A classification system and method for game platform messages

Abstract

The invention discloses a message classification system and method of a game platform. The system includes a spam message discovery module, a sample library, a model training module and a real-time processing module. The spam discovery module processes spam through the PU learning algorithm to obtain sample messages and store them in the sample library; the model training module uses the sample messages in the sample library, and trains similarity classification models or character-level CNN text classification models according to the game launch period; real-time processing The module calls the classification model of the model training module to discriminate user messages, and performs secondary discrimination through LP algorithm and keyword analysis, and feeds back the results to the spam classification module. In this way, the PU learning algorithm can be trained with a small number of samples, and solve the problem of lack of samples in the cold start process; through the character-level CNN text classification model, the recall rate and compatibility are improved; through the LP algorithm and keyword analysis, real-time defense is realized, and at the same time acquisition New sample messages build a similarity model to solve the cold start problem.

Description

technical field [0001] The invention relates to the field of data processing, in particular to a game platform message classification system and method. Background technique [0002] With the gradual development of the game industry, there are more and more spam messages in the chat interface of online games. How to classify the messages of the game platform (automatically filter out spam messages) to prevent users from being swiped and abused by malicious advertisements Even fraudulent information deception is a problem that every game company has to face. According to the differences in the technical strength and size of different game companies, the traditional solutions generally have the following types: [0003] 1. Keyword blocking: set keywords corresponding to spam messages, identify and classify user messages, and block spam messages after triggering; [0004] 2. Manual account blocking: operators search for spam messages and perform account blocking operations; ...

AI Technical Summary

Problems solved by technology

[0005] 3. Increase the difficulty of speaking: such as adjusting the speaking level, props speaking, etc.;

[0010] ①Sample collection problem: supervised learning algorithms rely on a large number of samples;

[0011] ② Unsatisfactory classification effect: The effect of the existing model scheme is not very satisfactory;

[0014] ⑤ Game cold start problem: How to classify game platform messages in a timely manner at the beginning of the game

Images