A new voter bot - newer, smarter, freer

TLDR;

I've made a bot you can use for free which is pretty dope. The idea is:

to automatically vote for posts that you would vote for anyway while also maximising curation rewards, bot by voting more regularly than a human can, and by picking great content which will probably be popular too.

You can check it out on GitHub, it runs as a Node.js server, designed for Heroku. There's a purple button you press to set it all up. Note, some assembly require but batteries included 🤓

There's a lot of features, but you'll have to read on find out... 😉

Voter - a new, smarter, freer bot

I haven't written a proper article on Steemit in 28 days. During that time, instead of writing articles, I've been using my free time to attempt to make a smart, free voter bot. I've named it Steem FOSSbot Voter, or Voter for short. 😁 FOSS stands for Free and Open Source Software. This means the source is freely available and open to (i.e. the license permits) modifying the code for your own purposes.

My original proposal for this idea is here.

There's already tons of voter bots and services, why do this?

All other well made voter bots I've seen are closed source, which means you cannot inspect the source or use it for yourself, and will be protected under a much more restrictive license. Most of these bots are commercial ventures, albeit often small scales ones, so there's often some cost, usually just casting a vote for the creators account. There's other bots that you can get added to, like a trail, or you can even pay someone to make a bot for you.

I want to make clear that I don't disagree with this approach or oppose these ventures (at least not simply because they are commercial). I do however believe in freely available software and I think voter bots are no exception. Steemit and the Steem blockchain are written as FOSS themselves, as well as countless other widely used software, so it's clear we all benefit from this approach, whereas only a few will benefit from the closed source model.

The second most different structural thing, is that you set up the bot to run as your own server, not on someone else servers, as is the norm. In stark contrast with any voting service, this means you do not give your posting key to another Steem user, your keys remain under your control.

Currently the bot is designed to work as a Heroku deployment. Heroku is a popular cloud computing service particularly popular with startups because it is free to use if you don't require much power. Thankfully our bot doesn't require that much and I believe it can be run entirely for free (though it's up to you to verify this, like any of my other claims 😅).

It should be possible to run the server locally, but I haven't tested it. It's next on my list, I'll be making installation instructions for this and making any changes to the code to make local install more simple.

I don't care about that, what does it do differently?

It's a pretty advanced bot, highly customisable, and allows you to tweak it in many different ways for your own needs and benefit.

Most of it is accessed through a basic but usable web app UI, accessible through your server URL. You can change the bot config, view logs and (my favourite!) stats about each time the bot was run, which votes were cast and how well it's doing for you.

It also sends you emails if you set that up, either every time the bot runs, or once per day.

Here's a few screenshots:

No, I mean how does it work?

Oh. 😅 Basically, Voter uses a scoring system for determining which posts to vote on. You use a web interface on your server to choose which aspects of a post you think contribute to a good post, and tell the bot how important you think each are.

There are a lot of aspects to choose from, which I call metrics, to borrow a term from analytics. They are each given short names, here's a few examples (there's 65 in total):

• post_up_voted_num_whale: Number of whales that already voted for the post
• author_is_followed: The post author is someone you follow
• author_repuation: Author reputation
• post_num_words: Number of words in the post
• post_any_tag_blacklisted: Are any of the post tags are on the blacklist?
• post_has_spanish_language_use: Post is (probably) in Spanish
• post_images_only: Post is mainly just pictures (less than 100 words and several images)

Already you can see we are looking deep into the post and information about the post, i.e. metadata. We're performing a little bit of Natural Language Processing (NLP) and trying to expose data which may be an indictor of quality.

And really the whole point is that the bot allows you to decide how to define what is high quality using this data.

Strategic and cultural curation

In my bot proposal a month ago, I suggested there were two styles of curation, strategic and cultural curation.

Cultural curation votes for posts based on the content of the post and other related cultural, and thus social aspects. It has an indirect affect on payout, and thus curation reward, because culturally "good" posts often receive good rewards. At least that's the idea, if everyone does it, especially the whales. You think it contributes to the culture of Steem in a positive way, however you wish to define it.

Strategic curation votes for posts that will yield the best curation reward, regardless of content substance. Any user of Steem will know that the "best" posts do not always get the highest rewards, and sometimes "undeserving" posts get very high rewards. The existence of betting, games, competitions, porn, etc. etc. shows this clearly.

It makes sense though to consider both aspects, cultural and strategy. They interact with each other. You can do this with the bot.

How does that actually work

Each metric is a numeric value which is multiplied by a numeric weight set by you, to get a part of the score for a post. These scores are added together to get the total post score. If this score is above a certain value, called the threshold, the post is voted on. The threshold automatically moves itself based on the average score of the last few posts, adjusted for how many votes you want to cast per day (and a little higher if you wish).

We call a collection of these weights an algorithm. You can set this in the web app. There are also some white / blacklists that you can add authors that you like / dislike to, as well as tags and categories, keywords and website domains.

Creating a good algorithm

A good algorithm starts with a hypothesis about what makes a good post. This should ideally be a balance between what gets good payout, and what describes "good" content.

Example 1, long form articles

As an example, let's say I like long form articles about interesting subjects, I see they get good rewards and I think they contribute to the culture of Steemit. So I hypothesise that good long form articles have more than 200 words, some images, not too many videos and that the best ones are usually written by someone with a bit of reputation, they've been here a while. Also, they are in English, because I cannot read another language and so would not vote for any in another language naturally.

In that case I might set the following algorithm:

• post_num_words in range 200 to 600, weight = -0.5
• post_num_links_image in range 0 to 5, weight = 10
• post_num_links_video, weight = -25
• author_repuation, weight = 1
• post_has_english_language_use, weight = 50

We're using ranges here in some cases to make the metric values more sensible. We don't care if there are more than 5 images, and don't want to over score based on that. This is similar for number of words and the reputation range. Also note that the post_has_english_language_use is a true or false, so if true then 50 will be added to score, if false nothing will. Metrics like this are clearly marks as Boolean in the list on the Edit Also page.

Example 2, artistic posts

Perhaps we're just interested in artistic posts. The hypothesis is that these kind of posts are pretty much just images with maybe a little text. The more images the better, the more words the less good.

• post_num_words, weight = -1
• post_images_only, weight = 500
• post_num_links_image, weight = 20

So for every word we loose 1 score post. If the post is images only (less than 100 words) then we add 1000, which is a lot but will high skew the score up for these kind of posts which we're mostly interested in. Additionally we add 100 for every image.

Example 3, the sell out

Say we just care about curation rewards and want to maximise that, completely disregarding content. Our hypothesis could be that certain authors consistently get high curation rewards, so with our Steemit account we follow them and then use one metric only:

• author_is_followed, weight = 1000

If the author is followed, their posts will always be voted on, up to our daily limit.

Reflection on the examples

The first two examples are contradictory in goals and represent two completely different ways of looking at quality. The interesting thing is that both of these kinds of posts might do very well and get good rewards, our hypothesis could be right in either case. What I like about them is that they are also heavily cultural.

The last example is very simple and might be effective, but it is not really intelligent. If the authors I follow stop producing content, or their content becomes poor quality, or even post too much, this will stop working. The problem with that approach is that it cannot react to changes in the system. However the first two methods will always look for different posts which fit the criteria.

In summary, I recommend you try to find out how to translate an idea you have about post quality that is a bit deeper than just who and what is popular now, because I've personally seen that Steemit is very dynamic and changes a lot, even from week to week.

Reflexivity with tests and stats

In order check if you algorithm is working you can perform a test run to get an idea of the scoring your algorithm and other settings will produce.

When the bot has actually run a few times, you can check the stats to see how things are going and adjust the algorithm accordingly. Perhaps your approach is not working at all, or maybe your weights are just off. It could be that one metric is contributing too much or not enough to the overall score. You can check this with the metric break down chart, pictured here:

I've used this and found it to be effective in improving algorithm design.

Okay okay, looks good but I'm not a coder

You don't need to be a coder to use this bot. There are detailed installation instructions, and as long as you follow them you should have the bot operational in half an hour.

To create your bot algorithm, you just need try to isolate what makes a good post and translate that into bot-speak. See the examples above.

Where people do find difficulty, I'd love to see people helping each other and figuring out how best to use it. It's a tool based on sound principles (I hope!) and even I as the creator don't know the absolute best way to use it!

Also, I will be hanging out on the # bot channel on Steemit.chat sometimes in the next while, hit me up if you've a question.

But bear in mind, this software is more or less "as is". I welcome criticism, bugs reports (logged through GitHub), contributions and free Steem / pizza, but do not expect to get customer level support and do not be disappointed when I won't built it for you.

This is a hobby project, a labour of love, and I just hope that it contributes a little bit to the community, puts some power into the hands of the dedicated, and leverages the unique property of Steem's curation rewards for everybody.

Thanks for reading!

As I said at the start of this post, I've been working on this project for about a month, and hope to get back to regular posting now. 🤓 I put in a lot of work so if you like it please consider re-Steeming this post and helping spread the word.

Also keep an eye out for @codydeeds, I believe he'll be looking at using the bot and might even do some tutorials.

And lastly, I've already written quite a lot into the docs for this project, which are located in the repository. I'll probably do a few posts explaining things more if there is an interest in the bot but there's a lot of information there if you can't wait.

😄 ✌️