Abstract: At present, the research on digital content in telecom industry is mainly based on the user insight of different preferences based on business caliber, and most of them are based on business experience, which is not conducive to the development and expansion of the scale of digital content users. To this end, this paper makes use of the historical data of large-volume customers and studies the digital content preference based on multi-label classification algorithm, so as to obtain various potential target customers, and finally recommend customers’ preferences through marketing to improve precision marketing ability. Firstly, desensitization data such as the basis and consumption attributes of M telecom users are taken as the data source, and the list of active users of video, music and reading in the last three months is obtained. The active dimension is manually annotated to obtain the initial data set. Because the positive and negative samples are not balanced, three samples are randomly sampled by multiple down-sampling method, and six algorithms including CC, ML-KNN and RakelD are used for comparative experimental verification. The experimental results show that the RakelD and ML-KNN multi-tag classification algorithms have better predictive ability in the perspective of user preference. Therefore, ML-KNN is adopted as the basic classifier of RakelD algorithm, namely RakelD_MLKNN method, to respectively predict the data sets with different positive and negative sample proportions, and the results are all better than the previous 6 existing common multi-label classification algorithms and traditional empirical selection methods.

Key words: digital content preference, multi-label classification, Classifier Chains (CC) algorithm, Multi-Label K-Nearest Neighbor (ML-KNN) algorithm, Random k labelsets Disjoint (RakelD) algorithm