The broad acceptance and deployment of smartphone and Internet of Things (IoT) technologies presents a first-class opportunity for strengthening product marketing via interaction between businesses and their in-store customers within physical stores. Knowledge and tracking of the physical proximity of customers to particular products can extend successful recommendation strategies for product marketing applied in online stores to physical stores. Thus, there is an opportunity for businesses to utilize the potential of smartphones and IoT for organizing advertising and promotion campaigns with better results compared to traditional techniques. This paper presents an IoT platform developed for proximity marketing, which collects information related to user locations in real time within the sales area and processed in correlation with historical data. The platform analyzes the aggregated data for optimal decision making by marketing departments and performs targeted interaction with consumers in real time through automated messaging. The IoT platform operation is based on Bluetooth beacon devices installed in the business sales area, combined with smartphones owned by customers. The aim of developing the ProximIoT platform is to improve the efectiveness of promotional activities put in place by businesses, as well as overall customer experience.
1. Introduction to consumers [2] as well as improved placement of products in the store, in areas that attract the most attention The concept of proximity marketing put forward by the [3].
ProximIoT 1 project relies on the intuitive notion that the ProximIoT is being developed by Terracom S.A., the location and movement of customers in relation to prod- department of Computer Science and Engineering, Uniucts is an indicator of purchasing interests. PrioximIoT versity of Ioannina, and the Institute of Computer Scicombines the advantages of physical (presence in the ence, FORTH. The system combines diferent tools and store) and digital stores (personalized information and technologies with the aim to recognize consumer prefbetter understanding of customer interests), thus creat- erences and subsequently address targeted advertising ing a new immersive experience and an efective channel recommendations. The system tracks and collects the for relevant and personalized advertising and marketing location of the customer inside the sales area with the information. Proximity marketing directly utilizes mo- use of IoT devices [4] (Bluetooth beacons) and a speciallybile devices to deliver advertising and marketing content; developed smartphone application. These location data thus, marketers take advantage of the ever increasing use are transferred to the Zastel IoT platform 2, which allows of smartphones by all consumers, sending them targeted the easy management of IoT devices and the development notifications and personalized ofers. This strengthens of IoT applications based on cloud technologies. It conloyalty and active participation of consumers, increases stitutes a vital part of the system providing all necessary motivation for direct in-store purchases [1], and reduces CRUD functions on the basic entities of the IoT platlabor-intensive marketing tasks. GDPR-compliant collec- form and the appropriate infrastructure, services and an tion and processing of consumer location data combined API to facilitate IoT applications development. Content with inferred user behavior (e.g., interests, preferences, Management System (CMS) has been developed which etc.) can lead to better targeting and personalized ofers is where the content of recommendations is derived, by processing collected data using machine learning techniques in order to generate targeted recommendations related to a customer’s interests. The development of profiling techniques ofers a personalized experience to Published in the Workshop Proceedings of the EDBT/ICDT 2023 Joint Conference (March 28-March 31, 2023, Ioannina, Greece) * Corresponding author. $ pmanousi@cs.uoi.gr (P. Manousis); dimitris.lvr@gmail.com (D. Louvaris)
© 2023 Copyright for this paper by its authors. Use permitted under Creative Commons License consumers based on their dynamic behavior within the 1 hCPWrEooUrctkReshtdoinpgpssIhStpN:/c1e:6u1r3-w/-0s.o7r3g/wACwttErwibUu.tpRiornoW4.x0oiInmrtekrinsoahtti.oocnpoal m(PCCroBYce4.0e).dings (CEUR-WS.org) 2https://www.zastel.com Input: Current user (), Current beacon ()
Output: The product to propose 1 _ = getProducts(, ); 2 _ = ∅; 3 foreach ∈ (()) do 4 _ = getProducts(, ); 5 foreach ∈ _ do
if ∈/ _ then 6 _ add(); 7 if bought then 8 . += similarity();
end 10 return topScore(_); else end end end . -= similarity(); sales area. The study of the dynamic behavior and movement of customers within the store (e.g., duration of stay near specific products) through pattern recognition can successfully lead to comprehension of customers’ buying power, motivation for purchases and habits, i.e., all elements that can increase business profitability. ports with visualisation of the movement of consumers and their interest in products. 2. System Architecture ProximIoT contains a ML and recommendations subsystem (M5) that receives data from M4, CMS and ProxThe ProximIoT architecture is modular, consisting of six imiotDB modules and applies ML techniques using Spark. subsystems that comprise a customizable and extensible M5’s functions include classifications of customers based system. The structure and information flow between on their choices and creating recommendations for purthem is shown in Fig. 1. chases based on customer history and profile similarity,
A core component is the ProximIoT database (Prox- navigation to points or aisles of the store with products imiotDB) which interacts with all subsystems and mainly of interest or discounts, and products in their wishlist manages location-related data. The IoT platform and available on the mobile application. smart push notification subsystem (M1) receive data from To achieve this, M5 performs a cosine similarity check the application that runs on the consumer’s mobile de- on the customers, based on the amount of time they vice, processes and stores the data on queues, from where spent on products, trying to find customers with similar they are consumed by the BigData Collection component buying interests as the current one. M5 selects the topk (M4). After processing, the platform forwards appropri- customers and locates the products they bought, that the ate marketing content to the client’s smartphone via push current customer has not yet bought. M5 ranks those notifications. The subsystem responsible for storing and products with a scoring function (based on the cosine managing information regarding products, customers, similarity) and proposes the one with the highest value ofers and transactions is the CMS (M2). to the user, as depicted in Algorithm 1.
The mobile application (M3) that interacts with Prox- Finally, a pattern detection component (M6) is responimIoT realizes the bidirectional communication with the sible for detecting interesting event patterns using the IoT platform, through which users automatically receive location-based events collected by the placed beacons at the corresponding information for each product based runtime. Pattern detection [5] is based on Esper3 Comon their location within a market store, as detected by plex Event Processing (CEP) system. Esper considers the beacons. The main functions of M3 besides forwarding most recent activity of a customer in a store and then product information are the projection of advertising using the Event Processing Language (EPL), a SQL-based campaigns for each store, product rating, and product language, M6 detects event patterns of interest which, comparison. The BigData collection and processing sub- in turn, trigger specific promotional notifications sent to system (M4) consumes data messages from Zastel and the customer through the smart push notification system. accepts product and customer data from CMS and Prox- For instance, if the customer is detected in the area of TVs imiotDB. Retailers can receive real-time insights and re- 3https://www.espertech.com/esper/ in a retail shop for at least 2 minutes, then a promotional message for a specific TV on sale is sent to the customer.
Overall, the information flow starts when M3 provides the location of the consumer to M1, while it also communicates with the CMS from where it receives information about a specific product, which the customer focuses on.
In addition, M1 continuously sends to M4 tracked user data (location, duration of stay in beacon’s range) for collection and filtering. M4 updates the recommendations (M5) and event pattern detection (M6), which in turn Figure 2: (a) Floor plan ; (b) Installed beacons analyse the information and provide the customer with personalised behaviour-based suggestions, which will be displayed on smartphone (M3) via M1. M1 communicates real customers. From that upcoming commercial pilot bidirectionally with M2, M3, M4, M5 and M6 subsystems we are expecting further validation of our experimental aided by the ProximIoTDB. results.
As already mentioned, ProximiotDB is required for the interconnection of the subsystems, through its API.
ProximiotDB stores the customers’ locations per session, 4. Conclusions and the aggregate of recommendations that took place, and acknowledgments useful detected patterns. ProximiotDB and CMS (that manages clients’ personal information and products’ in- In this report we described ProximIoT, a novel IoT-driven formation) constitute the entire system databases. From system for proximity marketing, aiming to drastically privacy perspective, note that no sensitive user informa- improve the efectiveness and automation of in-store tion (e.g. names, addresses, etc.) is stored in ProximiotDB; product marketing, advertisement, and promotion cameach user is represented by a unique random identifier. paigns.
The authors thankfully acknowledge funding by the 3. Current Status Greek Research Technology Development and Innovation Action “RESEARCH - CREATE - INNOVATE”, Operational Programme on Competitiveness, Entrepreneurship and Innovation (EΠ ANEK) 2014–2020, Grant T1E∆ KProximIoT is currently in an advanced development phase with several subsystems and functions completed and others in an optimization phase. Completed subsys- 04810. tems include the IoT platform with the APIs and CMS which have been deployed on the cloud, ProximiotDB 5. References along with the BigData collection module, deployed at the University of Ioannina, and the mobile application [1] P. Christodoulou, K. Christodoulou, A. S. Andreou, with many features and smart notifications continuing to A real-time targeted recommender system for superbe enriched. Furthermore, the recommendation system markets (2017). has reached the state where it can suggest products based [2] X. Liu, Y. Jiang, P. Jain, K.-H. Kim, Tar: Enabling on the customer’s current position and his similarity to ifne-grained targeted advertising in retail stores, in: the previous customers of the store (Alg. 1). A testbed Proceedings of the 16th Annual International Conferhas been developed at FORTH-ICS (Fig. 2) for the pur- ence on Mobile Systems, Applications, and Services, pose of applying integration tests on the implemented 2018, pp. 323–336. subsystems. We use Kontakt.io’s Bluetooth LE beacons, [3] F. D. Busche, Method and system for integrating spaconfigured to use iBeacon protocol with signal range tial analysis, and scheduling to eficiently schedule about 1 meter. The placement of beacons is made in a and monitor infrastructure maintenance, 2002. US way to avoid signal overlaps in space when possible. Patent 6,496,814.
Results from the controlled experiments with the cur- [4] P. Sethi, S. R. Sarangi, Internet of things: architecrent ProximIoT prototype indicate that tracking and ana- tures, protocols, and applications, Journal of Electrilyzing the location of customers with respect to promoted cal and Computer Engineering 2017 (2017). products over time can indeed provide useful insight as [5] C. Zeginis, K. Kritikos, D. Plexousakis, Event pattern to their interests and potential purchasing plans. Fur- discovery in multi-cloud service-based applications., thermore, a store located in Ioannina, Greece has already Int. J. Syst. Serv. Oriented Eng. 5 (2015) 78–103. installed on its premise Bluetooth beacons, and assigned on them the nearby products, so as to test ProximIoT with