The FashionBrain solution aims at combining data from different sources to support different fashion industry players by predicting upcoming fashion trends from social media as well as providing personalized recommendations and advanced fashion item search to customers.
The main objective of the FashionBrain solution is to consolidate and extend existing European technologies in the area of database management, data mining, machine learning, image processing, information retrieval and crowdsourcing to strengthen the position of European fashion retailers among their world-wide competitors.
The FashionBrain solution targets the two main actors of the fashion industry: European retailers and customers. We propose to gather and combine the sheer amount of data generated by (emanating from) different fashion industry multisectorial players.
The gathered data will be curated, analysed and used as input for machine learning algorithms. The outcome of the project will benefit retailers by reducing the financial efforts for Search Engine Optimisation (SEO) paid to the Web search engine. It will benefit both the retailer and customer by providing novel services to customers in order to improve their shopping experience thereby boosting their brand/company loyalty.
Sales & Marketing, Brand Manufacturers/Distributors, Retailers
A core business of Europe’s fashion industry is to acquire a deep understanding of customer needs and to predict next trends. Search engines and social networks are often used as a bridge between the customer’s potential purchase decision and the retailer. In order to reinforce Europe’s position in the fashion industry and better exploit its distinctive characteristics e.g., multiple languages, fashion and cultural differences, it is pivotal to reduce its dependence to search engines.
This goal can be achieved by harnessing various data channels that retailers can leverage in order to gain more insight about potential buyers, and on the industry trends as a whole. The main outcome of the FashionBrain solution is the improvement of the fashion industry value chain obtained thanks to the creation of novel online shopping experiences, the detection of influencers, and the prediction of upcoming fashion trends. Tangible outcomes will include software, demonstrators, and novel algorithms for a datadriven fashion industry.
Most existing fashion retailers in the mid 2000s decided not to invest massively in owning search engines but to pay global web search engines large amounts for SEO and advertisements. In order to alleviate the existing dependence from social networks and search engines, fashion retailers should be able to use their own tools and data to predict next emerging trends, and to acquire fashion related data by other means, for example by crowdsourced activities or by tailored user interactions.
We intend to:
- Shift traffic away from Web search engines to retailer’s mobile applications and domains.
- Create a novel shopping experience by making images searchable.
- Detect influencers to predict fashion trends.
- Share insights within Cross Industry Partner Network to create a Data Integration infrastructure.
The main technical challenges existing in the world of online fashion retail are due to the lack of data integration amongst different infrastructures and sources, complexity of the data workflow, scalability, and lack of training data for supervised models.
FashionBrain solution tackle these challenges proposing a data integration ecosystem based on core primitives of in-memory databases, deep learning over text, and crowdsourcing. These enables end-user applications like, for example, search product by image and fashion trend prediction.
Fashion time series reflect the evolution of user preferences to fashion items. These time series represent for example the sales of fashion items, the number of likes of an image on social media or the reviews in a retailer website along a period of time, etc. Compared to time series from other domains, the fashion time series exhibit two distinguishing features i) fast evolving trends and ii) visual representation of items. In fact, the increase of fashion sales or number of likes of a particular item means that users are showing an interest in this item and consider it to be fashionable. The frequently changing user preferences towards fashion items have put great challenge on the dynamic modeling of both users and items. Besides the temporal dynamics of fashion data, the visual appearance of fashion items plays a key role on user’s decision-making behavior.
Our solutions allow in-database implementation of fashion related machine learning tasks, thanks to the integration of Flair and IDEL into MonetDB, together with trend detection and prediction tools.