System and method for quickly identifying crank calls based on non-frequency characteristics of signaling

Abstract

The invention relates to the fields of telecommunication technology, big data and the like, in particular to a signaling-based non-frequency characteristic crank call rapid identification system. The system consists of a signaling acquisition unit, a historical signaling total database unit, a signaling characteristic classification modeling unit and a signaling monitoring and crank call interception unit. And the signaling acquisition unit is used for acquiring original signaling, converting the original signaling into a detailed call record, and then transmitting the detailed call record to the historical signaling total database unit for storage. And the signaling feature classification modeling unit uses the sample data to form a non-frequency feature crank call discrimination library. And the signaling monitoring and crank call intercepting unit monitors the signaling in real time, identifies the crank calls in a library by combining non-frequency characteristic crank calls, and intercepts or reminds the crank calls. The invention also comprises a method. According to the method, modeling is trained through full features, non-frequency features are collected in the recognition stage to judge and recognize the crank calls, and therefore the work of secondary sampling and repeated feature vector calculation in the recognition stage is avoided, real-time recognition is achieved, and the recognition accuracy is improved.

Description

technical field [0001] The invention relates to the fields of telecommunication technology, machine learning and big data, in particular to a system and method for quickly identifying harassing calls based on non-frequency characteristics of signaling. Background technique [0002] Harassing phone calls are a cancer in today's network society and real society, and the harm to individuals and the whole society is huge. In addition, a large number of illegal calls will occupy precious communication resources, directly leading to problems such as a drop in connection rate and network equipment congestion, which greatly reduces the experience of legal mobile users. [0003] The current governance solutions for harassing calls can basically be divided into three categories from the technical field: [0004] First, through user identification [0005] This method is to mark the calling number through user complaints, including telephone complaints, APP reports, etc., and then in...

AI Technical Summary

Problems solved by technology

[0010] However, the above three types of methods all have defects or even unsolvable problems:

[0011] For the first method, through user complaints or APP reports, the workload and labor costs of manual customer service are greatly increased, and some numbers may be maliciously reported, resulting in a large number of misjudgments

[0012] For the second method, speech recognition and audio comparison, the call records of the calling party and the called party must be monitored, which is too intrusive and suspected of violating user privacy. At the same time, a large number of storage devices need to be added. The technical implementation is complicated and greatly increases Software and hardware costs, long recognition cycle

[0013] For the third method, the method of big data + machine learning algorithm to model signaling features to identify harassing calls is the most promising method at present, but the identification of harassing calls based on machine learning modeling of signaling features is currently There is a serious flaw, that is, the modeling of harassing calls depends heavily on the frequency characteristics of signaling. When a new number that has never appeared in the full training set and test set arrives, because there is only one record, It is impossible to calculate the frequency characteristics of this signaling, such as call frequency, connection rate, called dispersion, etc. These frequency characteristics are derived from statistics based on a large amount of data, so it is impossible to quickly judge The category of the signaling

As far as the current method of identifying harassing calls based on machine learning algorithms + signaling features is basically a method of forming a blacklist library, it is slightly better based on short-term resampling, such as 5-minute granularity, based on 5-minute The sample size and the frequency characteristics are counted. However, the number of signaling that appears within 5 minutes is obviously far from enough, because there may still be only one occurrence of each signaling within 5 minutes, making it impossible to quickly identify and intercept online Purpose of harassing calls

Images