Spam Detection via Account Attributes and Bayesian Inversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting spam messages in microblogs are ineffective due to difficulties in maintaining accurate spam message sample libraries and interference from spammers, leading to delayed detection and misclassification of spam messages.
Innovation Solution
A method that determines abnormal message probabilities based on account attributes, using undiscriminating text segmentation and Bayesian calculations to assess the likelihood of a message being spam without relying on pre-collected sample libraries, allowing for self-learning and flexible spam recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual detection or behavior detection algorithms are used to build spam message sample libraries, then spam messages can be detected, but the detection is delayed by several hours and misjudgments occur frequently
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing the abnormal probability values for various account attributes (such as registration time, posting frequency, follower ratios) before spam messages actually appear. When a new message arrives, the system immediately retrieves these pre-computed probabilities and performs Bayesian calculation without needing to manually detect or analyze the spam message first. This eliminates the several-hour detection delay while maintaining high accuracy through the pre-established probabilistic model.
2Reliability
If word segmentation is performed on spam messages to detect them, then spam detection can be conducted, but spammers add interfering symbols or replace characters to avoid detection
Solution Approach 1:
The patent inverts the traditional detection approach by not analyzing the spam message content directly through word segmentation. Instead, it evaluates the account attributes of the sender (registration time, posting frequency, follower-to-following ratio, etc.) and calculates the probability that the account is abnormal. This inversion makes the detection immune to spammer evasion techniques like adding interfering symbols or replacing characters, because the detection no longer depends on the message text but on the sender's behavioral patterns.
Solution Approach 2:
The patent introduces account attributes as an intermediary between the spam message and the detection system. Rather than directly analyzing the message content which spammers can manipulate, the system uses account attributes (registration time, posting frequency, follower ratios) as intermediate indicators to infer the likelihood of spam. This intermediary layer protects the detection system from evasion techniques while still enabling accurate spam identification.
3Productivity
If traditional word segmentation and Bayesian calculation based on sample libraries are used, then spam probability can be calculated, but the method cannot handle most spam messages due to sample library inaccuracies
Solution Approach 1:
The patent changes the parameters used for spam detection from message-content-based parameters (word frequencies in sample libraries) to account-attribute-based parameters (registration time, posting frequency, follower ratios, etc.). By changing these fundamental parameters, the system achieves both high productivity in processing spam messages and high measurement precision in calculating spam probability, eliminating the limitations of traditional sample library approaches.
Data Source
AI summary
Disclosed is a method for detecting an abnormal message, comprising: diving a text of a detected message into a plurality of text segments; obtaining one or more account attributes of each text segment, and determining a publication proportion parameter corresponding to the account attributes of each text segment; determining a first factor corresponding to the account attributes of each text segment according to the publication proportion parameter; determining a second factor of the detected message according to the first factor corresponding to the account attributes of each text segment; and determining according to the second factor of the detected message whether the detected message is an abnormal message. Through the combination of publication account attributes of messages with undifferentiated text segmentation and the use of Bayesian algorithm, batches of junk messages of a microblog account are effectively limited, and the flexibility of junk message processing is improved.