Hello world! Welcome to report where I'm using machine learning to analyze tweets about specified topic and present results in form of various and easy to understand charts. The sentiment analysis algorithm has been developed as part of my Master Thesis in 2017/2018.

This report is currently being published exclusively here on Steemit.

Parameters

Today's analysis has been executed on tweets which contain word "bitcoin" and were published between 2011-01-01 and 2018-12-31. Detailed specification of the data is shown in the following list:

• Keyword: bitcoin
• From: 2011-01-01
• To: 2018-12-31
• Number of analyzed tweets: 50000
• Tweets per week: 119
• Language: en
• Geographical location: Not specified

Results

Sentiment

After downloading 50000 tweets between the specified dates, sentiment analysis has been executed on each and every one of those tweets. Sentiment score has been then aggregated over weeks and months, to lower the granularity of results on the time axis and then plotted as a following linechart.

^{Sentiment of tweets for keyword "bitcoin"}

My subjective comment on the chart: Wow :O Well this is pretty cool...we can see that the sentiment predicts market cycle by cca one year...while price was rallying in 2013, sentiment was getting lower and lower during the whole year before. During the slow bull market start, it got better and better. But in 2016, before the craziness of 2017, sentiment already started to turn opposite direction.

Aggregation using heatmaps

To show the general trend/pattern in the sentiment, linechart works great. We can see the bigger timeframe and estimate the long-term direction. But if you're interested in particular month or week, it's hard and in case of weeks actually impossible to see the change. Has an athlete put the great performance in particular match? Has the brand/company released a new line of product? So see such low lever changes, following 2 heatmaps are to be used.

^{Chart shows average sentiment per month where 0.59 is the worst and 0.74 the best achieved score}

My subjective comment on the chart: We can also clearly see spike in September 2018 when BTC was showing strength on 6k and we thought we've found a botton. To the mooon! Once we broke 6k, all hell broke loose and sentiment went toootally down. Also, 2011 is interesting - Coinbase announced selling 60 millions of $ worth of bitcoin in february and check that sentiment out. In June, controversial WikiLeaks started to accept bitcoin :) See the sentiment change? I do.

^{Chart shows average sentiment per week where 0.38 is the worst and 0.96 the best achieved score}

Most frequently used words

---

Another very interesting aspect to look into are the repeatedly used words using wordclouds. Even more interesting is to compare two wordclouds generated from different time - usually before and after some event/change happened. If you give this a second though, the problem here is that many short words (like "and", "or", "with" and so on) are used almost in every sentence and would also show up in wordclouds. To mitigate this, I've removed list of 153 so called stopwords. Additionally I've also removed words typical for this area listed in the end of the report*.

^{Most often used words in tweets containing word "bitcoin" before and after 2014-12-31.}

My subjective comment on the chart: Shit, my filter on the word bitcoin froms some reason doens't work here :D But in the early days, we can also see words like casino and gambling, showing the first usecases of BTC.

Most frequently used UNIQUE words

As we can see in the previous worldcloud, there are many words which are actually shared in both wordclouds. That makes all the sense as there are many areas which will be forever connected with bitcoin. But I went one step further and decided to create wordclouds which contain only unique words with don't appear in the opposite wordcloud.

^{Most often UNIQUE used words in tweets containing word "bitcoin" before and after 2014-12-31.}

My subjective comment on the chart: Hah nice...in Before graph, we can see MtGox :) And also Satoshi is being much more discussed. Dogecoin is there as well :) Makes sense :) In the After graph, it's more money oriented - crypto, cryptocurrency, market....We can also see Ethereum and BCash popping up :)

* words excluded from all 4 wordclouds are: yii,bit.ly,.ly,bitcoin, btc

BONUS - shaped wordcloud from all words!

This one is just for fun :) It's generated from 1000 most popular words in all tweets, not divided into before and after groups. Click it to open!

About project

This series of posts shows the power of machine learning and it's application in the real life. It also makes kind of symbolical point of analyzing Twitter and publishing it here on Steemit. Technology of the future is being used on the social media of the future ;)

Get your report

---

Twitter sentiment analysis reports are being sold for quite some dollars in the world outside of Steemit. In our tiny word of Steemit, such price would be way too much - that's why I'm offering to generate& send you a report with your chosen keyword and dates for a laughable price - 5 STEEM. Order 3 and get the fourth one for free :)

Interested in how's you favorite coin doing on Twitter? Or favorite athlete? Politician, actor or clothes company? . Just DM me and you'll get the full report under 48 hours :)

Hope you enjoyed! Matko.