Application of Abnormal Behavior Detection in Movie and Music Streaming Services

Abstract

In the study of abnormal behavior detection of users in movie and music streaming services, the traditional TCN (Temporal Convolutional Networks) ignores the complex relationship between multi-dimensional data, the accuracy of anomaly detection is average, and it relies on batch data processing and offline analysis, resulting in delays in the detection of abnormal behavior. This paper constructs an improved TCN model to collect user behavior data from movie and music streaming services, cleans, denoises and standardizes the data, and extracts representative time series features such as behavior frequency, time interval and behavior duration. An improved TCN model is constructed. The multilayer convolution kernel and residual connection are added based on the original convolution structure. The processed user behavior data is input into the improved TCN model, and the model’s output is used to determine whether the user behavior is abnormal. An incremental learning method is used to process batch user behavior data in real-time. Experimental results show that after 100 training rounds, the loss value and accuracy of the improved TCN model in the verification set tend to 0.04 and 0.93 respectively; the precision, recall and F1 score of the improved TCN model in the test set are respectively are 0.93, 0.91 and 0.92, which are significantly better than other detection models and have good detection performance. Under 900ms network delay, the model preprocessing time is reduced by 1 second, the model inference time is reduced by 1.6 seconds, and the post-processing time is reduced by 0.3 seconds after incremental learning, which reduces the delay of abnormal behavior detection.