Twitter and Fashion Forecasting: An Exploration of Tweets regarding Trend Identification for Fashion Forecasting

The fashion industry faces serious challenges in terms of accurate demand forecasting. While production decisions have to be made at an early stage, precise demand information only become available several months later. One main characteristic of the fashion industry is long time-to-market compared to short selling periods. Consequently, it is hardly possible to replenish successful products. Therefore, companies will have losses in terms of stock-outs or overstocked inventories. In order to avoid these losses accurate forecasts are needed. We suggest examining social media text data to support baseline forecasts. This research explores the question if the microblogging service Twitter can be an appropriate source for extracting relevant features in order to predict future fashion trends. Mainly we tackle the following questions regarding the Tweets: are fashion related topics discussed on Twitter? Can we extract information regarding colors, cuts, materials or fashion styles of a product? And if this is given how these words do occur together? For this purpose we collected Tweets which are either brand related, product type related or event related. The analysis is divided into two parts: In the first step, the pre-processing of the text data, we applied tokenization, stopword filtering, stemming and case transformation. In a second step, we applied Associations Rules in order to examine co-occurrences of the extracted words. The analysis shows that it is difficult to draw quantitative conclusions out of the data we obtained. This work is more a qualitative approach to the topic and we suggest validating our examination with bigger data set.