# How to manage the influence of variables for lsa cosine?

I'm building reccomendation system for movies and the following data

original_language popularity vote_average clastergroups companies_group sentiment        Horror  Science Drama
5                 7.284477   5.7          2             1                -7.571429e-02   0       0       0
5                 10.137389  7.8          0             0                -1.937500e-01   0       0       1
5                 9.026586   6.5          2             9                -1.394737e-01   0       0       0
9                 9.822423   7.6          3             1                2.272727e-02    0       1       0
5                 7.137117   6.9          1             8                1.191176e-01    0       0       0
16                2.030174   7.2          1             0                7.380952e-02    0       0       0
5                 7.438934   6.3          1             16               8.108108e-02    0       0       1
5                 11.970205  7.2          0             0                -1.041667e-01   1       0       0


With the function lsa::cosine(t(as.matrix(data))) I want to find similarities between movies and show the result to the user. But the problem is that the model is not very accurate.

In order to fix it, I want to put some significance to every variable. For example, genres of movies (science, drama etc.) are the most significt so I want my model to consider genres first. clastergroups are the least significant, I want to consider it a bit, but not as much as genres. How I can do this?