Nuevo seminario de la serie “Nuevas soluciones con Big Data” organizados por el UC3M-Santander Big Data Institute (IBiDat) donde se presentarán problemas reales en distintos campos y la solución aportada utilizando todos los datos disponibles. Lo seminarios intentan ser un punto de encuentro de profesionales y académicos para presentar problemas y analizar posibles soluciones basadas en Big Data.
El siguiente seminario será el viernes 26 de abril a las 9.30 en la sala 18.1.A12 del edificio nº 18 (Carmen Martín Gaite) del Campus de Getafe de la Universidad Carlos III de Madrid y finalizará a las 11:00.
En esta ocasión el seminario tendrá como título “ Towards online sales marketing activities: Machine Learning for estimating user purchase intention“ (the talk will be in English). Será presentado por José Andrés Martínez Moreno, Head of Data Science at Coolblue.
E-commerce has been substantially growing each year for the last decade. The global e-retail market value reached US $2.3 trillion in sales in 2017 (Statista, 2017). However, the rate of growth has declined and is expected to stagnate in most of the developed countries in the near future (Statista, 2018). Therefore, online stores need to find strategies to maintain market share in a volatile and very competitive market.
It is widely known, the number of visitors to a webshop is remarkable larger than in bricks and mortar stores, and conversion rates are dramatically lower for online stores in comparison to offline stores. Therefore, small steps in understanding online behaviour that boost the accuracy on predicting and prescribing actions, can lead to considerable increase in revenues and profit margins.
Having a certain level of knowledge of how likely a user is to make a purchase, allows marketing and sales professionals to perform personalized online and offline actions. In particular, retailers could implement personalized content to guide users sessions from potential buying-sequences to purchase outcomes and/or upselling.
It is common to find in recent literature Machine Learning approaches for calculating purchase intention, mostly using time-related features combined with a set of categorical variables. This set of categorical variables typically refers to sequences from the clickstream and most of the time is summarized aiming to reduce the very large set of different actions available in a webshop to a few dozen (at the most detailed level, there are tens or hundreds of thousands of different actions: clicking products, zooming, scrolling, filtering, information retrieval using internal search, etc). In some cases
researchers use additional information from past visits and purchases, so the user has to be identified and as a consequence, the method can not be used with the clickstream data from potential customers or unidentified customers.
In this talk, will be discussed some general details, the intuition and some relevant nuances of a Machine Learning approach for estimating users purchase intention; using the most detailed clickstream data at the session level, from anonymous users that interact with the webshop. The clickstream dimensionality is reduced using a knowledge discovery technique. The results demonstrate a very high prediction power achieving state-of-the-art results in both datasets: training and holdout.
Los seminarios están organizados por Carlo Sguera, investigador de IBiDat (firstname.lastname@example.org) al que puedes dirigirte para información adicional.
Los seminarios son de carácter abierto y gratuito. Se agradece confirmación de asistencia rellenando el siguiente formulario https://forms.gle/fyjgCtcaweihyfcN8