Predicting community members based on evolution of heterogeneous networks using a best community classifier and a multi-class community classifier

A computer-readable medium containing instructions for controlling a computer system to identify a community of objects of a designated type, by a method comprising:

• providing indications of objects of different types and of relationships between objects of different types, each relationship having an associated time;
• generating a heterogeneous graph with vertices representing distinct objects and edges representing the relationships between objects, each edge having a time period and a weight;
• generating a time vector heterogeneous graph from the heterogeneous graph, the time vector heterogeneous graph having a vertex for each vertex of the heterogeneous graph and an edge between objects representing a relationship between the objects, each edge having a time vector representing the weights of the relationship over time periods
• extracting features relating to objects of the designated type from the objects and their relationships as indicated by the generated time vector heterogeneous graph, the extracted features representing evolution of the features over time periods;
• providing training data including features of objects within the community, the training data including best community training data generated by, for each time period and each object of the designated type that is within the community for the time period:
• setting best community training data to the extracted features for that object;
• labeling the best community training data for that object with a rank of the highest ranking community that that object is within;
• and
• setting a time period for the best community training data for that object;
• and
• the training data further including multi-class community classifier training data generated by, for each time period, each community, and object within the community within that time period:
• setting multi-class community training data to the extracted features for that object;
• labeling the multi-class community training data for that object with the community;
• and
• setting a time period for the multi-class community training data for that object;
• and
• classifying an object of the designated type as being within the community based on closeness of the features of the object to the features of the training data that includes best community training data and multi-class community training data such that when the object is classified as a potential community member for multiple communities based on the best community training data, classifying the object as a potential community member for a single community based on the multi-class community training data.

A computer system that identifies potential community members of a community, the community members being committee members of a conference, comprising:

• a data store that indicates objects representing persons, conferences, and papers and indicates relationships between objects of different types, each relationship having a time period;
• a memory storing computer-executable instructions for
• a component that generates a heterogeneous graph with vertices representing distinct objects and edges representing the relationships between objects, each edge having a time period and a weight;
• and
• a component that generates a time vector heterogeneous graph from the heterogeneous graph, the time vector heterogeneous graph having a vertex for each vertex of the heterogeneous graph and an edge between objects representing a relationship between the objects, each edge having a time vector representing the weights of the relationship over time;
• an extract feature component that extracts features relating to persons from the relationships as indicated by the generated time vector heterogeneous graph, the extracted features representing evolution of the features over time;
• a generate best community training data component that generates training data for a best community classifier having a label associated with a person, the label indicating potential to be a community member, the training data being generated by, for each time period and each person who is a community member of a community within that time period:
• setting best community training data to the extracted features for that person;
• labeling the best community training data for that person with a rank of the highest ranking community that that person is a community member;
• and
• setting a time period for the best community training data for that person;
• a train best community classifier component that trains a best community classifier using the best community training data to classify the potential for a person represented by their features to be a community member of a community;
• a generate multi-class community classifier training data component that generates training data for a multi-class community classifier having a label associated with a person, the label indicating potential to be a community member, the multi-class community training data being generated by, for each time period, each community, and community member of a community within that time period:
• setting multi-class community training data to the extracted features for that person;
• labeling the multi-class community training data for that person with the community;
• and
• setting a time period for the multi-class community training data for that person;
• a train multi-class community classifier component that trains a multi-class community classifier using the multi-class community training data to classify the potential for a person represented by their features to be a community member of a community;
• and
• a classify person component that classifies a person as a potential community member using the best community classifier and when the person is classified as a potential community member for multiple communities, using the multi-class community classifier to classify the person as a potential community member for a single community;
• and
• a processor for executing the computer-executable instructions stored in the memory.