High dimension sparse data clustering system and method

High dimension sparse data clustering system and method

  • CN 101,266,621 B
  • Filed: 04/24/2008
  • Issued: 08/10/2011
  • Est. Priority Date: 04/24/2008
  • Status: Active Grant
First Claim
Patent Images

1. a high dimension sparse data clustering method is characterized in that, comprises the steps:

  • Step 1;

    extract the sparse features of each data object, calculate the diversity factor between the data object in twos according to the sparse features of each data object;

    Step 2;

    data object and ant are randomly dispersed on the two dimensional surface, and distribute a data object for ant, ant bears this data object;

    This two dimensional surface is the net region that boundary condition is arranged;

    Step 3;

    when ant moves to a net region, calculate the average similarity of other data object in current data object that it is born and the current net region according to diversity factor;

    Calculate ant to bearing the probability that puts down of data object according to average similarity, meet first pre-conditionedly if put down probability, then ant is put down it and bears data object;

    Step 4;

    when ant does not bear data object, select a data object of not born at random, according to diversity factor calculate this data object in it self net region, place with the average similarity of other data object, and according to the pick up probability of average similarity calculating ant to this data object, meet second pre-conditioned if this picks up probability, then execution in step 2, do not born and select data object otherwise choose one again;

    Wherein, in step 3 and the step 4, calculate average similarity according to the following equation;

View all claims
    ×
    ×

    Thank you for your feedback

    ×
    ×