Spam feature calculation device, spam features calculation method, and the program

摘要

PROBLEM TO BE SOLVED: To calculate a feature capable of appropriately discriminating a spam page even if the spam page is automatically generated.SOLUTION: A spam feature calculation device for calculating a feature used to discriminate a spam document comprises: topic leaning means which calculates a topic parameter composed of a sample number of words to each topic from a training text which is a non-spam document and stores it in a topic parameter DB; and feature calculation means which divides a text of feature calculation target into prescribed units, obtains aggregation of divided texts, calculates a topic vector using the topic parameter stored in the topic parameter DB for each divided text, and calculates the feature on the basis of a difference between topic vectors.