Don't put words in my mouth you troll. You don't understand anything.

Bayesian networks are acyclic, directed graphs. Markov networks are cyclic, undirected graphs. Markov networks are a better representation of the problem because words in a sentence are not distinct. If your Bayesian network uses words as their nodes, then you could not construct a sentence which had a word more than once without potentially having cycles. Because then your graph would have to have a cycle. Take the previous sentence as an example. I used the word "have" twice. In a Bayesian network, this would mean the nodes have and to would form a cycle as demonstrated below.




This is not a directed acyclic graph. Which means it is not a bayesian network. You would have to duplicate the "have" node (and many others) if you wanted a bayesian network to work like this. Or, you could remove the data replication and not be able to generate sentences with the same word twice. This is why Markov networks are better. Because they are undirected cycle graphs. You can generate a markov chain of words which more closely resemble human speech patterns. Bayesian networks are used for plotting statistical representations of independent cause and effect relationships. Grammar does not fit that.

Please. PLEASE. Shut the fuck up. You are providing nothing of value to any conversation you are in. And if you are just trolling, all you are doing is making yourself look like an idiot.

This post was edited by Minkomonster on Mar 18 2016 11:14am

ugh such a technicality why do i keep looking at this thread
you said you wanted to look at previous sentences not just go with an independent state
you're right you're going to have to duplicate the "have" node for any possible use of the word, in your model, with a directed model you simply say

would , could

go to
have
go to

1, 2, 3, 4

have only goes to 1 or 2 when preceded by would, only goes to 3 or 4 when preceded by could

if you don't direct it or take previous states into account you can go any fucking where

i'm not sure why you think that conditional independence and direction are unsuited for grammar or why you want to use an directed or cyclical process since one of the most basic rules of grammar are that subjects are followed by predicates but fuck man

all i fucking did was try to explain that he needs to use a goddamn graphical model instead of a chart i don't think either of these models will completely solve the problem



it's not really similar at all tbh language processing output is way too much work
i want to do the opposite of what you're doing really and give a set 70 possible responses come only when relevent statements are read

This post was edited by Ideophobe on Mar 18 2016 01:22pm

Then why reinvent the wheel, use chatscript or AIML.

pretty much what i'm doing
i'm going with pircbot api for java, it has any possible chatbot functionality i could need, with alot of cool extra features for a complete irc bot
i'll have to write all the line read logic myself i think for my idea to work but it should be pretty straight forward

This post was edited by Ideophobe on Mar 18 2016 03:55pm

It is. I wrote a dynamic call back based irc bot framework a year ago. Easy stuff.

http://www.sjtrek.com/trek/rules/
with some "Welcome to Qwarks" "what can i get ya" bartender talk to turn a channel into qwarks bar and quote rules of acquisition at people whenever they say relevant stuff

Going to probably write a wrapper around cleverbot and mitsuku tomorrow so I can see if I can train a sentence based markov chain off them. I'll need to leave it running on my server for about a week I'd imagine before I can get any sort of tangable database setup to test with.

Is it possible to do some kind of fuzzy string matchi g in SQL? I know if you use the LIKE clause and the % operator, but I am not sure how I would attempt to chop the sentence up.

For example if I had "hello my name is abduct" a good fuzzy match would be "hello %", but how to logically convert one to the other is another thing.

Confirmed