Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
How to Train and Build a Conversational News Chatbot (github.com/tzano)
108 points by tzano on May 4, 2018 | hide | past | favorite | 8 comments


Future parents will teach their children to never talk to computers.


This is cool, I bookmarked it to check out later.

I’ve always been curious about chatbots. I feel like in the right situations they can provide a good customer experience, but I’ve never seen them as being able to stand-alone as their own product.


I feel like when chatbot technology can be used with voice, it could be a gamechanger. Sort of like when you watch sci-fi like Star Trek, The Expanse, Iron Man, and how they interact with the AI. They're able to ask complex questions to a computer that is able to interpret it- like audio SQL calls in conversational form.

Or the most popular application would be AI to help people feel less lonely or make their lives easier like in the movie Her or the book Ready Player One.



This is neat but there’s a big gap between natural language search and conversation.


So this is an interesting topic for me, because I once tried to write a "conversational" Hubot plugin for a hackathon[1]

To me, "conversational" means the bot must inform its responses from the context of previous messages. When we decided to write this bot in 24 hours, we had no idea just how much Hubot does not want to be conversational! There is a plugin that you can use called Redis Brain that hardly makes things any easier, at the time we wrote this bot at least there were zero "conversational" plugins for hubot, and I wouldn't be surprised if there are still none.

An example of a simple conversation that Hubot is absolutely not designed to be able to have (and neither is this news bot, although it's very neat... edit: well, I've looked again and I could be wrong, it looks like the news-bot actually does maintain some state per-user...):

"Hey tebot, I have an idea"

"> OK, tell me your idea"

"Let's eat a whole wheel of cheese"

"> OK, I've recorded your hackathon idea"

So that's a basic conversation. In order to have a conversation (or, as a slack bot, potentially having multiple concurrent conversations with several different users) the bot needs to have a template for at least one or several possible conversation types, probably some trigger condition to begin the conversational context, and a path to progress the context towards an end (like joining or leaving the Hackathon team, or advertising that your team has a free spot for another member, or)

... did you get the idea yet that this bot is meant to be used at hackathons to help organize any independents into teams without that awkward socializing? To be a platform for solving problems like "I have experience in these three languages and I don't know anyone here, but I would like to work with a team that uses one of these languages."

Not to get too far off-topic but we were solving a problem that we actually had; this was actually the first any real Coffeescript experience that any of us had, as we formed a team not realizing that none of us were expert in any of the same languages!

Anyway I'm not trying to crap on what you've done but the headline grabbed me, because I remember how just getting the bot to have the most basic of conversations was so hard that it took easily the first 12 of 24 hours just to get a basic PoC, as it seems that was something the bot framework was just not designed to do!

[1]: https://github.com/kingdonb/hubot-malta


How do you install/run it from scratch?





Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: