Using Growing Neural Gas Networks for Clustering of Web Data as a Foundation for Marketing Automation in Brick-and-Mortar Retailing

Abstract

Even though more than 90 percent of retail executives agree to personalization in marketing being a top priority for them, only a handful deliver on this aspect. In brick-and-mortar retailing marketing actions that are conducted for large-scale grid based on an arbitrarily set distribution area - not tailored to single stores. This is mainly due to the lack of data and missing processes for automatization of marketing management. Even though the collection of data is easy to establish, the resulting data quality and condition is a problem for many clustering algorithms. Even methods of Machine Learning that are considered to be more robust to noise have issues with the unpredictable nature of such data collection, as they rely on a predefined structure. This paper presents a method for collecting a large amount of relevant location information for a major German brick-and-mortar retailer from online data sources and mapping them to a grid model of Germany at the level of 1 km2. The clustering is conducted using methods of Machine Learning, in particular Growing Neural Gas (GNG), a neural network variation. The resulting store clusters are then tested for the purpose of location specific marketing. The GNG adapts good to the noisy data and the practical quality is feasible.