Your question is interesting.. , some thoughts  hopefully of use..

1 Build a matrix  a matrix :  [P]=

             cols    word1    word2 ... word 100

row person 1

row person 2

'rowperson  100

2. Do PCA

3. Obtain scores for important( eigen values( vectors)

Cheers

It depends on exactly what you mean by "idea". 

Narasim's suggestion of doing PCA (or a similar factor analysis) is a good one. 

I recommend doing some research on latent dirichlet allocatlon (LDA), which is often used in natural speech processing tasks with similar goals. 

Thanks to both of you :) I appreciate both answers. 

Question please.... what could be cell values if i use that matrix for PCA? 

hi

The number of times that word appears for each row (person).

Cheers