To provide a general knowledge discovering system for efficiently realizing knowledge discovery by selectively sampling only data whose information quantity is large from a data base.

A learning means 104 allows a low rank learning algorithm inputted through an input part 107 to learn plural partial samples prepared by sampling from data stored in a high speed main memory 120, and obtains plural hypotheses. A data selecting means 105 estimates the information quality of each candidate data read from a large capacity data storing device 130 by using the hypotheses, and additionally stores only the data whose information quantity is large in the high speed main memory 120. A control part 106 repeats the processing prescribed number of times, and stores the obtained final hypothesis. Then, a predicting part 102 predicts a label value by the final hypothesis for data inputted to the input part 107 whose label is unknown, and an output part 101 outputs this predicted value.