Research On Recommendation Model Of Telecom Package

With the commercialization of the fifth-generation communication technology(5G)and the full implementation of the "number-transferring" policy across the country,competition among mobile operator has become increasingly fierce,which has caused major operators to adjust their strategies one after another.New strategies are adjusted to attract new customers and retain old customers.It is very necessary to recommend a suitable telecommunications package to old customers by using their historical behavior records and following their preferences.At the same time,with the development of computer technology and data science,machine learning methods based on user information are increasingly used to establish personalized recommendation models for users,providing techniques and methods for establishing telecommunication user package recommendation models.We adopted the user’s historical data from a domestic telecom operator.On the basis of basic processing and statistical description of data,analysis of user composition and user’s historical consumption behavior,and reconstruction of data characteristics,we combined and constructed the existing recommendation algorithms,built the Hybrid recommendation model,to compared and analysis the models,and then selected the optimal model to predict the user’s selected package for a series of packages for the operator.finally,we built a telecommunications user package recommendation model.There are several commonly used types of user recommendation models: contentbased recommendation,collaborative filtering,apriori-based and classification-based recommendation,and hybrid models.We explored the application of the newer XGboost modeling method based on apriori and classification algorithms,and it is integrated with the content-based recommendation algorithm to construct a hybrid recommendation model.The empirical analysis of telecommunications user historical data was verified by four evaluation indicators: precision,recall rate,F1 value and confusion matrix.We used the verification result to compare the prediction results between the hybrid recommendation model and the XGboost single model,and obtained the conclusion that the prediction precision of the hybrid model is highly.that means that the hybrid model’s forecasting ability is stronger,and the hybrid model is suitable for operators to recommend telecom packages for users due to their historical behavior data.The research conclusion is helpful to improve the service level of mobile operators,and successfully retain their original customers in fierce competition by improving the perception of their user.The empirical research on building a hybrid recommendation model shows that the hybrid recommendation model is a good development direction for building a user recommendation model.However,since this example only based on the 6 existing telecom packages provided by the mobile operators,the representative of the recommendation conclusion is limited,but the study of the model proved that the hybrid model is a better recommendation model than the single model,and has a relatively large development space.