Training needs reassessment at times requires extra time and effort in asking questions, doing surveys, or doing interviews. By extracting themes and key concepts from trainees' online narrative feedback after training, this research explored a mixed way of human and computational analysis on trainees' perception of training deficiencies to weave individual narrative feedback as supplementary features into a semantic knowledge map of training needs. The case study was conducted in an international company with over 100000 employees at dispersed locations. This research provides a method for any specific companies with unique characteristics to construct a dynamic knowledge map, leading to a possible automatic improvement of the training program.