Think about it, if you are in a bad mood you are more likely to use emojis like 😔 and 😟 and 😒, while if you are happy you probably use emojis like 😘 and 😄 and 😁 (and maybe ❤️ if you're lucky enough). I think if you parse the use of emojis in a persons recent messages, you could calculate to a high degree of certainty what psychological state that person is in at the moment.
An idea for the calculation could be to assign each emoji to a value, kind of like how simple versions of sentiment analysis works. You could then add up the scores of all emojis and come to a conclusion.
Another idea would be to throw some aspect of machine learning into the mix. If you for every calculation let the subject fill in how he or she actually feel, you could use this data to modify the values of emojis, and maybe come to conclusions for how different combinations of emojis affect the result. I imagine that this is how more complex sentiment analysis algorithms work, like the one used by the AlchemyAPI.
Right now I have a lot of projects up my sleeve, with Surview and Startup Camp, CETAC, and me and my buddy Jonathan's side project Mailbase, so I don't really see myself implementing this idea anytime soon. Hopefully though, I will find some time in a not too distant future.
Take it easy.