Method and apparatus for predicting risk, electronic device, computer readable storage medium

A computer-implemented data processing system for quantifying risk propagation in complex network data structures by addressing temporal risk conduction lag using a trained network, comprising:

• at least one processor;
• a display device;
• a training module;
• and
• a memory communicatively connected to the at least one processor;
• wherein,
• the memory stores in non-transitory memory instructions executable by the at least one processor, and the instructions, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
• generating an inherent risk probability of a to-be-tested object, the generating comprising processing static features representing basic attributes of the to-be-tested object, and extracting dynamic features of the object from historical data using a time-series feature extraction process;
• building and storing in non-transitory memory a relationship graph between the to-be-tested object and different associated objects, wherein the relationship graph comprises a plurality of nodes and association relationships among the nodes,
• and the building the relationship graph comprises extracting association relationships among the objects, creating entities in a graph database, creating node relationships in the graph database based on the association relationships, and constructing the relationship graph from the graph database;
• determining a primary conduction probability between any two directly associated objects in the relationship graph, comprising
• determining, based on the relationship graph, all associated objects associated with the to-be-tested object,
• determining conduction relationship pairs based on all the associated objects associated with the to-be-tested object, each of the conduction relationship pairs having a direct association relationship,
• determining, based on the relationship graph conduction edge features of each conduction relationship pair, and
• inputting the conduction relationship pairs and the conduction edge features of each conduction relationship pair into a pre-trained conduction probability model, stored in non-transitory memory, to obtain the primary conduction probability between any two directly associated objects output by the conduction probability model wherein the pre-trained conduction probability model comprises the trained network;
• determining, based on the primary conduction probability between any two directly associated objects in the relationship graph and the relationship graph, a multi-level conduction probability of the to-be-tested object by using the relationship graph to determine risk propagation through indirection association paths;
• determining a target risk probability of the to-be-tested object, based on the inherent risk probability and the multi-level conduction probability of the to-be-tested object;
• and
• transmitting instructions to display the obtained target risk probability on the device;
• wherein the memory stores in non-transitory memory the pre-trained conduction probability model, the conduction probability model represents a corresponding relationship between each conduction relationship pair having the conduction edge features and the primary conduction probability,
• said pre-trained conduction probability model is in communication with the training module,
• the training module is configured to train the conduction probability model by using conduction feature association pairs,
• the training module is configured to receive sample data, and is configured to process the sample data, comprising
• building conduction sample pairs based on the relationship graph, by using a sliding time window, comprising
• selecting a first sample that is at risk during an initial period of the sliding window as an initial node;
• determining, based on the relationship graph, a second sample associated with the initial node during a conduction period of the sliding time window as a termination node;
• and
• forming the conduction sample pair from the initial node and the termination node,
• labeling a pseudo-tag on each of the conduction sample pairs based on whether the termination node is at risk from the initial node in the conduction period, and
• perform feature processing on a conduction edge of each tag-labeled conduction sample pair based on the relationship graph, to obtain conduction feature association pairs, wherein a conduction feature association pair of the conduction feature association pairs comprises a plurality of conduction sample pairs with pseudo-tags and conduction edge features corresponding to the conduction sample pairs.

A non-transitory computer readable storage medium storing computer instructions for quantifying risk propagation in complex network data structures by addressing temporal risk conduction lag using a trained network, wherein, the computer instructions are used to cause the computer to perform operations comprising:

• generating an inherent risk probability of a to-be-tested object, comprising processing static features representing basic attributes of the to-be-tested object, and extracting dynamic features of the object from historical data using a time-series feature extraction process;
• building a relationship graph between the to-be-tested object and different associated objects, wherein the relationship graph comprises a plurality of nodes and association relationships among the nodes,
• and the building the relationship graph comprises extracting association relationships among the objects, creating entities in a graph database, creating node relationships in the graph database based on the association relationships, and constructing the relationship graph from the graph database;
• determining a primary conduction probability between any two directly associated objects in the relationship graph, comprising
• determining, based on the relationship graph, all associated objects associated with the to-be-tested object,
• determining conduction relationship pairs based on all the associated objects associated with the to-be-tested object, each of the conduction relationship pairs having a direct association relationship,
• determining, based on the relationship graph conduction edge features of each conduction relationship pair, and
• inputting the conduction relationship pairs and the conduction edge features of each conduction relationship pair into a pre-trained conduction probability model stored in memory to obtain the primary conduction probability between any two directly associated objects output by the conduction probability model;
• wherein the conduction probability model is used to represent a corresponding relationship between each conduction relationship pair having the conduction edge features and the primary conduction probability, wherein the pre-trained conduction probability model comprises the trained network,
• wherein the conduction probability model is trained by
• building conduction sample pairs based on the relationship graph by using a sliding time window, comprising
• selecting a first sample that is at risk during an initial period of the sliding window as an initial node;
• determining, based on the relationship graph, a second sample associated with the initial node during a conduction period of the sliding time window as a termination node;
• and
• forming the conduction sample pair from the initial node and the termination node,
• labeling a pseudo-tag on each of the conduction sample pairs based on whether the termination node is at risk from the initial node in the conduction period,
• performing feature processing on a conduction edge of each tag-labeled conduction sample pair based on the relationship graph, to obtain conduction feature association pairs, wherein a conduction feature association pair of the conduction feature association pairs comprises a plurality of conduction sample pairs with pseudo-tags and conduction edge features corresponding to the conduction sample pairs, and
• training a pre-built conduction probability network based on the conduction feature association pairs, to obtain the trained conduction probability model;
• determining, based on the primary conduction probability between any two directly associated objects in the relationship graph and the relationship graph, a multi-level conduction probability of the to-be-tested object, by using the relationship graph to determine risk propagation through indirection association paths;
• determining a target risk probability of the to-be-tested object, based on the inherent risk probability and the multi-level conduction probability of the to-be-tested object;
• and
• transmitting instructions to display the obtained target risk probability on a display device.