Eva: Adult CPA's Wet Dream

Okay, you've done SEO, you've set up a site, you've got your affiliate links and chosen your offers and landing pages. But have you tried the social angle?

Social Networks, Chat, and Apps are all over the Internet today. It seems like you can throw a stone and hit 20 social networks before it hits the ground. How do you leverage all of that?

Many people use automation when it comes to Adult and social networks. The problem with most automation though is that it's very limited to only specific actions. Let's say that you've got an account on a dating network and you're doing the standard stuff most affiliates with bots do...

The Standard Approach

1. get a list of leads in the form of user ID's on a site
2. go through the list and friend everyone on it, send a message/comment if possible
   
3. possibly spin your message to be more organic, use the user's name in the message
   
4. multiply this by more accounts using multithreading

Now, if you use the standard approach, chances are that if your list is well targeted and your message is great then you can get some pretty nice final conversion ratios after some tinkering. You might even get 7%! Don't get me wrong, 7% is great and you're making money, but is it better than 40%?

The main problem with the standard approach is that usually this gets your account deleted very quickly and you have to write your message generally and get all of your thoughts out in one message. Because it is general and because you are probably dropping your link right in it, many people on today's internet will look at your message and trash bin it directly. That is why some affiliates use the advanced approach. Less work, better results.

The Advanced Approach

1. Do some action on the site that will get attention (ie: Playing Match games or visiting user profiles)
2. Wait until you've got some messages in your inbox from all that attention whoring.
3. Auto-reply to these messages with one or more on-topic-but-general, spun replies, disregard what your new friend is saying and drop the link
4. Rinse-and-repeat simultaneously with multi-threading over multiple accounts

With the advanced approach, the advantage is that you get the benefit of the doubt. All of the people your bot interacts with are people who were for some reason interested enough to send a message, so this is almost like an opt-in lead. Using this extra trust factor you can gain some more points with them by customizing the replies based on the user's profile. Even then you are still running a small chance that the person is online and starts replying to you before you drop the link.

Regardless of ignoring their incoming messages, you can still pull off 10% or even 15%, but you have to take extra precautions to make sure your accounts are not deleted. Sometimes you will get reported, sometimes admins will review your communication, but every problem has a solution.

The Smart Approach: Eva

That was a lot of foreplay just to get to the juicy meat of this post, but if you're still with me at this point then it's totally worth it. What was the one thing you noticed about both of those approaches?

They lacked meaningful interaction with the prospect. These approaches are like walking around, (or sitting there,) with a magic baseball bat and hitting people over the head with it as they cross your path. Granted, if your starting list is targeted or your site profile is "off the chain" then you aren't just giving random people concussions... You manage to magically knock more of them right over to your affiliate link... but can you swing a 40% home run average?

Sorry for those of you who don't know or care about baseball. I'm American so it seemed like the most stereotypical metaphor (*simile?) I could use.

Eva is an intelligent chat agent specifically made for the Adult CPA vertical... yes, an intelligent chat agent. She's not just any back-alley eWhore bot. There are not many chat bots which I know of which focus on this vertical. Here are some of her non-secret features...

• Automatic Categorization of incoming messages by intention, polarity and sentiment/emotion (I'm going to try not to get geeked out in this list, bare with me!)
• Live training of the auto-categorizer, no need to compile new training
  
• A full Conversational Model template set up which you can add your offer to (like adding a widget to a wordpress site), including but not limited to smalltalk, information exchange, flirting, a pitch and a close. The model flows so that Eva pushes them toward the close (ie: your bad ass offer link)
  
• Personas - still is Eva, but using different personal data (ie: the info from your profile on the site you're automating can affect the flow of conversation)
  
• Full dynamic variables and entity recognition
• Data extraction from chat (get e-mails, phone numbers, social URLs, whatever information you want from your leads all nice and tidy in a spreadsheet)
• Reply based on any data (ie: if the user is younger, mention it.)
• Link replies to other actions
  
• Text spinning on replies
• Asking questions (takin' names! no really though, get more data out of them)
  
• Topic and intention tracking
• Full Feedback based training
• An API allowing your server or software like webDOM 4 to use Eva and chat, comment, or send replies intelligently on any site.

All of these fancy words are trying to say in short: A system that will chat with your leads and not break their suspension-of-disbelief. A system you can tailor to your needs and quickly generate the same CTR as if you took the time to chat to all these horny bastards yourself.

This project is just about ready for beta testing. I will be posting back in a couple of days with a progress update and possibly a date that I start the beta. I will also provide a demo chat for you all to feast your fingers on in my next post.

Here's some screens of the unfinished interface...



If you'd like to be a beta tester, you can DM me or reply

 

Verse Update - 2019

Hey guys and gals! It's been almost two years since I first started Verse and about a year and a half since I started this thread. The reason it's been taking so long is not because of difficulty in developing Verse, but more because it is currently a personal project and nobody is paying me for it yet. As you can imagine, I've had to take on many other paid projects, do some affiliate marketing, etc. while Verse has sat on the sidelines, slowly advancing. I have definitely been working on it, but work has been intermittent up until now. What has changed? Well, firstly I am itching to release the beta so that I can start applying more training to the AI. Currently there are about 7k incoming messages in the system total, and training has only been done on just over 1k of those messages. The training spans just over 250 different intentions. Needless to say, I'm almost ready to open the doors to a small group of beta testers as well as anyone willing to invest.

I have also made a couple of different Agents (means a different chat bot for a different niche). The first is Eva, an English speaking Adult CPA agent. The second is Hilda, a German speaking Adult CPA agent. The third is Vee, a multi-lingual bot which I had setup over the Messenger API to talk to potential clients about Verse and get them to register. Vee is currently on pause right now while more development is done.

Eva is the most advanced agent in terms of training and her conversation model maturity (I know it's a robot, lol, but it's easier to refer to it as "her"). Here is a sample conversation I just had today with Eva.

Me: hello
Eva: yo
Me: how are you doing?
Eva: hey cutie
Me: hahaha, you are so hot! I think you're the most beautiful woman on this site!
Eva: I actually don't use this site that much
Me: Oh yeah? what site do you use?
Eva: I don't have skype
Me: do you have whatsapp?
ERROR with API...

As you can see, she handles the beginning of a conversation very well, as well as detecting intentions for messages given the small amount of training and the relatively basic conversation model I've set up for her. The error with the Java API client happened because the agent did actually have an error, but it came from the fact that the API client was literally just worked on right before the conversation and the fact that Eva didn't have a reply set up for guy asking for whatsapp. The goal with Eva's "sell signup" conversation is either to get a guy to go to an offer OR give his email address (which is then extracted directly from the chat and put into a database for later use.)

Different Conversations = Different Methods

You know how every affiliate has his own workflow or method for how they generate leads? Sometimes they get the lead from skype and send to the site, sometimes they get the lead from a dating site and then later on reply over whatsapp in order to chat for awhile and then drop the link for example. In any method, the approach itself is usually restricted and limited based on the tools available to the affiliate at the time.

Verse is setup so that these limits disappear. Since it's an API service, there are literally no restrictions to which platform you can chat on. I am working currently on the Java API client, but also there will be API clients for PHP, javascript, command line, python, .NET, etc. This way any bot developer can incorporate replying to messages the smart way.

Of course, I will also be providing bots for different chat services and sites which use the Verse API to converse intelligently on the most popular platforms.

What is Left to Finish

I actually maintain a pretty large list of prioritized features that need to be developed in order for me to consider Verse complete and ready for beta testing. Among those are my current priorities:

1. Updating the training chat interface so that all errors can be inspected and corrected directly in the chat. This will make the training process much faster so that I don't bust my wrists switching browser tabs between all of this data.
   
2. Topic Pushing - You know how push notifications work right? they push out a notification to a person's phone or browser. Topic pushing is similar in that, in a conversation, you can either reply or deliberately push a topic by "changing the subject". There are a complex set of rules I have devised for pushing topics which include everything from what is known in AI as backpropagation to manually created links from a reply to a new topic. Both the underlying functionality and the interface elements for this are being finished.
3. Listeners - A listener is an asynchronous way to listen for answers in a conversation. For instance, if Eva asks "do you have any pets?" she will listen for an affirmative reply or a negative reply based on intentions found when the end-user responds. Listeners are asynchronous because the user could respond to a question or command a couple of messages later on in the chat. I already have pattern based listeners (for email, phone number, etc.) and intention based listeners set up. Now I am working on how the system extracts and stores the data gathered from these listeners.
4. Offers - Both working on the functionality and interface to setup and manage offers. This is perhaps one of the most useful features because offers are dynamic, meaning you can change them out just like you would change out banner ads on a website. Each offer can be attached to a topic, can be linked to a specific reply, or it can be an entire topic by itself. Offers include target; like a link, a phone number, a landing page, in-chat email registration, or contact for more information.
   
5. Synonymous parsing - There are many types of synonymous phrases in every language and many times slang is used in chat. For instance, I could say "what's the haps?" ... the intentional training might not always recognize the phrase as having the intention, so synonymous parsing takes certain semantic phrases and makes them equal to other phrases which are more standardized in that language. In the above example, the part of my message "the haps" can be marked as synonymous with "is happening" and thus the filtered final message can be processed as "what is happening" and the intention can then be properly recognized. This is only one small example of synonyms but there are at least 13 synonymous relationship types already available.

Conclusion

The above are only the top 5 priorities on my completion list and there are many others yet to complete in order for Verse to be ready for beta release. For me, the most important factor in completing verse is time.

I estimate that with some good investors, I could complete Verse beta within a couple of months, but without investors it could take upwards of six months. Either way it goes, Verse is getting better and is almost ready to start handling more complex conversations which will keep your potential leads chatting away for hours or maybe even days, gathering their data and establishing a bond with your lead before it drops the offer on them. Eva is no longer a wet dream, but a learning and very real option for generating leads, gathering lead data, and pushing offers through chat.

While I'm finishing up the top priorities listed here, I will also be training Eva and testing her out with a number of sites. The training corpus is still small, but the AI algorithm is pretty robust as you can see from the example.

If you are interested in Verse/Eva at all, let me know in a direct message here and we can chat person to person

 

Hey, thanks for the reply @Graybeard, I can always count on you to add some value to my posts

As far as the way the system works, it uses a customized version of the Maximum Entropy (MaxEnt) algorithm along with n-grams to perform statistical calculations and categorize an incoming message based on training data. The same way many industry chat bots do it now. This training data has tagged intentions as well as language tags. Though when actually explained, most AI doesn't seem like magic or even AI anymore, so sorry for the spoiler guys! no machine wars just yet!

The difference between this and other MaxEnt systems is that it can be trained live and in-line so that it calibrates without having to reset the system and update the training data manually. Besides that, it also has what I like to call "nested intentions". This allows for the system to be specific about intentions, but still include contextual ambiguity enough to make a pretty solid guess as to what topic the conversation should end up in. For instance, if a training message was a greeting with a flirt, the intention tag for that message would be #greet.flirt ... so the root intention is a greeting, and the sub-intention is a flirt. Intentions can have as many levels as possible for a heightened level of specificity, but usually one to three levels is all that is necessary. The more training data that is tagged, the less mean error exists in the categorization algorithm and thus the more accuracy there is in calculating the intention of incoming messages.

So since it uses intentions instead of keywords, it means that all possible words and phrases from training messages are actually linked to the message and intention. This is where MaxEnt statistics come in and finish the job of finding the proper intention for the incoming message. After that, the conversation model serves as a map to follow as to what topic the message's intention falls under in the conversation.

The idea of the conversation model is that it is based on a real conversation. If you talk with anyone, there will always be a current topic your conversation is on. Whether you're greeting them, or you are talking about politics, your conversation naturally flows from one topic to the next and generates an overall "flow" or a path through all possible topics (a multi-dimensional sparse space) the conversation could have. Sometimes one person changes the subject based on an internal relationship that they follow from one thing to the next. A conversation can begin talking about politics and end up talking about how to properly grow cabbage, but it usually follows a smooth path through other topics to get from point A to point B.

So both people in the conversation can push the current topic from one place to the next. If one of them has an agenda or goal they wish to get to in the conversation, they usually choose a path which best represents a smooth transition through topics from where it currently is back to making their point. Sometimes they take the shortest path and force the conversation back to that topic. Sometimes it just naturally or smoothly gets there. So overall the idea with the conversation model is to allow for enough topics to be talked about, and free range between those topics by both the user and the agent. When there is a goal in mind, Topic flow paths can be included to coerce the conversation to the point of a goal, in this case, an Offer. Think of it like a telemarketing sales script: you want the salespeople to follow the script, but as they do, they also learn how to improvise better ways to reach their goal and illicit some action from the person on the other side of the call. That is what I consider more unique about my homegrown approach.

Hahaha, that was way longer than I expected it to be.

In regards to your comment about hooking this up to a VR character or a webcam girl, the system already as a planned hook system which can do post-backs to other software. There is a "task/action" system which can be used for this, and actions can be linked to a specific reply from a specific topic, or any condition within the system.

Machine Learning and AI are both industry parallelisms which basically mean the same thing, but also almost mean nothing really. All current AI and machine learning uses big data and statistics in order to brute-force the semblance of intelligence. It doesn't actually matter if the system passes a complicated Turing test, but more that the end-user views it as an intelligent system. Once you take all the smoke and mirrors away, we're all just gathering data and running statistics on it, then applying those statistics to some action. A more accurate word than AI or "machine learning" to describe an intelligent system would be applied statistics. That is only my opinion, but if you look enough into the algorithms I linked to above you will find the word "statistics" right alongside "machine learning", "neural networks", "probability" and "AI".

So yes, your idea is very possible. There have been multiple ideas as far as applications beyond just adult CPA, but really nobody is giving adult CPA some chat bot love. I mean there are type 1 bots which just output a list of messages over time, and there are type 2 bots which are advanced because they use 10,000 phrase long keyword lists and have them all linked directly to replies... but currently there is nobody with an "AI chat bot platform" for the CPA world that is focused on generating leads from conversation and implements a REST API. I hope to be the first when this is ready for public release

 

Thanks, I will check those out. Yes, with this the presentation is what matters. I know that it Eva is targeted at CPA marketing right now, but the Verse platform will definitely be useful in many other verticals.

And yeah, I think some rules are quickly being turned into laws, etc. about bots being presented as people, etc. So I will find a good way to angle it such that it can be used for anything. I like your idea about the novelty thing, but how do you propose I make money with a VR girl, etc? Do you mean I should still use it as a marketing method (wherein she sometimes sells things to the user?)

Really though, novelty and adult CPA aside, the chat-bot realm is starting to gain a lot of traction and really the point of even joining the already surging market for this is to start out with a niche. Whether it be adult, etc. time will tell.

When I started webDOM I didn't even know what CPA was. I chose the niche "indie musicians looking for more real exposure"... but what ended up being the perfect niche was affiliate marketers and people wanting to sell things online. It took some interaction with clients for me to realize I had the niche wrong. As soon as I switched the home page and started to angle the pitch toward affiliate marketers, it started to take off with the help of some of my clients.

I'm hoping to start at a couple of niche verticals for Verse where there is a need but low market saturation and then have it naturally transition to a more optimal niche as people start using it and providing feedback. Thanks for the idea, if you have any others for this one, please do share.

 

LOL, YES! I will definitely try that out when this is ready. *pulls Blender and Unity off of the shelf and dusts them off*

Actually, there are quite a few option for realistic human 3D models. MakeHuman is one of them that I have used. The VR models could have different clothes that the user picks, age, race, etc. can all be compiled by makehuman into a humanoid model, fully rigged and poseable. New animations for the rigs could be added to the system as it grows... hell if it makes enough money one could just hire some talented animators and spit out new animations all day.

The chat could hook into a server that generates the models given user requirements, and then start a video chat. Facial expression recognition libraries can be used to give further hints to the VR girl about what the user likes. HAHAHA, actually with QuickFap, one of my other small projects I gathered a large matrix of porn tastes over a short period of time. I could probably use that to get an output of associated "toys" and other assorted items. Hell there are tons of freely available 3D environments too.

Here is the quickfap intro video Caution [NSFW] "Because there's no time to waste"

LOL, anyway, yes, great idea, thanks!

 

Update

To read the list below this one, you need to understand first what a couple of the terms are

• Agent - a "chat agent" is what you would normally call a chat bot. It's a collection of training, conversations, and semantics. An agent can be multilingual or only use one language
• Persona - this is an instance of an agent that has it's own set of attributes which can be used in replies
• Attribute - A variable value that you can set which acts as some statistic about the agent, for instance the attribute "favorite.color" should hold the Persona's favorite color. Let's say you set this to "blue". In the reply: "I really love $(self.favorite.color)" the variable $(self.favorite.color) would resolve to blue and the final reply would be "I really love blue". There are a wide range of already existing attributes including location, gender, age, and things like favorite.videogame or passtime. Attributes go beyond just the self, and can be used in replies to reference attributes the system has gathered about the "other" Entity in the chat. The value of an attribute just like a Reply can be spun text, so more variation in the final reply will be possible. For example "origin.city" can have the value: {LA|la|los {A|a}ngeles|Los Angeles}
  
• Entity - Anything whether Agent or a real person chatting. In Verse any message sent has an entity which it originates from and an entity which it is being sent to. It's important to mention that Personas are the "entity" of an "agent". So If I have the agent Eva, I can have an entity called "jean" which represents Eva on a specific chat platform. Jean is a Persona of Eva, but also Jean is an Entity with Attributes.
• Interface - Verse can chat on any interface. It is an API after all and does not have it's own chat platform. So, an interface can be any site, app, or chat platform which you would use the Verse API with.
• Dialogue - Dialogues are just a chat session between two Entities. Each dialogue has it's own ID and allows Verse to track conversations, etc. A dialogue can be cross-interface. There is a planned feature that can track the same Entity across interfaces based on a number of factors.
  
• Conversation - A Conversation is both similar and dissimilar to a real conversation. A Dialogue is closer to the natural meaning when you say "I had a conversation with him." When you see the word Conversation in Verse, it refers to the conversation model for a specific Agent. A Conversation has Topics that can either be hard-wired together in a flow or be disperse an unconnected.
  
• Topic - A collection of Replies which can be linked to Intentions that trigger the topic. Think of it as a real topic in a real conversation. When you have a real conversation it flows from topic to topic. Topics are the base structure of Conversations. Intentions are linked to topics so that Verse can keep track of where and which direction the Dialogue is going, plus provide response in the form of a Reply which is on topic.
  
• Intention - These are at the heart of the conversation model. An intention is much like in other chat agents. It represents the "speech act" which the user is communicating in a message. For instance, if the user asks: "Hey, can I have your phone number?" it could be automatically categorized under the intention "get.phone". In Verse, Intentions have structure which allows them to have levels, when a person requests something, usually the root intention I use in training is "get", then followed by what they want. There are many ways to ask for the same thing, so the training smooths out that chaotic space and allows for acute or more ambiguous sensing of intention. Intentions are slowly becoming standardized across all Agents for many different speech acts. The system currently can categorize for (recognize) around 245 intentions.
• Reply - A reply is what exists in topics. A reply can be spun (using standard text spinning syntax) and include variables or Attributes about the "self" or "other", including other objects like "time". A reply is usually linked to an Intention in a Topic so that it can be chosen as the response for any incoming message. Non-linked replies in a Topic can be fired when Intentions lead to the topic or when a "push" happens which means that the Agent basically "changes the subject", usually moving toward a goal.
• Listener - Will listen for a specific Intention and then cause data extraction to take place. Think of this as the ability to make a statement or ask a question, and based on the reply from the user, you can update or set a new Attribute of the user's Entity. If you have a Reply that asked: "Do you like animals?" you can set up a Listener that listens for the intentions #affirm or #negate. Based on which intention comes in the following replies, you can then set the attribute "likes.animals" to "yes" or "no" for the user's Entity. A listener can also include an expiry setting so that you can stop listening for an answer after a certain number of messages or a certain amount of time.
  
• Condition - Can be based on any of the variables in the system, including Topic data, Dialogue data, Entity data, Intention data, etc. A condition can be attached to a Reply and will cause it to be the chosen Reply over the other alternatives that match the incoming Intention in that Topic.
  
• Action - Allows for an action to be taken. Actions can be taken on the Verse server, with specific services built into the system, or can be included in API reply data so that a bot using Verse can take a specific action. Actions can be attached to replies such that when the reply fires, the action happens simultaneously. You can use actions to do things like send an email, send data back to your bot, trigger basically anything. Imagine it as a postback to external services.

I have been working hard on Eva and have completed the following...

• A Java Client API for Chat
  • Pretty versatile, allows for three different actions at the moment: adding an interface, adding an entity, and getting a reply.
  • Easy one-line configuration so you can get your bot using Verse in a couple of lines of code
  • Retention of configuration data
  • I also included the API key
• API Keys section of web app
  • You can request or view an API key for your user account and a specific Agent
• Chat Log on web app
  • Shows the incoming messages from chat Entities and the replies used broken up by topic, along with extra data about how the Agent chose the reply: intention found and topic used
• Personas section of web app
  • View Personas on your account (in specific chat agent)
  • Edit and Add Persona Attributes for each persona
    • TODO: Add cycles (free, hourly, daily) for scheduled changes to attribute values
      
    • TODO: Generate a Persona from an Entity
      
    • This also shows you a list of all "mentioned attributes" in replies from the conversation model
• Deleting Replies in the Interface is now possible
  
• Inline training in the Training Chat page
  • Now when an error happens (ie: can't find reply, topic, intention) it shows the solution
  • If a reply cannot be found, Verse shows an input for a reply in the found topic
  • If a topic cannot be found, Verse shows a topic input for the found intention
  • If an intention cannot be found, Verse asks for you to annotate the message by adding an intention
• Did some more training on live data
  • Since I finished the Java client, I tested out with Eva in a live chat situation, and since Eva's current conversation is set up to have getting a phone number or email as a goal, The system had about 70 different Dialogues with real people, and as a result has gotten 3 emails and 8 phone numbers. A larger portion of those 70 did not reply after the first reply because of natural causes, like they weren't on site, etc. On all of those, it was only a greeting, which Eva has a lot of training on, so it was handled with the correct reply, just the other chatter's disappeared. This training allowed me to also update the conversation model a bit, adding more Replies and linking more Intentions in the right Topics
• TODO: Entity data report
  • I'm currently working on this. It will be a nice interface that allows you to see and filter the data gathered by the Agent. There is already plenty of data to go through, so I'm thinking I'm going to go with a sort of "pivot chart" hybrid type of report... or just doing filters so that you can quickly get the data you want like (name, age, phone) where you require a phone to be part of the data.
• TODO: Listeners
  • Need to finish interface for listeners so new ones can be created easily and will start listening after a specific Reply is fired.
• TODO: Conditions
  • Need to finish interface for conditions. This will allow you to create and attach a condition to a Reply so that it will fire if the condition is met. Since a reply can forward to a Topic or fire an Action these sort of conditional replies can act as transitions to other topics, or even be used to require some data from the user.
• TODO: Intentional Math
  • I will reveal more about this later after it's finished, but let's just say that it will be a great way to set up sequential and contextual equations who's product is a final intention which was not previously part of the categorization. An example: #want.meet > #flirt = #finalize ... meaning that if a #flirt follows the user saying something about meeting, then include the #finalize intention in the process of replying. The #finalize intention could be one that has no training data, but is linked to some Topic in the Conversation.

Anyway, that's the update. I will have some screenshots of some of this stuff shortly. I have to blur out some of the pieces of the screens since this is real data I will be showing. I am on a roll, and thanks to everyone for giving me feedback and likes, etc. I have a great feeling about this project!

 

Screenshots as promised earlier

The first shot is the training chat with an example input message that generates an error with in-line solution.


I think these speak for themselves, but want to call attention to the quality of the intention categorization system. It is already pretty accurate, and the conversation model is mature enough to handle quite a few contingencies. It's interesting though how you almost don't have to even guide the conversation through Topic pushing, just replying as it is now, the conversation is driven by the user toward the goal, the Agent only provides the proper queues in replies.

Also in the last shots you can see how Attributes are setup for a Persona of Eva and how they can be spun text as well.

 

Hahaha, well someone's going to have to run to the liquor shop! These drinks don't make themselves!