Current vector space model, for instance TF/IDF, has not yet taken into account the relations between terms; it only combines the term frequency in a document and the inverse document frequency in whole database to identify importance-score (weight) of a term respect with the document. Here we discover lexical relations among terms in the document to improve the vector space model TF/IDF. The weight generated from TF/IDF for each term, which is improved by lexical relations among related terms in the document. We evaluate the proposed method using documents selected from Wikipedia. The result shown that the proposed method is significant effective.