I am currently working on emotions for my chatbot, my current system works like this: there are different emotions (and the system can accomidate varying numbers of emotions). They are on a scale, with two emotions on either end and a "no mans land" in the middle: happysad for instance is happy when it is > 0.5 and sad when it is < 0.5) It is split into two different functions: doemotion and getemotion. Doemotion's role is to recalculate emotions based on user input (eg. lol would increase the "happysad" emotion) Getemotion's role is to calculate (via thresholds, such as happysad > 0.95 or happysad < 0.20) which (if any) emotional messages are to be displayed (like :-) :( or I'm Heigh!) 1. Should i randomly change the emotions? (either per input or ?) 2. Is there a better way to do this/achive this effect? in my doemotion function, when changing the values, i use emotion = clamp(emotion * newvalue * 2, 0 ,1) Is this a wize thing to do? From, Nice coder

hi,
well, if it was me, I would use perceptrons that output a percent value rather than a bool, and I would make each perceptron take as many factor as I need, including the defaut character of the chatbot, and a random value. and then I would rather use an update function, that calculate that percent of each perceptron, and decide what emotions to use...
again, that's just what I would think according to my own logic.
hope it helps.

Dark Metamorphosis- How would i train the perceptrons?
Currently my "problem" is that my bot either:
1. sounds like its under the influence of alcohol or
2. Sounds a little tamer, but very "emotional"

Currently my values are a little far of 0.5,
Here are the two lines (from the data file)

Good|Bad|Hate|love|LOL
0.85|0.2|0.1|0.99|0.70

Now if i go LOL twice, it gives a "Very happy" responce. if i go hate once i get a "Very sad" responce. Should i move these values coser to 0.5?

Nothing here is a bool, unless i've got i heigh bot which had its happyness clamped. Every quentity is a double, so i could have quite a few extra decimal places...

From,
Nice coder

Try doing some research on previous work. I've seen quite a lot of papers on reactive facial expressions for VR agents which would be very applicable to you (i.e. replace facial expressions for emoticons).