The Connected Consumer use case utilizes the BigDataStack environment to implement and offer a recommender system for the grocery market. All of the data that are used to train the analytic algorithms of the use case are corporate data provided by one of the top food retailers companies in Spain. The final goal of the scenario is to make use of the capabilities of BigDataStack to produce in-store tailored recommendations to customers at the on-line applications of the food retailer.
What is the business objective?
This scenario aims to create a collaborative-filtering recommender system that produces recommendations of new products to the food retailer’s customers. Thus helping the food retailer to improve the user experience of some of their current applications (e-commerce site, loyalty app, etc.) and at the same time improve their customers’ loyalty.
What’s the BigDataStack differentiator?
Many retailers nowadays are offering recommendations of products and promotions to their customers. However, these recommendations are sometimes calculated weeks in advance to a certain promotional campaign. To make matters worse, many times the calculation of these recommendations is done using outdated customer segments. In this context, the challenge is to be able to speed up the process of calculation of recommendations so that users can be offered personalized suggestions of products. In order to fulfil this objective, the system implemented aims at providing the capability to make in-store consumer-tailored predictions to its users. As soon as an interaction between a user and a product arrives to the system, the analytic flow needed to calculate the recommendations for a certain user is triggered. In this way, the system is offering fresh recommendations to its users.
Who are the actors in the use case?
-
Data Owner: BigDataStack offers a unified Gateway to obtain both streaming and stored data from data owners and store them in its underlying storage infrastructure that supports SQL and NoSQL data stores. In the Connected Consumer use case the Data Owner is the food retailer.
-
Data Scientist: BigDataStack offers the Data Toolkit to enable data scientists both to easily ingest their analytics tasks by utilizing a declarative paradigm, and to specify their preferences and constraints to be exploited during the dimensioning phase regarding the data services that will be used (for example preferences for the data cleaning service)
-
Business Analysts: BigDataStack offers the Process Modelling Framework allowing business users to define their functionality-based business processes (through declaratively-defined models) and optimize them based on the outcomes of process analytics that will be triggered by BigDataStack.
-
Application Engineers and Application Service Owners: BigDataStack offers the Application Dimensioning Workbench to enable application owners and engineers to experiment with their applications and dimension it in terms of its data needs and data-related properties
-
Grocery Consumers: BigDataStack offers a performance environment enabling the achievement of the goals set by the use case in terms of resource provisioning and execution of analytics based on given service level objectives. Based on that, the customers of the food retailer are end-users of any of the applications of the food retailer (e.g. someone who is browsing in the e-commerce site) and will obtain the recommendation results following the successful execution of the analytics pipelines on BigDataStack infrastructure.
What are the key activities these actors are involved in?
The Connected Consumer use case can be broken down into the following main activities:
-
Definition of the analytics for the recommender. This activity aims at defining the business processes and the main analytical tasks that the recommender needs. The main actors in this activity are the Business Analyst and the Data Scientist.
-
Deployment of the application services. This activity aims at testing the capability to deploy on top of BigDataStack the application services of the recommender system. This activity applies to several actors: the business analyst, the data scientist and the application engineer. The former sets the time constraint for the whole process of recommendations. The data scientist calculates the service level objectives (SLOs) needed for each of the application services that compose the application. The application engineer tests several configurations before proceeding to do the final deployment.
-
Re-deployment of the application services. This activity demonstrates the capability of BigDataStack to make run-time adaptations in the configuration of the application services due to changes in the incoming data. The activity applies to the application engineer and the data owner.
-
Visualize recommendations. The data scientist wants to know which recommendations the system will propose for a given user. Main actors in this activity are the data scientist and the business analyst.
-
Provide recommendations. The External Applications need to provide recommendations to their users. E.g. a new App oriented to strengthen the loyalty of the food-retailer’s end-users wants to have the feature of suggesting new products to its users. Main actor in this activity is an end user of the client applications of the recommender.
-
Collect feedback. The Data scientist wants to have further information about the success or failure of the recommendations provided by the recommender. For this reason, the system is prepared for collecting feed-back from the external systems about the usage of the recommendations. Main actors in this activity are the data-owner and the data scientist. This activity aims at testing the capabilities of BigDataStack to ingest data through the Gateway.
Datasets
All data that have been used to train the analytic algorithm of the use case are corporate data provided by the food retailer. The dataset contains information about its clients. However, GDPR aspects have been taken into account before sharing the data with the consortium.
Next steps
In the next steps both business and integration objectives are expected to be fulfilled.
Business:
-
Introduce a clustering algorithm for Segmentation of customers in the current scenario that will help to produce more accurate recommendations. The current algorithm is calculating the recommendations based on the shop of the purchases.
-
Create a model for prediction of the next purchase of a given customer. The goal is to predict what the customers will need in their next purchase based on both the corporate data (history of purchases and product characteristics) and open data, i.e. predictive shopping list.
Integration:
-
Use CEP and Gateway for injecting corporate data in real-time such as order items, products, etc.
-
Full integration with the end-user tools of BigDataStack such as Process Modelling and Data Toolkit, since we want to be able to inject new analytical models.
Read the full report on Use case description and implementation
