What you can do is to show you sentences that are as vague as possible but can get answers from a large number of words.

The QQ chat robot we see is actually the same thing ~ you say a word, it will first filter out modal particles, such as ah, ah, ha, such as the weather is really good today. Ah, get rid of this word

Then find the key word, weather, good.

A large number of sentences are written in the system in advance, and one of the characteristics of * * * is ambiguity. It looks for a sentence that can be associated with both weather and good words: yes, it is.

With this sentence, you can answer "good weather" or "bad weather". Then you feel smart and will answer you.

The other is silly perfect matching, which is to save as many sentences as possible in the database and answer them automatically once they match. You speak so many Chinese characters that there are many different ways to combine them at will. Of course, if the database is large enough, this is not a problem.

As for how to write the paper, it's quite troublesome, but I think it might be good to play by myself. It can be written around the following aspects: function, prospect, principle (the two I mentioned above, there may be others) and future development direction (expanding database and fast search).