The growing number of constructs in behavioral research presents a problem to theory integration, since constructs cannot clearly be discriminated from each other. Recently there have been efforts to employ natural language processing techniques to tackle the construct identity problem. This paper compares the performance of the novel word-embedding model GloVe and different document projection methods with a latent semantic analysis (LSA) used in recent literature. The results show that making use of an advantage in document projection that LSA has over GloVe, performance can be improved. Even against this advantage of LSA, GloVe reaches comparable performance, and adjusted word embedding models can make up for this advantage. The proposed approach therefore presents a promising pathway for theory integration in behavioral research.