Spin networks arise in many areas of physics and are a key component in both the canonical formulation (loop quantum gravity) and the path-integral formulation (spin-foam gravity) of quantum gravity. In loop quantum gravity the spin networks are used to construct a countable basis for the physical Hilbert space of gravity. The basis states may be interpreted as gauge-invariant wavefunctionals of the connection. Evaluating the wavefunctional on a specific classical connection involves embedding the spin network into a spacelike hypersurface and finding the holonomy around the network. This is equivalent to evaluating a ``g-inserted'' spin network (a spin network with a group action acting on all of the edges of the network). The spin-foam approach to quantum gravity is a path-integral formulation of loop quantum gravity in which the paths are world-histories of embedded spin networks. Depending on the spin-foam model under consideration the vertex amplitude (the contribution a spin-foam vertex makes to the transition amplitude) may be represented by a specific simple closed spin network. The most important examples use the 6j-symbol, the 15j-symbol, and the Riemannian 10j-symbol. The semiclassical treatment of spin networks is the main theme of this dissertation.

To show that classical solutions of general relativity emerge in the appropriate limits of loop quantum gravity or spin-foam gravity requires knowledge of the semiclassical limits of spin networks. This involves interpreting the spin networks as inner products and then treating the inner products semiclassically using the WKB method and the stationary phase approximation. For any given spin network there are many possible inner product models which correspond to how the spin network is ``split up'' into pieces. For example the 6j-symbol has been studied in both a model involving four angular momenta (Aquilanti et al 2012) and a model involving twelve angular momenta (Roberts 1999). Each of these models offers advantages and disadvantages when performing semiclassical analyses. Since the amplitude of the stationary phase approximation relies on determinants they are easiest to calculate in phase spaces with the fewest dimensions. The phase, on the other hand, is easiest to compute in cases where all angular momenta are treated on an equal footing, requiring a larger phase space.

Surprisingly, the different inner product models are not related by symplectic reduction (the removal of a symmetry from a Hamiltonian system). There is a connection between the models, however. On the level of linear algebra the connection is made by considering first not inner products but matrix elements of linear operators. A given matrix element can then be interpreted as an inner product in two different Hilbert spaces. We call the connection between these two inner product models the ``remodeling of an inner product.'' The semiclassical version of an inner product remodeling is a generalization of the idea that the phase space manifold that supports the semiclassical approximation of a unitary operator may be considered the graph of a symplectomorphism. We use the manifold that supports the semiclassical approximation of the linear map to ``transport'' features from one space to another. Using this transport procedure we can show that the amplitude and phase calculations in the phase spaces for the two models are identical. The asymptotics of a complicated spin network, and thus the fundamental amplitudes of loop quantum gravity and spin-foam gravity, may be computed by first setting up an inner product remodeling and then picking and choosing which features of the calculation to perform in which space.

In this dissertation we first introduce the remodeling of an inner product and the semiclassical features of the remodeling. We then apply the remodeling to the well-studied cases of the 3j-symbol and the 6j-symbol. Finally we explore how the remodel procedure applies to more complicated spin networks such as the 15j-symbol and the g-inserted spin networks of loop quantum gravity.