Encrypted traffic QoE detection method and system based on multi-task learning and hierarchical classification

Abstract

The invention discloses an encrypted traffic QoE detection method and system based on multi-task learning and hierarchical classification, and belongs to the technical field of computer software. According to the method, the classification of the QoE indexes of the encrypted traffic is realized by performing automatic feature extraction on the trend chart by using the CNN, and meanwhile, a multi-task learning algorithm and hierarchical classification are utilized, and a plurality of QoE indexes are combined for mutual auxiliary learning, so that implicit features in the QoE indexes can be automatically identified; the higher QoE index classification accuracy is achieved; traffic characteristic changes generated by time and network changes can be well coped with; and a very good classification effect under a plurality of different video traffic data sets can be achieved.

Description

Technical field [0001] The present invention is directed to the method and system of flowing Qoe detection under the encrypted flow scenario, which belongs to the technical field of computer software. Background technique [0002] According to CNNIC (China Internet Information Center), the statistical report of China Internet Development Statistics shows that as of June 2020, my country's netizens reached 940 million, which increased by 36.25 million in March 2020, and the Internet penetration rate was 67.0%. 2.5 percentage points were increased in March 2020. In such a market environment, there have been more than 300 online video websites in China. After these years, the BAT giant and capital mergers have been eliminated, there are only 10 large-scale video sites. [0003] Under such a fierce competition, whether it is the industry or the jurisdiction, the traditional network performance indicator is not enough to judge the satisfaction of the user, and the user experience (QOE...

AI Technical Summary

Problems solved by technology

Although this method based on machine learning can achieve very good classification accuracy, it requires a lot of manual feature extraction work and expert experience, and the model trained in this way may not be able to maintain such a high performance when transferred to other databases. Accuracy, not robust enough

QoE detection based on deep learning, using deep learning to do QoE detection no longer requires manual extraction of traffic features, which avoids a lot of manual operations, and in terms of automatic feature extraction and accuracy, deep learning methods have shown better effect, some researchers proposed to use CNN and LSTM (long short-term memory network) to extract traffic features (M.Lopez-Martin, B.Carro, A.Sanchez-Esguevillas, and J.Lloret, "Network traffic classifier with convolutional and recurrent neural networks for internet of things, "IEEE Access, vol.5, pp.18 04218 050, 2017.), transforming traffic from one-dimensional data into two-dimensional images and then using CNN to process them can solve traditional processing The time-consuming and low-efficiency problems encountered by image data, combined with LSTM can achieve a very high QoE index classification accuracy, but its disadvantage is that the robustness of QoE for video traffic from different websites is not good enough.

Images