DescriptionThis dissertation is comprised of three essays on the impact of social media on trend prediction in the fashion industry. The essays include: 1) an evaluation of the Twitter social media platform for fashion trend detection by using a machine learning approach and text mining techniques, 2) a design for a forecasting algorithm to predict fashion trend dynamics by utilizing big data analysis based on Amazon sales, and 3) methods for enhancing forecasting by using sentiment analysis of Twitter corpora.
One of the biggest challenges for fashion apparel companies is accurate demand planning. Conventional apparel consumers rush into obtaining emerging fashion styles. This phenomenon forces many fashion companies to deviate from the classic biannual supply chain cycle and shorten their go-to-market strategy. Consequently, there is no time for brands to learn consumers’ demand by utilizing conventional methods such as pre-sale.
Using historical demand data to accurately forecast consumer preferences is especially difficult in the current field of fashion. The classic models based on historical sales have low reliability for sales planning in an era in which fast fashion dominates. Fast fashion intensifies and accelerates several distinguishing characteristics of the fashion business, including demand uncertainty, high stock-out costs, a high risk of obsolescence, short product lifecycles, and long lead time.
At the same time, fashion industry consumers are active Web 2.0 users. One of the most popular platforms is Twitter, where people publish personal opinions and discuss experiences related to emerging fashion trends in real time. The first essay in this thesis examines whether trending fashion styles and features, such as colors, cuts, fabric types, etc., may be mined from Twitter fashion corpora. The second essay investigates the value of using data on fashion trends identified on Twitter in order to enhance the predictive power of forecasting models of fashion sales. Amazon apparel sales data were used to evaluate the model. Customer sentiment information may also be mined from Twitter fashion corpora. Sentiment analysis techniques are frequently applied along with natural language processing (NLP) for sales prediction in many industries. The third essay investigates whether using sentiments mined from Tweets as an additional predictor for fashion trends forecasting leads to performance improvement. A sentiment variable is added to forecasting models presented in the second essay and the resulting forecasting error is evaluated.