The embodiment of the invention relates to a method and a device for grouping user portraits, which relate to the data analysis field and can enable user portrait grouping to be more stable and accurate. The method comprises steps: based on the label of each user portrait, a label network G (V, E) is built, wherein V presents a vertex set in the label network, each vertex presents one user portrait, E presents an edge set in the label network, and each edge presents a connection line of two vertexes corresponding to two user portraits having at least one common label; based on the label network, a line graph network G' (V', E', W') is built, wherein V' presents a vertex set in the line graph network, each vertex presents one edge in the label network, E' presents an edge set in the line graph network, each edge presents a connection line of two vertexes corresponding to two edges with the common vertex in the line graph network with the label network, and W' presents a weight set for edges in the line graph network; community division is carried out on vertexes in the line graph network; and a community division result is converted to a user portrait grouping result.