Quantitative analysis of Raman data is always affected with a lot of caveats since the intensities are not exclusively dependent on the existence of phases but also the local crystal quality which is why Raman is more of a quality instrument for many materials. For some systems there are characteristic peak intensity ratios that allow to make quantitative statements about the material quality, so you can check if that's the case for you (I'm not a cement guy, I do carbon materials) but for a quantitative phase analysis I would rather suggest powder diffractometry or something like that.

I'd have to know more about what it is you're doing. I've used Raman data for material science (complex compound clay materials especially, some glass, some metals) and compared it with other methods (numerous: XRD, XRF, PIXE, NAA, ICPMS, SEM, various spectrometry/spectroscopy, various chemical and other physical tests...). I use a lot of principal component analysis. basic plotted ternary diagrams for what I do can sometimes do the trick. Linear regressions, blah, blah, blah. However, what I'm looking for in ceramic materials are different sources; therefore, what's really important for my analyses is consistent groupings or clusters among samples no matter what technique and the identification and removal of what is generally high variation among all, versus specific atomic, molecular, mineralogical, etc. that are consistent within groups and vary between groups. Hope this possibly helps. More info on what you're actually doing and I may have some more advice. Thanks. I do like using the Raman and Raman data, however. Been a good method for our purposes thus far.

Quantification for Raman is challenging, as it is a scattering technique. The orientation of crystals also plays a vital role in as well as the polarization dependence of your spectrometer. You have to convert a linear polarized laser beam to the circular polarized. It eliminates the dependence of crystals orientation in your sample because they are random. Besides, it would be useful to add scrambler to the optical path of the Raman signal to eliminate the polarization dependence of grating efficiency. After that, just run either multipoint measurements or map of a certain area and take the average spectrum. Compare each spectrum with average, and you will find what Std you would expect. This will be a reference point for comparison with other samples. If you have a couple of reference samples, you may try to use PLS. Peak fitting also may work but not always, and it requires to subtract background, which may be different for different samples/spots, etc.

After deconvolutions find the area under the curve and compare the areas with each other, this may help