Graph search and visualization for fraudulent transaction analysis
In an embodiment, a process for graph search and visualization includes receiving a query graph, and calculating one or more vectors for the query graph, where the one or more vectors each identifies a corresponding portion of the query graph. The process includes identifying one or more graphs similar to the query graph including by comparing the calculated one or more vectors for the query graph with one or more previously-calculated vectors for a different set of graphs and outputting the identified one or more similar graphs. The comparison with the previously-calculated vector(s) may be based on previously-calculated vector(s) processed by grouping the one or more vectors into at least one group of vectors, identifying a representative graph for each of the at least one group of vectors; and storing the at least one group of vectors and a respective identified representative graph.
1. A system, comprising:
one or more processors configured to:
receive data, wherein the data includes transaction data entries;
generate one or more graphs using the received data;
calculate one or more vectors for the one or more graphs, wherein each of the one or more vectors represents a respective graph embedding corresponding to at least a portion of the one or more graphs, wherein the calculated one or more vectors are aggregated with one or more previously determined graph embeddings corresponding to previous transaction data entries;
determine that the one or more vectors belongs to at least one group of vectors based at least on the one or more vectors matching a data pattern associated with the at least one group of vectors including by building a hierarchy of clusters based at least on the aggregated one or more vectors with the one or more previously determined graph embeddings; and
identify a representative graph for each of the at least one group of vectors; and
a memory coupled to at least one of the one or more processors and configured to store the at least one group of vectors and a respective identified representative graph.
2. The system of claim 1 , wherein the one or more processors are configured to calculate the one or more vectors including by being configured to calculate the one or more vectors for the one or more graphs, each vector including a component that is a percentage of nodes of each node type in a corresponding graph.
3. The system of claim 1 , wherein the one or more processors are configured to calculate the one or more vectors including by being configured to perform unsupervised learning to calculate the one or more vectors for the one or more graphs, each vector including a first component that is a unique descriptor of a corresponding graph based on a topology of the corresponding graph and a second component that is a percentage of nodes of each node type in the corresponding graph.
4. The system of claim 1 , wherein building the hierarchy of clusters includes using successively aggregating learned node representations into clusters to obtain a single representation of each of the one or more graphs.
5. The system of claim 1 , wherein the one or more processors are configured to group the one or more vectors into the at least one group of vectors and to identify the representative graph for each of the at least one group of vectors including by being configured to use a clustering model.
6. The system of claim 5 , wherein at least one of model performance metrics and model diagnosis analysis is determined for the clustering model.
7. The system of claim 6 , wherein the clustering model is tuned in response to user input in response to the model performance metrics and the model diagnosis analysis.
8. The system of claim 1 , wherein:
the one or more processors are configured to update the at least one group of vectors including by:
receiving new data, wherein the new data includes transaction data entries;
generating one or more new graphs using the received new data;
calculating one or more new vectors for the one or more new graphs, wherein the one or more new vectors each identifies a corresponding portion of the one or more new graphs; and
updating the at least one group of vectors based on the calculation of the one or more new vectors; and
the memory is configured to store the updated at least one group of vectors.
9. The system of claim 1 , wherein the calculated one or more vectors includes a list of vectors corresponding to a set of data smaller than the received data.
10. The system of claim 1 , wherein:
at least one of the one or more graphs is represented by at least a subset of the calculated one or more vectors; and
the number of vectors in the at least a subset of the calculated one or more vectors is smaller than the number of transaction data entries.
11. A method, comprising:
receiving data, wherein the data includes transaction data entries;
generating one or more graphs using the received data;
calculating one or more vectors for the one or more graphs, wherein each of the one or more vectors represents a respective graph embedding corresponding to at least a portion of the one or more graphs, wherein the calculated one or more vectors are aggregated with one or more previously determined graph embeddings corresponding to previous transaction data entries;
determining that the one or more vectors belongs to at least one group of vectors based at least on the one or more vectors matching a data pattern associated with the at least one group of vectors including by building a hierarchy of clusters based at least on the aggregated one or more vectors with the one or more previously determined graph embeddings; and
identifying a representative graph for each of the at least one group of vectors.
12. A computer program product embodied in a non-transitory computer readable medium and comprising computer instructions for:
receiving data, wherein the data includes transaction data entries;
generating one or more graphs using the received data;
calculating one or more vectors for the one or more graphs, wherein each of the one or more vectors represents a respective graph embedding corresponding to at least a portion of the one or more graphs, wherein the calculated one or more vectors are aggregated with one or more previously determined graph embeddings corresponding to previous transaction data entries;
determining that the one or more vectors belongs to at least one group of vectors based at least on the one or more vectors matching a data pattern associated with the at least one group of vectors including by building a hierarchy of clusters based at least on the aggregated one or more vectors with the one or more previously determined graph embeddings; and
identifying a representative graph for each of the at least one group of vectors.