It'd be nice if the graphics made it clear what "sentiment: yellow" meant. A good graphic is its own summary. Especially since apparently the sentiment values are in an incredibly narrow range:
- what does "sentiment" actually measure? ("sentiment related" adjectives in posts, vs. "objective" adjectives?)
This is basically just detached data that's getting upvoted because... it's... pretty? The page as posted has literally no information, it just shows uninterpretable data that no one can draw any conclusions from.
I would just have one graph with all the companies. You're using one graph per company to show both sentiment and number of mentions, but only the former is interesting.
Like others have written bar plot is not ideal here, xy-curve plot would be much clearer.
> Ye quite worthless graphs without explaination. Also the sentiment and the amount of comments should just be two lines not colorcoded bars. xyy-plot.
You can kind of infer that blue means bad and yellow means good with some background info:
Facebook: yellower > quote bluer, starting around 2017 (coincides with a lot of criticism of disinformation spreading there/Cambridge Analytica).
MSFT: bluer -> yellower, guessing this is linked to them being less hostile to open source, etc.
Amazon: yellower -> quite bluer
Google: yellower -> bluer
> How do you even measure sentiment? The amount of curse words?
The sentiment ranges should be normalized. As-is, at a glance, it looks like HN loves Microsoft (very yellow!) and hates Amazon (so blue!), when in fact Microsoft's sentiment scale maxes out at a value below the minimum of Amazon's sentiment scale.
(Also, I'm assuming higher value = more positive, but who knows?)
Visualization tip - if you want to focus on sentiment, encode the sentiment as the size of the bars (or x/y-position of lines/points). What you've done is assign number of posts to the bars and encode sentiment as color which makes it more difficult to read.
(I would also pick a line over bars for this but that's a minor point.)
These should be normalized against each other and number of comments on HN over time. How do I know what’s just total growth of HN versus something meaningful for these terms?
Summary: Apple, Amazon, Google getting worse; Microsoft getting better; Facebook is Satan. That chimes with my experience of the changes over the past few years....
I'm in the UK education sector which seems to have become a Microsoft shop. Teams seems fine to me, but I think it's only because they nicked the idea from Slack. Outlook is... better than it was, but still no match for Gmail. OneDrive is incomprehensible. All the Office 365 stuff just seems like a big overcomplicated mess.
I just noticed, the sentiment numbers are completely different by chart for color. Shouldn't these be normalized for direct 1-1 comparison, or do these numbers not function this way?
If they are comparable, then interestingly Amazon is by far and away the most liked, which I find surprising. AWS launch posts lifting them up I guess?
All of the other trends seem to match. The slow turn on Google is cool to see in literal colors!
I can tell facebook sentiment has turned and google just a bit.
These graphs are useful for what they are, but from a data viz perspective it could be better.
It's too hard to tell small movements in sentiment patterns occur over time because the bars are too close. The x-axis isn't labeled, I don't know from a glance how much time each bar encapsulates and not least the sentiment color axis is unlabeled.
To make it better, I'd want a line graph of sentiment over time, as well as a smoothed average. Any major events with a spike in sentiment could be labeled on the graph. Ideally the sentiment scores would be places in some context that is a good reference.
This is cool, but I'd love to see a plot with all the axes (including color gradients) held constant across all charts so that they can be directly compared.
It's not a perfect solution, as I'm sure some of the discussion on these companies is loaded with jargon that wouldn't be interpreted accurately by a general sentiment analyzer.
In my view amazon is being fairly honest in comparison to other scumbags: yes they mistreat their workforce and cheat on reviews, but it's all part of the game, they didnt create the game.
Looking at the colouring (that reflects sentiment, right?) Amazon is pretty unpopular - and taking a rough mean over the past few years, similarly disliked as FB.
This visualization is a mess. What does "sentiment" mean? Why are the ranges for each different? What does "comments" mean? Comments about the company or on threads about the company? Is it normalized for total comments on HN?
The sentiment trend on Microsoft doesn't surprise me. The recent moves into .NET Core / .NET 5 has been received very well by those trying to build/maintain complex line of business applications. I have personally contributed to the sentiment metrics shown here by sharing my positive experiences on HN over the last few years.
Moving off .NET Framework felt like getting out of jail to me. We can now look at additional platforms and our confidence in the platform is better than ever.
This would be easier to read if the number of comments was dropped in favor of making the Y-axis be the average sentiment, and X-axis time (one month or quarter year buckets).
I used Facebook as a reference. Since it's all blue, I guess lower score denotes bad sentiment. Extrapolate it and you get to see it make sense.
People are more positive about Microsoft now than before, Apple and Netflix get an even sentiment through the years, Google and Amazon used to be yellow then turned blue since 3 years or so. And good old Facebook is perpetually in the blue.
- what do the colors represent?
- are marginally higher numbers marginally better/worse sentiment, statistically significantly better/worse sentiment, etc?
- what does "sentiment" actually measure? ("sentiment related" adjectives in posts, vs. "objective" adjectives?)
This is basically just detached data that's getting upvoted because... it's... pretty? The page as posted has literally no information, it just shows uninterpretable data that no one can draw any conclusions from.