Natural language processing in bot automation

[social engineer]

Ok, I may ruffle some feathers posting this but is anyone incorporating NLP into their bots yet.

I ask this purely as for a recent project I had to incorporate an NLP into a project so that users would not bypass the premium version of the project.

In the end instead of using the node library natural (https://www.npmjs.com/package/natural) i decided to use pythons Natural language toolkit (https://www.nltk.org/) as an API whereby all private messages would be sent and processed by my script and sent back if it deemed the user wasn't trying to get bypass premium features (This may sound strange without explaining the niche).

Anyway, this same NLTK API could with alterations be used to make bots appear humans.

Think about an ewhoring bot.

It messages a user on POF (or more likely with pof it waits for someone to message it)
it sends the sent message to the NLTK API.
The NLTK runs sentiment analysis and other tools to work out what the message is refering to and then sends back the analysis to the bot.
The bot then answers the questions (Question: whereabouts in london do you live? Bot Answer: I live in Kensington) and then sends questions with easy to reference responces.
The bot then sends them onto the lead (POF doesn't allow me to send risky images but you can see them on my profile here)

E-whoring dosn't have to be the only use NLP could be used for responding to Instagram messages human-like or even to craigslist ads.

Its one of those subject which I'm surprised is not talked about more on this board.

I'm no expert by any means on the subject. I've just dabbled in it to solve a problem (a rudimentary self-learning spam filter basically), but it seems to be very powerful stuff.

 

[PetroK]

Insightful points on your remarks as AI on marketing bots is coming soon.

 

[MaxGuard]

Really great points as AI can be next version of spinners to create or spun content and get great results. just my two cents.

 

[xpizzle]

I've been working on a bot that can do photo classification and identify the objects within the photo to then generate the comment, still based on a template though. This is very niche specific though and I have been training based on objects you would see in this niche.

I wanted to generate comments based on other peoples comments, but with a lot of bots commenting as well and comments such as "cool photo" doesn't really help generate a very good comment.

 

[bangando]

it seems like spam emails is the most used example in those nlp tutorials or books. I've played with it last year and i was trying to target or locate users based on words or tags they've used. But it's like you said, it needed a lot more refining since using a location tag/word on social media doesn't mean that you actually live in that specific area

 

[TechShadow]

Do people actually fall for "eWhoring" ?

 

[Neareo]

I might be biased, as our chatbot platform [neareo(dot)com] doesn't support Natural Language Processing, but I think NLP is overrated. It is much more efficient currently to use diagram-based scenarios and text-to-speech to avoid inadequate bot answers like "I see" every now and then. Consider a diagram-based scenario as a funnel that leads the visitor to a specific target that YOU define, versus some open/non-linear scenario where the user will end in some dead end. Where would you prefer to send him/her as a marketer?

 

[Cv7]

I might be biased, as our chatbot platform [neareo(dot)com] doesn't support Natural Language Processing, but I think NLP is overrated. It is much more efficient currently to use diagram-based scenarios and text-to-speech to avoid inadequate bot answers like "I see" every now and then. Consider a diagram-based scenario as a funnel that leads the visitor to a specific target that YOU define, versus some open/non-linear scenario where the user will end in some dead end. Where would you prefer to send him/her as a marketer?

I did read that NLP had an average 60-70% precision rate so you might end up with a long-drawn conversation (when I suppose your objective should be conversions and not just robot-human babbling). Conversational interfaces on the other hand are much straighter to the point but trying to get the user/victim on the other end to click your link or do X within 2 minutes of starting the convo might arise suspicion.

Im an ideal scenario you could use NLP to start off the conversation and program a simple timer of maybe X minutes until the CI starts being used for the responses in tandem (NLP + CI hybrid).

Example: 10 min of back and forth generic NLP convo bs, then AI comes through with:
" so do you wanna {option1} or {option2}? "
OR
" you wanna add me on {option1} or maybe do xx on {option2} later when I'm free? "

Fill the brackets with whatever links or sites containing your ads/leads and I could see those conversations being very monetizable.

I assume this is what your bot does?

There must be lots of libraries/databases with troves of refined data for conversational AI? I'm not well versed in the subject but its definitely something I'll learn more about in the future.

 

[WhateverHat]

Do people actually fall for "eWhoring" ?

Looks like people who are already on such sites, made their mind

 

[Miraz0]

Do people actually fall for "eWhoring" ?

Many many people do.... And "eWhores" (mostly mans now...) makes good money...

 

[Neareo]

Example: 10 min of back and forth generic NLP convo bs, then AI comes through with:
" so do you wanna {option1} or {option2}? "
OR
" you wanna add me on {option1} or maybe do xx on {option2} later when I'm free? "

Fill the brackets with whatever links or sites containing your ads/leads and I could see those conversations being very monetizable.

Interesting thought. But I see one downside, you might loose the visitor/lead before the timer gets hit