Charting Stacks

42 Sentiments

Share via Twitter Share via Facebook Share via Linkedin Share via Reddit

It is always an interesting exercise to reflect on what you have done in the past. And as an analyst a lot of what you do is very public, which makes this an even easier exercise. As I finish up here at RedMonk, I thought it would be good to go back to some of the areas I have had the opportunity to look at over the last few years, select some specific pieces and get a sense of my overall sentiments.

This also gives us an opportunity to compare the results of the sentiment analysis APIs from Amazon, Google and Microsoft and see how they differ. For those that really want to spend some time comparing what the machines say versus what was written, the complete list of pieces is below.

My groupings were Serverless, Kubernetes, Hashicorp, Cloud Foundry, Kafka, Thought Pieces, Frameworks and Docker.

The nice outcome, from my perspective, is that – according to the machines – I am normally on the neutral to positive side for most areas I have written about. That reflects interest, but it also, to me at least, reflects my overall optimism about the work I have been privileged to observe over the last few years – the cloud native space in particular is just getting started, and the cake is going to be very, very large.

Looking somewhat closer, AWS and GCP mark me as neutral sentiment. Microsoft seems to find me a little more excitable, and generally further into the positive area

The Services

I used a number of simple API offerings for this analysis

It is worth noting that in a world where packaging matters, Azure Text Analytics was by far the easiest to get started with.

The Sentiment Formats

Each of the providers have a slightly different format for how they return sentiment scores.

AWS Comprehend

AWS returns most likely sentiment as well as scores for mixed, positive, neutral and negative. I have used the most likely sentiment as our value here, and map other providers to it. I have merged mixed and neutral.

AWS has a 5000 character limit on the document size you can submit via its API and recommends splitting the document into pieces. I split at 4500 characters, and where I have had to do this, I have averaged the sentiment scores returned.

Azure Text Analytics

Azure returns a single value in a range of 0 to 1. For the purposes of our analysis we have allocated the following weightings, to map to the AWS scale that we are using.

  1. 0 – 0.25 Negative
  2. 0.25 – 0.75 Neutral
  3. 0.75 – 1 Positive

Similar to AWS, Azure has a 5120-character limitation on the document size that you can submit via its API and recommends splitting the document into pieces.  As with AWS, I split at 4500 characters, and where I have had to do this we have averaged the sentiment scores returned.

Google Cloud Natural Language

Google provides results in a range of -1.0 to 1.0 with a magnitude value for emotion. Google was the only provider able to deal with document segments over 5000 characters. Every result we got from Google was between 0 and 0.3, so we have marked all as neutral.

If people are interested in the actual code behind this and the raw results, please let me know on twitter. It needs a little clean up, but I am happy to drop the relevant bits onto github.

42 Pieces









And thus, with the number 42 in my head, ends my time at RedMonk. It has been a lot of fun. Thank you to all the team, to all the people across the industry I have worked with over the last few years and to everyone that has taken the time to read what we publish. Please do stay in touch – I can be found on Twitter and Linkedin.

P.S. RedMonk are hiring.

Disclosures: AWS, Microsoft and Google are all current RedMonk clients.

No Comments

Leave a Reply

Your email address will not be published. Required fields are marked *