In some embodiments, applying the graph convolution operation based on the weights comprises: applying, by one or more processors, the graph convolution operation based on at least one of the following: a first set of weights specific to types of the relationships indicated by the edges between the set of neighbor nodes and the given node, and a second set of weights each determined based on the numbers of nodes having edges connected with respective neighbor nodes in the set of neighbor nodes and the number of nodes in the set of neighbor nodes.

In some embodiments, applying the graph convolution operation comprises: applying, by one or more processors, a first graph convolution on the set of neighbor nodes based on the first weights to obtain a first intermediate representation; applying, by one or more processors, a second graph convolution on the set of neighbor nodes based on the second weights to obtain a second intermediate representation; and combining, by one or more processors, the first and second intermediate representations to obtain the word representation.

In some embodiments, the edges comprise a first directed edge from the given node to a further node in the set of neighbor nodes and a second directed edge from the further node to the given node, the first directed edge indicating a first relationship from a given word represented by the given node to a further word represented by the further node, the second directed edge indicating a second relationship from the further word to the given word. In some embodiments, the first set of weights comprises a weight specific to a type of the second relationship instead of to a type of the first relationship.