ML-India is pleased to announce its ninth machine learning meetup in its Gurgaon chapter which was held on 26th March, 2017.
In the last few meetups, they had been hosting some reputed researchers to speak on popular machine learning topics. The talks were received very well by the group members. This time around, owing to popular demand, they decided to organize a hands-on tutorial on sentiment analysis of tweets. They analysed what Twitterati feels about our political leaders like Arvind Kejriwal and Narendra Modi. The idea was to get people started with a basic problem in data science.
They had around 15 people attending the hands-on. There was a mixed crowd ranging from students to people who are using machine learning to trade in stock markets. There were three guys who just had their startup acquired by a Chinese company and were exploring machine learning on images.
They began with a quick introduction to iPython to acclimatize participants who weren’t accustomed to it, and then quickly moved to introducing popular packages like pandas, numpy and scipy, which are required in data science. This built the required momentum for everyone to get their hands dirty on actual code. Everyone was able to add/ remove cells in ipython and do basic stuff. They then introduced the Twitter API and the functionality provided by python wrapper twitter-python. The participants then got credentials for their twitter accounts and setup the twitter-python module on their local instances. By this time they become comfortable posting updates on twitter, fetching friends and their tweets.
Following this, they discussed a few approaches of performing sentiment analysis on a piece of text and talked about various alternative approaches which can be tried for sentiment analysis. They introduced TextBlob package which has a trained naive Bayes classifier trained on movie reviews. Participants used this package to analyse 1000 tweets for each of the political leaders - Narender Modi, Yogi Adityanath, Rahul Gandhi and Arvind Kejriwal. They discussed the anomalies in the analysis by looking at actual data in CSVS. Many quick ideas were floated to fix the analysis and make it more accurate for gauging actual popularity.
Click here to download the slide deck from the talk.
They concluded the meetup with a feedback session. Some participants suggested that more sources of data like quora, reddit or facebook could be used for further sessions. The guys working on stocks were able to relate these concepts to measuring sentiment on different pieces of news and then predicting the change in price in a particular stock. Another seasoned participant stated that an analysis found 90% of data on twitter to be useless and uninformative, and hence, extracting useful information is a big challenge.
The beginners in data science showed special excitement in these hands-on sessions. They said it was very useful and they were able to follow 30-40% stuff despite being very new to this area.
Click here to join the group.
If you wish to start a chapter of ML-India in your city, we would be more than happy to help you out and get it started. Write to us!