The Research Of Forecasting For Fashion Products’ Popular Trends Under The Internet Environment

The reason why a product is popular is that it can reflect the popular trend in modern times. Therefore, it’s necessary for fashion products enterprises to be able to identify and predict the popular elements.Nowadays, the most widely used method for fashion trend forecasting is qualitative prediction method which is based on the knowledge of the domain experts. But this kind of method is subject to the professional skills and experience level of the experts. In recent years, the quantitative method is used to forecast fashion trends. For the quantitative method, the data is mostly transaction data. On the one hand, based on the characteristics analysis of transaction data, a combined model is constructed to forecasting the popular trend of the fashion product. On the other hand, the correlation analysis between web search data and transaction data have already became the hot research problem recent years. The Web search data reflects the consumers’attention of product’s popular elements. To give full play to the Web search data, this paper integrated the Web search data and transaction data as uniform datab source to conduct predict model, the model realized the offline and on-line prediction separately.Weather a product can be popular or not is influenced by the psychological needs of consumers, and can be reflected by their behaviors. In the long term, consumer behavior is regular to some extent, while the short-term consumer behavior, affected by various factors, shows great instability, which causes great trouble to short-term prediction of popular elements of fashion products. The uncertainty of consumption behavior is due to the fact that both linear and nonlinear phenomenon exist in the changes of product sales data. In the face of this data feature, based on the ARIMA time series model and ELM extreme learning machine model, this paper conducts the linear modeling and nonlinear modeling on sales data, respectively. Then, integrated by a weighted voting mechanism, a combined model is constructed. In the construction of ELM extreme learning machine model, the method of multiple running is adopted to get the average to solve the problem of unstable operation results, and the method of trend extrapolation optimization is used to set the optimal number of hidden layer. With industry data of garment trading in Taobao collected from January 2010 to July 2013, the combined model proposed in this paper is used to predict the short-term clothing color’s popularity. Experimental results illustrates that the developed combined model is more reasonable and accurate by comparing with the single predict model.The Web search data, not only reflects the focus attention and personalized demands of users, but also implies users’social or economic behaviors. This paper presents a prediction model both on historical sales data and the Web search data. Considering that the demands of fast fashion products are always changeable and fast fashion products update very often, the so called extreme learning machine (ELM) is developed to forecast the popular trends of fast fashion products. ELM has been proved to have a better generalization performance and a strong learning ability. By comparing with the traditional statistical prediction models, our model is more accurate after introducing the Web search data. Moreover, it is able to find out the inflection points of popular trends. At the same time, the OS-ELM model realized the on-line prediction.