Hi Yufan,
MolProbity developer here. You have asked a very reasonable question.
The "angle" reported in the C-beta position outliers is a dihedral defined by the residue's N-CA-idealCB-modeledCB positions. If you are looking at the markup kinemage, that's N-CA-purple sphere-modeled CB. This dihedral indicates the direction of the CB deviation relative to the ideal CB position and the geometry of the residue. You can imagine the CB deviation in a radial coordinate space, with the ideal position at 0, and the modeled position at some distance and angle.
Generally speaking, the deviation distance is the most intuitive and interpretable measure, but the angle - the angular position of the modeled atom in radial space - may be useful in some applications. On the MolProbity website, after validating a structure, you can find a "Cbeta deviation scatter plot" that uses this radial representation of CB geometry.