Résumé
In the fashion market, the emergence of the Internet and Big Data has profoundly changed the competitive environment as well as the relationships with clients. Indeed, in the era of Big Data, social networks, mobile applications, and various recent technologies allow the collection of massive amounts of data in real-time at a low cost. It has therefore become crucial for fashion retailers to optimize the exploitation of this data, extracted from their information systems or from social networks, in order to retrieve relevant information for the various decision-making processes such as the optimization of the assortment of collections and the quantities to be restocked in stores.More precisely, within the scope of this thesis, massive data and artificial intelligence tools are used to answer the problems of fashion trends detection and sales forecasting, in different contexts. The proposed model relies on different types of data to predict short-term trends and demands: (i) Internal company data such as historical transactional data and data from previous collections, and (ii) External data from social networks, such as images and texts from different sources. The methodology developed is composed of three steps. First, the extraction and processing of data from social networks allow the identification and quantification of the most relevant fashion trends. Secondly, historical transactional data from within the company is used to establish sales forecasts by taking into account the influence of factors controlled by the company such as commercial actions, price variations, sales periods, etc. Finally, the detected fashion trends are integrated with the previously established sales forecasts in order to more accurately predict future demands, and consequently, evaluate the usefulness of heterogeneous data analysis for fashion demand forecasting. The proposed model can be used to forecast store replenishments of current products and/or to generate a short-term supply of new products. The proposed model is implemented and evaluated on real data from a French ready-to-wear retailer
Source: http://www.theses.fr/2022CLIL0031
.
