classify the previously predicted comments as positive and negative. There are multiple classifier algorithms, but the best fit for this particular situation was SVM Classifier, its accuracy rate was around 80%!
Challenges I ran into
The accuracy rate of 80% is well appreciated, but I did try and execute the data with other classifier algorithms. As it goes without saying, I chose the one with the highest rate, SVM Classifier.
Accomplishments that I'm proud of
This is my first Hackathon, and regardless if I win or not I will always remember this. Adjusting the posts to increase likes and positive comments can massively improve the promotions of AFH. We can attract members in masses, when our best-performing posts are promoted, which is well appreciated, because together we will be helping women everywhere! To be abled to build a small-scale application for a women driven organization to help acquire other empowering empowering women is my most proudest achievement.
What I learned
i did use various complex algorithms to build this model. Even though I was familiar with them on their own, combining them all and making them compatible with each other was definitely something new for me.
What's next for LIKULATOR.
An application running with such great potential to give extensive results, to make promotions easier. We can of course make improvements, add more data, improve accuracy, or even change the pathway through which we obtain these data. Likulator is just a stepping stone!