Consumer behavior analysis based on Internet of Things platform and the development of precision marketing strategy for fresh food e-commerce

OID coding

A proposed system, based on IoT technology, for product identification is presented and its specific structure is illustrated in Fig. 1. The scheme is designed to be compatible with the OID coding methods of other identification systems, thereby enabling unique product identification and eliminating any ambiguity in traceability objects  (Sheykhi, Ashtiani & Khajehoddin, 2021). Incorporation of GPS/GIS technology during transportation guarantees timely and secure delivery of products to consumers. Additionally, through the logistics information platform, consumers can gain access to pertinent information about the aquatic products in circulation, including their source, production and processing procedures, transportation and distribution, and after-sales service. This information can also be utilized to trace problematic products through the OID system, thus enhancing the quality of customer service  (Wei & Lv, 2019). Key traceability information is meticulously recorded, along with a digital summary of the production, distribution, retail, and regulatory processes involved in associated consumer electronics products. The blockchain extranet traceability solution is employed to generate digital summaries of the detailed production, logistics, distribution, and retail information pertaining to each stage in the supply chain of consumer electronics OID identifiers. The detailed traceability information is then stored in the respective identification management servers of traceability companies.

This system allows users to retrieve the key traceability information of consumer electronics products stored on the blockchain. Furthermore, the national public service platform of IoT Identification management enables users to access the detailed traceability information of these products. By comparing the digital abstracts, users can determine whether any tampering has occurred, thus ensuring the transparency, immutability, and traceability of all product information throughout the supply chain process.

RFID application

The sales link in the cold chain logistics of fresh products serves as the terminal link and is the only link that directly connects with the consumers. The manner in which data and information regarding the aquatic products are processed in this link can have a direct impact on the sales volume and traceability of these products. The system employed in this link is depicted in Fig. 2 and involves the use of RFID technology for reading the product information during the stocking and sales process. Temperature sensors are also utilized to monitor temperature changes in the aquatic products area (Casella, Bigliardi & Bottani, 2022). A video monitoring system is implemented to record the sales process of the aquatic products, and 3G and 4G communication technology is used to transmit information regarding the sales of fresh products.

To ensure the seamless delivery of fresh products to the supermarket, it is essential to verify the alignment between the transported products and the supermarket’s order records. To accomplish this, a handheld RFID reader is employed to extract information from the RFID tag affixed to the aquatic products. This information encompasses the product name, origin, processing manufacturer, storage warehouse, and transportation details. Subsequently, this information is amalgamated with product sales data, encompassing shelf time, shelf life, sales quantity, and the responsible party’s name, and integrated into the RFID tag. Once the correctness of the information is verified, the product is then handed over, stored, and placed on the shelf. To facilitate this process, compact fixed read-write devices are installed on the supermarket shelves, capable of automatically retrieving pertinent information when the aquatic products are either added to or removed from the shelves. This information is subsequently uploaded to the supermarket’s background computer system.

In the Internet of Things system, it is an effective method to cluster users by using k-means algorithm combined with RFID tag data. Firstly, the RFID tag data used by the user is obtained by RFID technology, including user identification, timestamp and location information. Then, these data are preprocessed, including cleaning, denoising, and handling missing values, to ensure the accuracy and completeness of the data. Next, appropriate features are extracted from the RFID tag data, such as the user’s behavior pattern, usage frequency, and time distribution. The extracted features are standardized to eliminate the differences between the features. Then, the k-means algorithm was applied to cluster the standardized feature data, the number of clusters K was set, and the clustering results were optimized iteratively. By analyzing the clustering results obtained by k-means algorithm, the characteristics and behavior patterns of different user groups can be understood. Finally, according to the clustering results, personalized services, recommendation strategies and market positioning were formulated to meet the needs of different user groups. This clustering method based on RFID tag data and k-means algorithm can help the iot system to better understand the user group, provide personalized services and optimize decision-making, thereby improving user experience and system performance.

Algorithm description

First, the initial clustering center of the k-means algorithm is improved by the following procedure.

Definition 1: The average distance of all sample elements in the data set $\mathrm{Z}=\left\{{\mathrm{x}}_1,{\mathrm{x}}_2,\dots ,{\mathrm{x}}_{\mathrm{n}}\right\}$ is: (1)${\displaystyle {\mathrm{d}}_{\mathrm{c}}=\frac{2}{\mathrm{n}\left(\mathrm{n}-1\right)}\sum _{\mathrm{i}=1}^{\mathrm{n}}\sum _{\mathrm{j}=\mathrm{i}+1}^{\mathrm{n}}{\mathrm{d}}_{\mathrm{ij}}}$ where, d_ij is the Euclidean distance data object x_i and x_j.

Definition 2: The density value ρ_i of x_i is the number of sample objects in the region centered on x_i and with d_c as radius.

Definition 3: The local density of x_i is calculated as: (2)${\displaystyle {\rho }_{\mathrm{i}}=\sum _{\mathrm{i}\ne \mathrm{j}}\chi \left({\mathrm{d}}_{\mathrm{ij}}-{\mathrm{d}}_{\mathrm{c}}\right)}$ when χ(x) > 0, χ(x) = 1 , otherwise χ(x) = 0 . (3)${\displaystyle {\text{dist}}_{\mathrm{i}}=max\left\{min\left({\mathrm{d}}_{\mathrm{il}},{\mathrm{d}}_{\mathrm{i}2},\dots ,{\mathrm{d}}_{\mathrm{in}}\right)\right\}\left(\mathrm{i}=1,2,\dots ,\mathrm{n}\right)}$ where: d_i1, d_i2 are the Euclidean distances from the samples i to v₁ and v₂.

In this article, Eq. (2) is used to find the largest density point in the sample ρ_i and use it as the first initial clustering center c₁. Then I select the remaining initial cluster centers according to the maximum-minimum distance criterion combined with the density method; The sample object x_j is selected by Eq. (3), and the object with the highest density among all sample points within the range of its average distance d_c is selected as the second initial clustering center c₂. The process is repeated until the KTH initial clustering center c_k is found.

The subsequent phase involves enhancing the calculation of the similarity metric. Initially, the similarity measure between the samples is determined by weighing the distance between them. In case the information entropy of a feature is meager, it indicates the feature’s indistinguishability; hence, it may be attributed a reduced weight. Conversely, greater information entropy calls for higher weighting of the feature to augment the clustering efficacy (Benmammar, Taleb & Krief, 2017).

The specific algorithm steps of the improved k-means algorithm are as follows:

Input: Data set Z , number of clusters K;

Output: clusters formed by clustering Ci, i = 1, 2, …, K

Step 1: K initial clustering centers were determined using a density-based approach combined with the maximum and minimum distance methodology.

Step 2: The weight distance formula was utilized to calculate the distance between each sample xi in the dataset and its corresponding clustering center ci.

Step 3: Samples were classified into their respective clusters based on the weight distance.

Step 4: The mean value of the objects within the same cluster was computed and used to update the clustering center.

Step 5: The process was repeated from Step 2 to Step 4 until the clustering center converged or reached the maximum iteration limit.

Marketing strategy development

The tasks can be categorized into three groups based on the personalization scheme, data governance, and analysis of task recommendations: (1) Calculation of user behavioral preferences, (2) analysis of purchase preferences for fresh food (Zhang et al., 2019), and (3) exploration of user behavior and purchase patterns in the dimensions of time and age when all user profiles have similar work requests. When ample memory resources are available to execute the task, the algorithm for task completion time and the distribution of user consumption behavior may be influenced. Thus, I assume that these factors follow a normal distribution. After calculating the similarity, it can be dynamically adjusted across dimensions to align with the corresponding marketing strategy and facilitate recommendation completion.

In the context of the Internet platform, the publicly available dataset is typically utilized for collection purposes, followed by processing and cleaning of the data, and analysis of user portraits and fresh goods purchasing behaviors. The user portrait is mainly based on the personalized purchase preferences of the user and their publicly expressed preferences. The recommendation strategy for user portraits, age distribution, and other relevant factors is then developed to generate responses and facilitate real-time recommendations under the IoT platform, as depicted in Fig. 3.

The recommended strategies generated by the platform are subjected to analysis. This analysis involves a comprehensive examination of the strategies’ components, such as their proposed actions, resource allocations, or decision-making processes. The aim is to gain a thorough understanding of the strategies’ structure, objectives, and potential implications within the given context.

Subsequently, the performance of the recommended strategies is evaluated using two specific methodologies: k-means clustering and the double-closed value method. The k-means clustering technique groups the strategies based on their similarities or patterns, allowing for a better understanding of strategy clusters and their distinct characteristics. This analysis enables the identification of different strategy clusters and the assessment of their respective performance. (4)${\displaystyle Q=\sum _{j=1}^n{q}_i\cdot L,0\le q_i\le Q.}$

Additionally, the double-closed value method is employed to evaluate the effectiveness and accuracy of the recommendations. This method involves the identification and analysis of frequent patterns or behaviors within the recommended strategies. By discovering these frequent patterns, the method provides insights into the consistency and reliability of the recommendations. The evaluation of the recommendations is represented by a variable denoted as Q, which serves as a measure of their quality or effectiveness. Moreover, the length of the recommendation is denoted by the variable L, which captures the extent or complexity of the recommended strategies.

In situations where the effectiveness and accuracy of the recommended strategies are unclear, a revision and dynamic adjustment of the strategies can be performed. This adaptive approach allows for continuous improvement and refinement of the strategies based on the evaluation results. By iteratively assessing and refining the recommended strategies, the aim is to enhance their performance and align them more effectively with the desired goals and objectives.