doc: clarify the effect of loop edges in Graph.similarity_inverse_log…

…_weighted()

src/_igraph/graphobject.c

@@ -15790,6 +15790,15 @@ struct PyMethodDef igraphmodule_Graph_methods[] = {
    "Each vertex is assigned a weight which is 1 / log(degree). The\n"
    "log-weighted similarity of two vertices is the sum of the weights\n"
    "of their common neighbors.\n\n"
+   "Note that the presence of loop edges may yield counter-intuitive\n"
+   "results. A node with a loop edge is considered to be a neighbor of itself\n"
+   "I{twice} (because there are two edge stems incident on the node). Adding a\n"
+   "loop edge to a node may decrease its similarity to other nodes, but it may\n"
+   "also I{increase} it. For instance, if nodes A and B are connected but share\n"
+   "no common neighbors, their similarity is zero. However, if a loop edge is\n"
+   "added to B, then B itself becomes a common neighbor of A and B and thus the\n"
+   "similarity of A and B will be increased. Consider removing loop edges\n"
+   "explicitly before invoking this function using L{Graph.simplify()}.\n\n"
    "@param vertices: the vertices to be analysed. If C{None}, all vertices\n"
    "  will be considered.\n"
    "@param mode: which neighbors should be considered for directed graphs.\n"