Key Contexts of Thematic Words

According to the type of input, this T-LAB can be used for two different purposes:

A) to extract lists of meaningful context units (i.e. elementary contexts or short documents) which allow us to deepen the thematic value of specific key terms;
B) to extract the context units which are the most similar to sample texts chosen by the user.

Here are some explanations for the two above cases.

Case (A)

Unlike Concordances, which allows the extraction of all elementary contexts in which the selected key words are just present (occurrences), and unlike Word Associations, which allows the extraction of all elementary contexts in which the selected key words are in matching pairs (co-occurrences), this tool allow us to extract the elementary contexts in which each selected key word is associated with other words (multiple co-occurrences) defining its thematic field.

It works in the following way:

1- the user chooses a thematic word "X" (see "Muslim" below);
2- T-LAB proposes a list of words (max. 50) whose co-occurrence values with "X" are the most significant;
3- the user can remove irrelevant items from the list provided (just double click each item);
4- after clicking 'Extract Key Contexts' T-LAB assumes that the user list is a query vector and computes its association indexes (i.e. cosine coefficients) with all the elementary contexts of the corpus or of the selected corpus subset.

The output provided, both in HTML and TXT format, contain a list of the most significant key-contexts of "X", listed according to the descending order of their association indexes.

The 1-4 steps can be reiterated for "n" thematic words.

Case (B)

It works in the following way:

1- the user copy/paste a text (Max 5,000 characters) in the appropriate box;
2- after clicking the 'extract key contexts' button, T-LAB transforms the input text into a query vector and computes its association indexes (i.e. cosine coefficients) with all the elementary contexts of the corpus or of the selected corpus subset.

The output provided, both in HTML and TXT format, contain a list key-contexts which are the most similar to the input text.

N.B.: In such a case the similarity measure doesn't take into account multi-words the strings of which, either with or without the underscore ('_') character, do not correspond to the analysed text.

The above 1-2 steps can be reiterated for "n" sample texts.
.