AMA Transcript: May 19, 2022.
For your reading pleasure, we made a transcript of Steven’s last week’s AMA! Translations will follow in the next coming weeks, so please bear with us.
Today’s AMA covers three major parts: 1/ overall progress, 2/ deep dive on the Echo platform which I think everyone has a lot of questions about, and 3/ the Q&A part.
Network development + Staking Update:
Okay, so a quick progress update. In terms of staking, we’re doing extremely well since the last AMA: total staking increased by 34M, and more importantly, we’ve increased total delegation by 58M. This is great, as it means our overall delegation rate is going up, so thank you to everyone who has participated in staking as well as delegation to help secure the Taraxa network. In terms of security, the network has been running pretty safely: there was one bug on the mainnet and there was a small bug on the testnet. Overall, everything has been running pretty stably: we haven’t had any crazy crashes, so this has been going well.
In terms of our development towards the native token conversion, we’re 67% there. All of the stability features that we wanted to get into layer one have been finished, so that’s really great. The next step is really focused on security as well as economics: the economic features are almost all completed. We have also signed with a very reputable security auditor Halborn who did security audits for projects like Avalanche. Right now, we’re ramping up our team and their team to see how we can work together smoothly. Once we finish the security audit, we’ll submit the report to the exchanges that the Taraxa Foundation has contractual relationships with — KuCoin, Gate, and AscendEx for them to tell us how long it will take for them to integrate our conversion into their exchange. There will be many tests prior to the actual conversion, so we will be mirroring the ERC-20 onto our test network many times so that everyone can check it out and test your wallets to make sure things work out the way they should. We’ll be doing tutorials on how to access your funds on the interaction network during these tests so everyone will have plenty of opportunities to test it out prior to the actual conversion.
A reminder to delegate your stake:
Now first, let’s celebrate the fact that the un-delegated rate has dropped from 13.5% to 7%, a drop of over 6%. Once again, thank you to everyone who has staked and delegated! There are still 7.45% of speakers who have not delegated: please delegate or else you won’t be receiving any yields.
Okay, the next one is the social listening app, and we’ve mostly been making progress on the data collection and analytics layers. On the data collection layer, things are going a little bit slower than expected because we’ve had the progress at 99% for several weeks. What we’ve seen is that we are getting our bots kicked off from Telegram groups. Still, we have many ways to solve this problem and the ultimate way is to finish a fully decentralized architecture, and this is the direction we’re going right now. The original plan was to build up the data collection network in a way that’s pretty centralized and then slowly decentralize it as we go. Anyway, it seems right now that decentralized architecture is actually a way to go because if it’s too centralized, social platforms tend to kick you out, so we are moving towards a more decentralized architecture. So we’ll keep everyone updated on our progress as well once we make some breakthroughs on how to defeat the problem. On the analytics side, we have been using the Telegram data to test it out, and this is going to be actually the focus of this AMA.
Taraxa’s Tech Stack.
The focus of this AMA is to give you a sense of what we’re currently building on the Echo platform. A quick recap for people who might be new to the project, or never heard of Echo: Taraxa has several layers of our tech stack: at the bottom, there’s a layer one and then there’s a platform layer. The platform layer enables different types of use cases as we’ve talked about: Echo for social listening, Marinate for informal business transactions, and Helio for machine-generated data.
On top of Echo, we’re going to be building different types of applications that make use of social data. One of those is Hype — a way for projects to really maximize the way their marketing campaigns are building communities in a more effective way without spam or waste of resources. It’s about being able to identify trends before they are being talked about on social and becoming big. So, these are some of the things that we hope we can enable, and that’s what Echo is about.
Echo platform: stack + current development.
Now, let’s do a deep dive into Echo: we’re going to talk about the data layer and specifically the analytics layer because we’ve been getting a lot of questions about that from the community. Inside Echo, there’s a data collection layer that collects publicly available social data in a decentralized manner, so that’s exactly what we’re doing right now, starting with Telegram data. The next one is the analytics layer which takes the data that has been collected in a decentralized way and conducts analysis to convert them into quantifiable social signals. These signals are things that applications can make use of to address various pain points. The third layer is the application layer: things like Hype, or trend-spotting applications, i.e. the applications using the signals to solve specific problems.
What are we doing right now with the data layer and the analytics layer? On the data layer, we have an architecture that’s sitting in 2,800 Telegram groups collecting all data periodically, it is very stable. As I mentioned earlier, the issue here is that if we try to scale that, we start getting new bots and getting banned. So what we’re doing right now is going towards a more decentralized architecture. Our short-term goal here is to hit around 8,700 Telegram groups because those are the crypto-related groups that we could find, but, of course, later on, I’m sure the community will let us know if they find more of those and submit to us. But that’s the minimum goal that we actually want to hit, and we’re very very confident we can get there.
The second part is the analytics layer that’s about taking the data from group discussions on Telegram and then trying to figure out how to analyze and quantify them. There are three areas that we’re specifically looking for right now and they’re already related to the Hype app. Hype’s goal is to figure out how you can set up marketing campaigns that reward the type of behaviors you would like to see on social media, i.e. the things that you definitely want to see are things that are critical to building your community: spreading the right messages, not spammy, being talked about in the right context. And more importantly, actually creating a lasting impact in the communities where your project or ad campaign is being talked about.
There are three areas that we’re really looking at inside the analytics layer: relevance, quality, and impact. Right now we’re mostly focused on relevance which is basically asking ‘is this message relevant at all to your project or is it relevant at all to your ad campaign?’, i.e. the most basic filtering quantifiable signal.
Using Echo analytics for the Hype app:
First of all, just to re-reiterate what the Hype app really is: it’s basically looking at how can you hype up your project to maximize community-building impact. That’s really what you want as a crypto project: to build up a community of fans in a sustainable way, i.e. your project’s being continuously talked about on social media.
What are the exact pain points in crypto marketing that Hype is trying to resolve?
If you’re trying to promote your project today, there’s a good chance that you’re wasting 99% of your spending, while getting a minimal amount of impact on social media. The reason why that happens is that you can’t differentiate in real-time things like spam and relevancy. It’s a recurring thing that happens in lots of crypto ecosystems where you pay people to do certain things, but it all gets consumed by people like bounty hunters, whose job is literally to go out and consume social bounties — they don’t really care about your project. The exact same thing happened with Axie Infinity, where most players are people from the Philippines who do this for a living, not really caring about the game being fun: so when the incentives stop, they stop playing the game. So, when you start doing these marketing campaigns, they all get consumed by people who are probably not going to be future community members, and they’re not going to influence or impact anybody who potentially could be a future community member. All your campaign budget is going down the drain, and you’re not really getting anything out of it.
Types of quantifiable metrics Hype is looking at:
There are three different types of quantifiable metrics, and then there we have lots of sub-metrics that we’re looking at relevance, quality, and impact. Relevance means looking at a specific message and trying to figure out if it is valid uh for payment, or to reward this type of behavior, — sort of a filtering category that answers ‘are they even talking about the project at all?’ Quality is the second category: quality answers ‘Are the discussions meaningful or is it just spam?’ A little bit of spam can be good, but if that’s too much, people be attaching a negative sentiment to your project — definitely not the way to build a sustainable community.
The third, and most important one is impact: do you know how you actually trigger any kind of response or discussion inside the social media platforms. This one is really important, and this is really the final goal. So, the quality factor is sort of an intermediary goal, but the impact is really what you want eventually — generating discussion inside the right groups on the right social platform, ultimately bringing new members into your community.
Let’s dig a little bit deeper into these different categories. In the relevance category, we have baseline admittance which answers ‘does this actually make it right for someone like you to be talking about my project?’ This basic relevance is actually interesting if we consider this across different dimensions: has the person who’s posting it, ever posted crypto-related content in the past? If someone who’s been posting about cats for a while is now posting about crypto on your behalf, those posts are not going to be very relevant as that’s obviously not a crypto influencer. The same goes for the group that this person is posting in: if it’s a group where the subject matter is not relevant to your project or your project’s industry vertical, that’s also not so relevant. The third one is what the content of the message is, i.e. the explicitly stated content of the message: if the message says something about Taraxa, is there something to attract attention that is related to your project. And the fourth one is contextual intent: is it contextually related to your message or not, because in platforms like Telegram and Discord, you might be talking about something but not mention the project, so if someone says ‘Taraxa has just released the testnet’ and ‘Oh yeah, I heard about that’, that second sentence doesn’t really have the word ‘Taraxa’ inside. So, contextually you need to figure out if this is also talking about Taraxa or not.
So, baseline admittance basically answers if the message fits into our consideration for reward: is it at all related to your project your ad campaign, or even your industry vertical. This is very important because the vast majority of messages on social media platforms are not related to what you’re trying to promote, so that needs to be filtered.
The second category is campaign differentiation. At different stages of the project you might want to market different things, so maybe this month you’re talking about your testnet, the next month you’re talking about alpha mainnet, the third month maybe you’re talking about a Dapp partnership. Okay, we have all these messages that are now related to my project and being talked about in the right context, in the right groups, and posted by the right people. So what you’re really trying to do here is to figure out which of your campaigns are they most closely related to, because you might have different rewards for different campaigns, or recency requirements, where you only want to reward the newest campaign. You can now have all these variables so you want to know which campaign it’s related to.
This also goes to another core philosophical concept that we have for the Hype app, and for marketing in general. We believe that marketing shouldn’t happen in fits and starts. So marketing and community building (or hyping in general) should be consistent: it should just happen all the time in a continuous stream that rewards positive behavior. Once you have that continuous stream, another problem is to differentiate between various types of emphasis, because having a continuous stream of hype doesn’t mean you have different emphasis across time. So, campaign differentiation actually helps you to figure out which campaign specifically is most closely related to.
The next one is quality: is it spam, is it informative, and of higher quality? The first sub-category here is crowdedness: are these messages being posted at a very frequency or not. Frequency is very relative, can be inside one group, or across different groups. It’s also very relative because certain groups are very talkative, and if you have a group of 100,000 people with 10,000 people online and they’re all talking, then the feed scrolls up very quickly on your phone or your computer. So if you’re posting at a relatively higher frequency in those groups it might be okay because you know as soon as you post something it gets scrolled up within 10 seconds, so you might need to post it a few times in order for that to work. But if you’re in a smaller group of 20 people online, and no one’s really talking, then you just post the same thing five times, then that’s spam, which only hurts your project.
The next thing that we want to look at is sameness: are the posts highly similar or not? If people are just copy-pasting the same content, there’s definitely going to be a very steep drop-off or reward decay, when it comes to things that happen very frequently and are exactly the same. That’s basically a definition of spam, which you don’t want to reward.
The third one is impact. You have visibility and interactivity, so this is really an end goal: to help your project gain more followers, and more community members by instigating more discussions about your project. So, visibility is about how many people saw your campaigns: in traditional SEO there are things like impressions and engagement, and you can gauge that just by seeing how many people are online, or in Telegram you can actually see how many people saw this message. The more visible, the better. We can actually get those signals and figure out how visible are these messages.
The next category is interactivity. This one is really important, as it instigates further on, follow-up discussions, so you can imagine someone talking about Taraxa launching a testnet inside a group, and then other people start talking about this topic with you, explicitly replying ‘hey I’m interested in this’ or ‘I heard about that’. Or, they have contextual discussions where they don’t specifically reply to but talk about something in close proximity to what you’re saying. So, these are the things you want to see: your ad campaigns to generate discussions on social networks. This is the holy grail: the more of this, the more interested the community is in what you’re doing, potentially bringing more community members.
Another thing that’s missing here is sentiment: is the person talking about you saying positive or negative things about you. The sentiment is very easy to detect from a natural language processing perspective, and we don’t necessarily take a view to only reward positive sentiment — sometimes talk is talk, and any talk is good, so if you’re famous you’re famous even if you’re a little bit infamous. That depends on what result a project really wants to get, so we’re leaving it up to them to reward only positive sentiment messages or both.
All of this is done through a combination of basic search and natural language processing technologies that we’re using. NLP is really a type of AI algorithm that has been to have some basic understanding of what humans are saying. That’s very exciting and a very interesting field of technology that we’re leveraging here.
So, these are the things we’re working on. Right now we’re mostly focused on the relevance section: is this message even related to your project or it’s a post, or a group is even related to your project right, and also ad campaign differentiation.