Research on Name Popularity Based on Neural Time Series Prediction

Le résumé fourni par la source

Parents usually consider many factors when choosing a baby name, such as the popularity of the baby name. However, predicting the baby name time series is an important means of judging the popularity. In view of the lack of use of current time series prediction methods in baby name dataset, this paper selects the frequency of American baby names in 139 years from 1880 to 2018 as the research object, establishes three autoregressive baseline models of ARIMA, LSTM and DeepAR, and combines historical data to predict the frequency of names in the next five years. The experimental results show that the Mean Absolute Percentage Error of ARIMA, LSTM, and DeepAR were 33.85%, 23.49%, and 20.23% respectively. The overall prediction accuracy of the LSTM neural network model significantly outperforms the traditional ARIMA model and is more suitable for multiple time series forecast analysis. Compared with the single value prediction, DeepAR model of probability distribution prediction can better fit the time series of name usage frequency, and the prediction accuracy of the model is also better than that of ARIMA model and LSTM model.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.