Processing sequential interaction data

A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:

• obtaining a dynamic interaction graph constructed based on a dynamic interaction sequence, wherein:
• the dynamic interaction sequence comprises a plurality of interaction feature groups corresponding to a plurality of interaction events arranged in chronological order;
• each interaction feature group comprises a first object, a second object, and an interaction time of an interaction event that involved the first object and the second object;
• the dynamic interaction graph comprises a plurality of nodes including, for each interaction feature group, a first node that represents the first object of the interaction feature group and a second node that represents the second object of the interaction feature group;
• and
• for at least a portion of the plurality of nodes, the first node or second node is connected, by edges, to two leaf nodes that represent two objects in a previous interaction feature group corresponding to a previous interaction event in which an interaction object corresponding to the first node or the second node was involved;
• determining, in the dynamic interaction graph, a current sequence corresponding to a current node to be analyzed, wherein the current sequence comprises one or more nodes within a predetermined range reachable from the current node through edges and positional encoding of each of the one or more nodes relative to the current node in the dynamic interaction graph;
• and
• inputting the current sequence into a Transformer-based neural network model, wherein the neural network model comprises an embedding layer one or more attention layers;
• obtaining, at the embedding layer, N embedded vectors based on, for each of the one or more nodes, node features comprising attribute features of an object represented by the node and the positional encoding of the node;
• combining, at each attention layer, input vectors based on a degree of correlation between N input vectors obtained from a preceding layer to generate N output vectors, wherein the N input vectors to a first attention layer of the one or more attention layers comprises the N embedded vectors, wherein combining the input vectors based on the degree of correlation between the N input vectors obtained from the preceding layer to obtain the N output vectors comprises:
• using a first transformation matrix Q, a second transformation matrix K, and a third transformation matrix V to transform each of the N input vectors to obtain a first intermediate vector, a second intermediate vector, and a third intermediate vector corresponding to each of the N input vectors;
• for any ith input vector in the N input vectors, determining a degree of correlation between the ith input vector and each input vector based on a dot product operation performed on the first intermediate vector corresponding to the ith input vector and each second intermediate vector corresponding to each input vector;
• performing a weighted combination on the third intermediate vector corresponding to each input using the degree of correlation between the ith input vector and each input vector as a weighting factor to obtain an ith combination vector corresponding to the ith input vector;
• and
• obtaining the N output vectors based on each combination vector corresponding to each input vector;
• and
• determining, by the neural network model, a feature vector corresponding to the current node based on the N output vectors generated by the one or more attention layers.

A computer-implemented system, comprising:

• one or more computers;
• and
• one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing one or more instructions that, when executed by the one or more computers, perform one or more operations comprising:
• obtaining a dynamic interaction graph constructed based on a dynamic interaction sequence, wherein:
• the dynamic interaction sequence comprises a plurality of interaction feature groups corresponding to a plurality of interaction events arranged in chronological order;
• each interaction feature group comprises a first object, a second object, and an interaction time of an interaction event that involved the first object and the second object;
• the dynamic interaction graph comprises a plurality of nodes including, for each interaction feature group, a first node that represents the first object of the interaction feature group and a second node that represents the second object of the interaction feature group;
• and
• for at least a portion of the plurality of nodes, the first node or second node is connected, by edges, to two leaf nodes that represent two objects in a previous interaction feature group corresponding to a previous interaction event in which an interaction object corresponding to the first node or the second node was involved;
• determining, in the dynamic interaction graph, a current sequence corresponding to a current node to be analyzed, wherein the current sequence comprises one or more nodes within a predetermined range reachable from the current node through edges and positional encoding of each of the one or more nodes relative to the current node in the dynamic interaction graph;
• and
• inputting the current sequence into a Transformer-based neural network model, wherein the neural network model comprises an embedding layer one or more attention layers;
• obtaining, at the embedding layer, N embedded vectors based on, for each of the one or more nodes, node features comprising attribute features of an object represented by the node and the positional encoding of the node;
• combining, at each attention layer, input vectors based on a degree of correlation between N input vectors obtained from a preceding layer to generate N output vectors, wherein the N input vectors to a first attention layer of the one or more attention layers comprises the N embedded vectors, wherein combining the input vectors based on the degree of correlation between the N input vectors obtained from the preceding layer to obtain the N output vectors comprises:
• using a first transformation matrix Q, a second transformation matrix K, and a third transformation matrix V to transform each of the N input vectors to obtain a first intermediate vector, a second intermediate vector, and a third intermediate vector corresponding to each of the N input vectors;
• for any ith input vector in the N input vectors, determining a degree of correlation between the ith input vector and each input vector based on a dot product operation performed on the first intermediate vector corresponding to the ith input vector and each second intermediate vector corresponding to each input vector;
• performing a weighted combination on the third intermediate vector corresponding to each input using the degree of correlation between the ith input vector and each input vector as a weighting factor to obtain an ith combination vector corresponding to the ith input vector;
• and
• obtaining the N output vectors based on each combination vector corresponding to each input vector;
• and
• determining, by the neural network model, a feature vector corresponding to the current node based on the N output vectors generated by the one or more attention layers.