This invention discloses an automatic simplification method and
system for psychological scales based on
word embedding and
deep learning. It primarily employs
word embedding technology to semantically vectorize each item and all item texts of the original psychological scale, obtaining item vectors and a set vector of all items. The
semantic similarity between each item vector and the set vector is calculated and defined as the semantic
discriminant of the corresponding item. A simplified scale is generated by
ranking the items according to their semantic
discriminant scores. The simplified scale is then administered to the target group, and the item
discriminant, reliability, and validity indices are analyzed to complete the validation of the reduced psychological scale. This invention combines semantic modeling techniques from
natural language processing with nonlinear modeling from
deep neural networks, overcoming the problem that traditional methods cannot consider item-related aspects semantically. It ensures the integrity of the scale's theoretical dimensions and can be widely applied to various scenarios, including the construction of large-scale psychological assessment systems for different groups and the development of intelligent
psychological testing tools.