This paper proposes a method to improve factor models with text information based on the factor model construction process. We analyze the model using the KJ method. However, because this model often has many unidentifiable latent variables due to the lack of observed variables, we need to filter and change the model. To decrease the number of unidentifiable latent variables, we use principal components analysis for keywords in text data and the principal component scores are allocated to the unidentifiable latent variables. Further, we could confirm that GFI (goodness-of-fit index) and AGFI (adjusted goodness-of-fit index) are improved by using the factor model for an online-game analysis.