Analyzing the impact of features embedded in customers’ reviews of refurbished mobile phones on customer satisfaction

Abstract

Tremendous amount of online reviews are posted every day on different online platforms which act as a valuable data reservoir for understanding customer satisfaction and analyzing customer's purchase intentions. Success largely depends on the retention of customers, which again depends on customer satisfaction (CS) levels, so understanding customer satisfaction is fundamental for the success of refurbishing products. Thus, this paper seeks to identify the features associated with customer satisfaction of refurbished mobiles by analyzing the online reviews collected from Amazon.com through ANN based prediction model. Later it was substituted with other classifiers namely Random Forest (RF), XG Boost (XGB), K-Nearest Neighbors (KNN) and Decision tree (DT) to assess the accuracy of our Refurbished Mobile Customer Satisfaction (RMCS) prediction model. As a result of analyzing 400,000 reviews on refurbished mobiles, it was found that ANN outperformed other classifiers in predicting RMCS.

Keywords

Customer satisfaction online reviews ANN refurbished mobile

1 Introduction

The rapid advancement in information technology (IT) has led to increased consumption of mobiles, as a result of which millions of tons of e-waste is produced every year causing hazardous impact on the environment. In order to mitigate the effects caused due to generation of e-waste, business firms are now switching to more sustainable options such as refurbishing. Refurbishing is the process of restoring the used product to new working condition by repairing its faulty or damaged parts and making some changes to enhance product's look and to extend its lifespan.¹ Refurbishment gives second life to the products which can be economically and operationally beneficial for the companies.² Most of the past literature focuses mainly on the operational dimensions, strategic and managerial dimensions. But, customer, which is one of the most crucial and essential element for the prosperity and success of the refurbishment process has not yet received much attention. Thus, it is crucial for the business firms to know their customers and to identify the key features which influence their satisfaction levels.

Customer satisfaction is defined as customers’ overall contentment and fulfillment with service or product and also reflects the customers’ perception of the value received compared to their expectations.³ Customer satisfaction with refurbished mobiles can vary based on several factors such as price, performance, quality, warranty, appearance, brand and seller reputation and other innovative features of mobile. Generally, customers tend to be satisfied with refurbished mobiles when they meet their expectations and provide good value for the price.

Thus, customer satisfaction is important to measure because high levels of satisfaction are associated with repetitive intentions to purchase and positive word of mouth.⁴ Previous studies measured customer satisfaction mainly through customer surveys.⁵ However, surveys are sometimes expensive and time consuming and may not capture the full context or nuance of a situation. Therefore, some other sources of data such as online reviews can be considered helpful to better understand and analyse customer satisfaction.

Online customer reviews provide unprecedented amount of information to customers in order to evaluate services before making a purchase and the customers can also share their experiences post purchase. Analyzing online reviews to understand customer satisfaction is therefore crucial and an interesting area of the research for several reasons. Firstly, online reviews may contain valuable information that can assist the firms in improving the sustainability of the refurbished items.⁵ Secondly, firms can determine customer preferences and the features or specifications that matter most to customers. Finally, firms can identify common issues or defects and work to improve their refurbishment processes, leading to higher customer satisfaction.

Understanding customer satisfaction from user generated reviews is classified as follows: (1) mining the customer satisfaction dimensions,⁶ and (2) modelling customer satisfaction.⁷ Mining reviews provide quick insights into customer sentiments and common issues but modelling can uncover hidden patterns and insights and provide a deeper understanding. Unlike mining online reviews, modelling reviews measure the impact of customer satisfaction dimensions on customer satisfaction. So, modelling is considered an effective way to predict customer satisfaction for refurbished mobiles.

Previous research studies have tried to analyse customer satisfaction using structural equation modelling and regression models.⁸ But these techniques being linear in nature solely assesses the linear association between the predictor and dependent variables and may oversimplify complex and complicated problems.⁹ Some other studies used gap analysis to measure the gap between customer's expectations and perceptions of the services as an indicator of service quality.¹⁰ Other studies used opinion mining and sentiment analysis,¹¹ Latent Dirichlet allocation (LDA) and topic modelling¹² and content analysis. Although these techniques identified important features of customer satisfaction but they failed to examine the importance of these features and their influence on customer satisfaction.

In order to overcome these advantages, an Artificial Neural Network (ANN) based prediction model was constructed to understand customer satisfaction by applying it on the collected online reviews. ANN is a fundamental tool in machine learning that is capable of modelling nonlinear relationships through a simple structure. ANNs are well suited for capturing the behavior of nonlinear systems with proper accuracy and low computational efforts. ANNs can even discover and model intricate patterns in data, allowing them to capture nuances that regression models might miss. ANNs can even capture context and sentiments better than many rule-based sentiment analysis method.

The varied outcomes of the prior models used suggest the necessity for additional research. Furthermore, to enhance the understanding of customer satisfaction through online reviews, the current paper is dedicated to address the following set of questions:

RQ1: Which are the refurbished smartphone features embedded in online customer reviews that are helpful in predicting RMCS?

RQ2: How can we explore the refurbished smartphone features by implementing the ANN-based model to predict RMCS?

RQ3: How can we assess the importance or significance of individual feature in estimating the RMCS?

To provide answers of the above mentioned questions, an ANN framework is suggested for rigorously studying and modelling customers’ satisfaction via online reviews from e-commerce website. Firstly, data was collected from Amazon.com and then this data was pre-processed to clean up the noisy data. Then, using CountVectorizer technique, the features and attributes associated with customer satisfaction were extracted from online reviews. And after that an overall RMCS was predicted by developing an ANN based modelling framework, with the derived features serving as input to the proposed model. Later, to validate the robustness and accuracy of the model proposed in this study, the results obtained through ANN model was compared with other classifiers like Random Forest (RF), XG Boost (XGB), K-Nearest Neighbors (KNN) and Decision tree (DT). Finally, the proportional importance of each extracted feature was examined to better understand the customer satisfaction.

ANN is the type of machine learning model based on biological neural network consisting of neurons and the weights representing the relationships within the data.¹³ K-nearest neighbor (KNN), a supervised machine learning technique, is used to make predictions according to labels of k nearest neighbors of test instances.¹⁴ Random Forest (RF), developed by Breiman in 2001¹⁵ is an algorithm used in machine learning for both regression and classification tasks.¹⁶ RF generates multitude of decision trees from sample data and the decision is determined by a majority vote of the individual trees.¹⁷ Decision Tree (DT) is a fundamental machine learning method which build a tree-like model consisting of three kinds of nodes and makes decisions at each node by evaluating the features and their values.¹⁸ Extreme Gradient Boosting (XGB) is applied for classification and regression problems to boost the weaker learner to become stronger using the decision tree algorithm.¹⁹ XG Boost aims to correct errors made by previous trees making it a robust and high- performance technique.

The subsequent sections of this study are organized in the following manner: Section 2 will delve into the literature review, while the methodology employed in the current study will be outlined in third section. Section 4 will reveal the findings of this paper. Theoretical and managerial contributions will be expounded upon in section 5. Discussions and conclusions drawn from the study's results will be laid out in section 6. Finally, in section 7, the limitations of the research and directions for future investigation are suggested.

2 Literature review

Online reviews refer to the evaluations and feedbacks posted on internet by customers about a product or service highlighting both positive and negative aspects, helping others make informed decisions before purchasing.²⁰ Online reviews has transformed the buying environment as customers often rely on online reviews when they are unable to judge a product in person in order to mitigate risks regarding quality of the product and seller's truthfulness.²¹ Also, online reviews are a valuable source of feedback for the firms as they provide them with the chance to broaden their understanding and knowledge of customer behavior, preferences, brand image and trends.²² So, it is important for researchers as well as e- marketers to develop a thorough understanding of customers from online customer reviews.

Measuring Refurbished Mobile Customer Satisfaction (RMCS) is essential for maintaining and growing a successful business while staying attuned to customer needs and preferences. Most of the previous studies focuses mainly on studying customers’ perceptions, customers’ purchase intentions and customers’ attitudes towards refurbished mobiles, not much attention is given to exploring customer satisfaction. Therefore, there is a need of further examination to gain better understanding of the factors influencing satisfaction level of customers towards refurbished mobiles. For this purpose, this study aims to expand knowledge about Refurbished Mobile Customer Satisfaction (RMCS).

Table 1 provides the brief summary of research done in the refurbished mobile literature.²³ Applied sentiment analysis to investigate costumer attitudes towards refurbished phones by analyzing online reviews collected from twitter and also proposed the most appropriate selling strategy. Their results showed that customers’ viewed warranty, seller's reputation, price and quality as the most important attributes.²⁴ Investigated the factors influencing customer satisfaction of the smartphones and employed structural equation modelling to discern the connections among these satisfaction related factors. By employing Kano model and ensemble neural network²⁵ assessed how customer sentiments impact various CSDs of customer satisfaction. LDA was used to extract CSDs from user opinions and the sentiment orientations of online reviews regarding the derived CSDs were identified using SVM. Using sentiment analysis¹¹ explored CSDs towards refurbished smartphones and provided in-depth insights into actual consumer behavior concerning refurbished mobiles. Their results indicate that products characteristics such as function, battery health and appearance are the features of refurbished mobiles that customers are more concerned about. Using sentiment analysis²⁶ identified and analyzed various factors that influence behavior of customers’ in adopting refurbished mobile phones in both emerging and advanced nations utilizing social media data.

Table 1.
Previous research related to refurbished mobile customer satisfaction.

Researcher Objective of the research Technique used Key attributes

Sharifi and Shokouhyar ²³ Investigates customer attitudes towards refurbished products Content, sentiment and descriptive analysis Battery; performance; screen; internal storage; camera

Suchánek and Sobota ²⁴ To model customer satisfaction of smartphones Structure Equation Modelling Design; customer satisfaction; image; hardware; smartphones; software

Bi et al.²⁵ Analyzing online reviews to model customer satisfaction Ensemble neural network combined with the impact-oriented Kano model ENNM; LDA; customer satisfaction dimensions; product development; SVM

Nasiri and Shokouhyar ¹¹ Exploring customer satisfaction
dimensions towards refurbished smartphones Latent Dirichlet allocation (LDA), sentiment analysis Refurbishing; CSDs; Smartphones;
Consumer behavior perception; comparative Reviews

Seifian et al.²⁶ Explores customers’ purchasing behavior towards refurbished mobiles Sentiment analysis Refurbishing; intercultural analysis; smartphones; opinion mining

Researcher	Objective of the research	Technique used	Key attributes
Sharifi and Shokouhyar ²³	Investigates customer attitudes towards refurbished products	Content, sentiment and descriptive analysis	Battery; performance; screen; internal storage; camera
Suchánek and Sobota ²⁴	To model customer satisfaction of smartphones	Structure Equation Modelling	Design; customer satisfaction; image; hardware; smartphones; software
Bi et al.²⁵	Analyzing online reviews to model customer satisfaction	Ensemble neural network combined with the impact-oriented Kano model	ENNM; LDA; customer satisfaction dimensions; product development; SVM
Nasiri and Shokouhyar ¹¹	Exploring customer satisfaction dimensions towards refurbished smartphones	Latent Dirichlet allocation (LDA), sentiment analysis	Refurbishing; CSDs; Smartphones; Consumer behavior perception; comparative Reviews
Seifian et al.²⁶	Explores customers’ purchasing behavior towards refurbished mobiles	Sentiment analysis	Refurbishing; intercultural analysis; smartphones; opinion mining

In contrast to earlier research, the suggested method in this research deviates in two significant aspects. According to our understanding, this research is the first to incorporate ANN based prediction model to measure RMCS through online reviews. ANN-based RMCS model exhibits the subsequent benefits in comparison to the previously used models: 1) ANN can model complex, non-linear relationships among different CSDs and customer satisfaction; and 2) ANN is a model that adapts based on data and is self-adjusting which doesn’t inherently make explicit prior assumptions about RMCS. By using ANN- based model to predict RMCS, this study surpasses the drawbacks identified in prior literature that intricate connections between various CSDs and customer satisfaction cannot be adequately conveyed. We have also substituted ANN with other classifiers like Random Forest (RF), XG Boost (XGB), K-Nearest Neighbor (KNN) and Decision tree (DT) to analyze the accuracy of our RMCS prediction model. Utilizing the developed model, we uncover the significance of individual features in forecasting customer satisfaction concerning RMCS, a dimension not adequately addressed in prevailing models relying on text-analytics, content analysis and sentiment analysis-based customer satisfaction prediction models.^11,26

3 Methodology

The objective of this paper is to uncover the factors associated with customer satisfaction regarding refurbished mobile phones, as expressed in online user reviews. We intend to extract these factors from electronic-word-of-mouth (eWOM) data and utilize them to forecast the satisfaction levels of customers buying refurbished mobile phones. Our research methodology involves four main stages: collecting data, preparing the data, extracting relevant features, and creating a predictive model for customer satisfaction among users of refurbished mobile devices (ref. Figure 1).

Figure 1.

Research framework.

Figure 2.

ANN Structure.

3.1 Data collection

Amazon.com stands out as a prominent e-commerce platform.²⁷ For this research, we utilized a dataset comprising 400,000 online reviews of refurbished mobile phones extracted from Amazon.com. These reviews were organized into a CSV file, containing details such as product title, brand, review content, product ratings, and price for unlocked mobile phones sold on the platform. The product ratings provided by users ranged from 1 star (indicating low satisfaction) to 5 stars (indicating high satisfaction).

3.2 Data pre-processing

In this study, we specifically considered reviews where customers provided both textual feedback and overall ratings for refurbished mobile phones. Reviews lacking product ratings were excluded from our analysis. A total of 413,840 reviews of refurbished mobile phones were chosen for in-depth examination. The review dataset served as input for Python-based code. Several procedural steps were undertaken in the text cleaning or textual pre-processing such as omission of stop words and punctuation, word tokenization, applying stemming, and altering the case. In order to provide the condensed version of the review content, frequently occurring words without substantial meaning, such as “and,” “between,” “or,” “is,” “am,” and “are” were removed. Subsequently, special characters and spaces were removed to clean the text. The eWOM text was then split into words, and the Porter stemmer library developed by²⁸ was incorporated into the Python code to perform the stemming process. Ultimately, the conversion of all the words to lowercase was implemented to minimize the probability of repeated word instances.

3.3 Extraction of features

Following the pre-processing of data, the CountVectorizer method, also known as the term frequency matrix, was employed to create a matrix representing word frequency. This technique transforms the text in an integer-valued matrix, generating a sparse matrix that depicts the word- frequency (Garreta & Moncecchi, 2013). Initially, the 10,000 most frequently used words in online reviews concerning refurbished mobile phones were identified. From these 10,000 features, our analysis focused on the 50 most common features that were identified from the pool of 413,840 customer reviews (eWOM) posted on Amazon. These features derived from the term frequency matrix are illustrated in the table. The 50 most commonly employed words capture the essential facets of refurbished mobile phones. They encompass various features reflecting customer sentiments about refurbished phones, including: (a) Mobile phone accessories: “charger,” “data cable,” “ejector,” “earphone,” and “case.”; (b) Battery attributes: “lithium,” “light,” “backup,” “nano,” and “mAh.”; (c) Financial considerations: “price,” “cost,” “cheapen,” “utilitarian,” and “worthy.”; (d) Performance and speed: “RAM,” “processor,” “motherboard,” “Bluetooth,” and “resolution.”; (e) Screen and display features: “brightness,” “FHD (Full High Definition),” “LCD (Liquid Crystal Display),” “AMOLED,” and “inch.”; (f) Storage specifics: “internal,” “external,” “GB,” “memory,” and “space.”; (g) Camera-related aspects: “lens,” “camera,” “video,” “images,” and “visibility.”; (h) Appearance considerations: “compact,” “slim,” “weight,” “design,” and “look.”; (i) Innovative mobile phone features: “biometric,” “facial recognition,” “NFC (near-field communication),” “MCC (Mobile cloud computing),” and “sensor.”; (j) Warranty features: “warranty,” “guarantee,” “quality,” “period,” and “extended.”^11,29

3.4 Refurbished mobile customer satisfaction (RMCS) prediction model

Consider a collection of reviews denoted as set R, comprising refurbished mobile phone assessments $(r_{1}, r_{2}, r_{3}, \dots \dots, r_{n})$ . Each review is represented as a set of mattributes, A = {a₁, a₂, … … , a_m}. The customer satisfaction prediction model generates overall refurbished mobile ratings (RMR) as the output, with mfeatures (A = {a₁, a₂, … … , a_m}) serving as the input. RMR is determined based on a five-point star rating system (from I to V stars), where I star signifies the lowest satisfaction level and V stars indicate the highest contentment with the refurbished mobile experience.²⁶ Emphasized the direct correlation between RMR and the overall performance of refurbished mobile phones. Consequently, RMR reflects the sentiments of previous customers regarding the refurbished mobile devices they have purchased. The predictive model for customer satisfaction with refurbished mobile phones can be expressed as follows:

R M R_{i} = R M C S (A)

(1)

where A represents the attribute vector corresponding to reviews r_i, where RMCS_i belongs to the set ℝ ⁿ for all i. The RMCS (Refurbished Mobile Customer Satisfaction) function is determined through the training and structural configuration of Artificial Neural Networks (ANN). The

RMCS concerning ANN is not directly accessible to users; rather, it is inherently embedded within the architecture of ANN, being a result of its design and training process.

In this research, we employed a single hidden layer feed-forward artificial neural network (ANN) consisting of 10 neurons, as illustrated in Figure 2. Our objective is to predict customer satisfaction with refurbished mobile phones using a review dataset

We identified 50 key features that reflect customer satisfaction with refurbished mobile phones. These features include items such as charger (a₁), data cable (a₂₂), ejector (a₃), earphone (a₄), case (a₅), lithium (a₆), light (a₇), backup (a₈), nano (a₉), mAh (a₁₀), price (a₁₁), cost (a₁₂), cheapen (a₁₃), utilitarian (a₁₄), worthy (a₁₅), RAM (a₁₆), processor (a₁₇), motherboard (a₁₈), Bluetooth (a₁₉), resolution (a₂₀), brightness (a₂₁), FHD (a₂₂), LCD (a₂₃), AMOLED (a₂₄), inch (a₂₅), internal (a₂₆), external (a₂₇), GB (a₂₈), memory (a₂₉), space (a₃₀), lens (a₃₁), camera (a₃₂), video (a₃₃), images (a₃₄), visibility (a₃₅), compact (a₃₆), slim (a₃₇), weight (a₃₈), design (a₃₉), look (a₄₀), biometric (a₄₁), facial recognition (a₄₂), NFC (a₄₃), MCC (a₄₄), sensor (a₄₅), warranty (a₄₆), guarantee (a₄₇), quality (a₄₈), period (a₄₉), and extended (a₅₀). These features were used as inputs in our ANN model, while the overall refurbished mobile ratings (RMR) served as the output.

Our goal was to minimize the prediction error of the model that predicts customer satisfaction with refurbished mobile phones (RMCS) concerning the feature set denoted as Z′ = {a₁, a₂, … … , a₅₀}. We trained this model to predict RMCS for the testing sample of reviews for refurbished mobile phones, represented as r_l.

4 Results

This study involved the execution of three experiments to examine how review text influences the prediction of customer satisfaction with refurbished mobile phones. Initially, we used Artificial Neural Networks (ANN) to predict satisfaction levels. Subsequently, we substituted ANN with other classifiers such as Random Forest (RF), XG Boost (XGB), K-Nearest Neighbors (KNN) and Decision tree (DT) to assess the accuracy of our Refurbished MobileCustomer Satisfaction (RMCS) prediction model. Their performance was compared with ANN followed by an evaluation of the importance of each attribute in predicting customer satisfaction.

4.1 Goodness of fit measure for ANN based prediction

The Table 2 presents the adequacy measures refurbished mobile customer satisfaction (RMCS) prediction model based on artificial neural networks (ANN).³⁰ Established that mean square error (MSE), root mean square error (RMSE), and residual sum of squares (RSS) are deemed satisfactory if they are less than or equal to 0.50, 0.75, and 0.08, respectively. Examining the table reveals that MSE (0.24), RMSE (0.49), and RSS (0.00095) fall within the acceptable range.

Table 2.
Goodness of fit results.

Measure Value

MSE 0.2469908

RMSE 0.4969817

RSS 0.0009531

R ² 0.8505486

Adjusted R ² 0.8505486

Measure	Value
MSE	0.2469908
RMSE	0.4969817
RSS	0.0009531
R ²	0.8505486
Adjusted R ²	0.8505486

The r-square and adjusted r-square share identical values, both set at 0.85, indicating that 85% of the variability is explained by the RMCS prediction model. Hence, the goodness of fit for the RMCS prediction model stands at 85%. These findings further substantiate the validity of proposed feature set in relation to refurbished mobile customer satisfaction (RMCS), as evidenced by MSE, RMSE, RSS, r-square, and adjusted r-square in the context of RMCS prediction.

4.2 Results comparison of ANN with four different classifiers

The results of the customer satisfaction prediction model for refurbished mobile phones (RMCS), utilizing classifiers such as ANN, KNN, RF, XGB and DT, are displayed in the Table 3. The data clearly demonstrates that ANN outperforms all other classifiers investigated in this study, achieving an impressive accuracy of 85.4%, as depicted in the accompanying Figure 3. These findings highlight that ANN is notably more effective in predicting RMCS compared to the other classifiers examined in the research.

Figure 3.

Accuracy with multiple classifiers.

Table 3.

Comparison results.

Name	Accuracy	F1	Recall	Precision	AUC
ANN	0.854	0.836	0.854	0.832	0.799
KNN	0.839	0.819	0.839	0.811	0.720
RF	0.844	0.772	0.844	0.712	0.781
DT	0.767	0.776	0.767	0.787	0.604
XGB	0.850	0.833	0.850	0.827	0.795

Previous research has emphasized that relying solely on accuracy is inadequate for evaluating classifier's reliability. Thus, F1-measure, area under the ROC curve (in which higher values denotes improved performance), recall, and precision for all the classifiers were also evaluated as detailed in the provided Table 3 .

Table 3 shows that ANN achieves superior F1-measure (83.6%), precision (83.2%), and recall (85.4%) in comparison to RF, KNN, DT, and XGB. This indicates that ANN outperforms RF, KNN, DT, and XGB as the most effective classifier for predicting customer satisfaction levels with refurbished mobile phones.

The ROC curves for ANN, RF, KNN, DT, and XGB are analyzed in the Figure 4, where the AUC values (area under the curve) for ANN (0.808) surpass those of KNN (0.720), RF (0.784), DT (0.599), and XGB (0.795). Just as observed with accuracy, ANN emerges as the top classifier for predicting customer satisfaction with refurbished mobile phones. It achieves a higher area under the curve (80.8%) compared to KNN (72.0%), RF (78.4%), DT (59.9%), and XGB (79.5%). This underscores that ANN efficiently predicts RMCS compared to the other classifiers utilized in this research.

Figure 4.

ROC curves for various classifiers.

In line with earlier studies, this research indicates that ANN exhibits superior performance compared to SVM, KNN and NB.¹³ The findings further confirm the effectiveness of the selected features in predicting customer satisfaction with refurbished mobile phones, as evidenced by ROC curves, recall, precision, F1-measure, and accuracy.

4.3 Feature-specific result assessment

This research involved examining how individual features contribute to predicting customer satisfaction with refurbished mobile phones. The goal was to comprehend the influence of different features derived from reviews on predicting customer satisfaction. In Figure 5, the significant importance of the 50 features employed in the analysis is illustrated.

Figure 5.

Relative importance of important attributes.

We observe that the terms “lithium,” “light,” “backup,” “nano,” and “mAh” are the five most crucial features impacting customer satisfaction with refurbished mobile phones. These features are frequently mentioned by users of refurbished mobile phones and are all related to the device's battery. This finding aligns with previous research, which emphasized the significance of cell phone batteries, a detail often noted by cautious customers.^31,32 Hence, our results find support in existing studies.

The attributes “amoled,” “brightness,” “FHD,” “LCD,” and “inch” play a significant role in predicting customer preferences for refurbished mobile phones, according to RMCS analysis. These features are identified as the second most frequently mentioned by customers. Notably, all these characteristics fall under the category of mobile phone screens. Previous research has consistently highlighted the importance of the screen in customer satisfaction and sales, as indicated in the studies executed by³³ and.³⁴ Hence, our findings align with the conclusions drawn in these earlier studies.

The factors “visibility,” “camera,” “video,” “lens,” and “images” hold significant influences on predicting customer satisfaction with refurbished mobile phones (RMCS).³⁵ Established a direct association between mobile camera quality and satisfaction levels of customers. Additionally, features such as “internal,” “memory,” “external,” “GB,” and “space” play a crucial role in overall customer satisfaction, all focusing on the memory aspect of refurbished mobile phones. This underscores the importance of the “camera” as a pivotal component in forecasting customer satisfaction for refurbished mobile devices.²⁶ Further corroborated this by indicating that battery life, internal memory, camera, and display were the components most commonly mentioned by customers in mobile phones.

The factors “Quickerbluetooth,” “motherboard,” “ram,” “processor,” and “resolution” play a crucial role in determining RMCS. These attributes focus on the speed and performance of refurbished mobile phones, which are the most common indicators of their functional quality, as noted in research by.²⁶ On the other hand, “NFC,” “biometric,” “Facialrecognition,” “sensor,” and “MCC” are significant predictors of RMCS, emphasizing innovative features in refurbished mobiles. Studies by²³ have shown that mobile phones with innovative features receive higher ratings from customers, directly correlating with their satisfaction levels.

Additionally, features like “Compact,” “weight,” “design,” “slim,” and “look” influence RMCS by highlighting the appearance of refurbished mobile phones. Investigation carried out by²⁶ and³⁶ has consistently emphasized the importance of appearance in shaping customer satisfaction.”

“The factors “Price,” “cheapen,” “utilitarian,” “worthy,” and “cost” are important attributes of refurbished mobile phones that significantly contribute to predicting customer satisfaction. These features reflect the financial considerations associated with refurbished mobile phones. Research by²⁹ has emphasized that financial motivations positively impact the likelihood of procuring refurbished smartphones.”

Customers’ contentment or satisfaction regarding refurbished mobile phones is significantly influenced by factors such as “warranty,” “guarantee,” “quality,” “period,” and “extended” services. The presence of these attributes underscores the pivotal role of warranty in predicting customer satisfaction with refurbished mobile phones. This observation finds support in previous research investigations carried out by²⁹ and.²⁶ Conversely, the significance of features like “charger,” “data cable,” “earphone,” “case,” and “ejector” is diminished in forecasting overall customer satisfaction when it comes refurbished mobile phones. This indicates that features related to accessories exert a diminished influence on customer satisfaction within the domain of refurbished mobile devices.

5. Managerial implications

Reviews significantly impact the market for refurbished mobile phones.²⁶ found that customers’ purchasing decisions are influenced by the reviews they read on websites. Therefore, managers should pay close attention to the content of these reviews, as it can enhance customer satisfaction with refurbished mobile phones. Efficiently addressing customer queries raised in reviews can improve customer satisfaction. This study has managerial implications for online businesses, suggesting that analyzing textual content in reviews can predict customer satisfaction with refurbished mobile phones (RMCS). This approach can also boost customer loyalty to refurbished mobile phones.³⁷ The study clarifies the practical application of our proposed RMCS prediction model for supporting managers in real-world situations.

Firstly, this paper emphasizes the need to prioritize customer satisfaction with specific features of refurbished mobile phones, namely the battery (“lithium,” “light,” “backup,” “nano,” and “mAh”), screen (“amoled,” “brightness,” “FHD,” “LCD,” and “inch”), and camera (“visibility,” “camera,” “video,” “lens,” and “images”). These features play a more significant role in predicting customer satisfaction (RMCS) compared to others.²³ Identified that these aspects strongly influence the purchasing decisions of refurbished mobile customers. Therefore, manufacturers of refurbished phones should concentrate on these features to enhance customer satisfaction and influence their buying behavior.

Secondly, the findings of this research underscore the significance of employing artificial neural networks (ANN) to forecast customer satisfaction with refurbished mobile phones (RMCS). The precision, accuracy, F1-score, recall, and AUC of these predictions can aid remanufacturers in identifying the features that most profoundly influence customer satisfaction, as indicated by their importance in RMCS prediction. This study also compared the results of ANN with those of other models such as RF, KNN, DT, and XGB.

Interestingly, the disparities between the outcomes of ANN and the XG boost classifier were relatively minor, suggesting that the XG boost-based prediction model can also effectively gauge customer satisfaction levels through reviews. Previous research has similarly emphasized the supremacy of ANN over traditional statistical methods in diverse fields like stock forecasting³⁸ and human capital management.³⁹

Moreover, the significant implication of the ANN-based RMCS prediction model presented in this study is that manufacturers can swiftly assess customer satisfaction levels regarding refurbished mobile phones by analyzing real-time reviews. This capability was previously unattainable through conventional statistical methods outlined in existing literature. Typically, manufacturers allocate a substantial portion of their budget to analyze customer feedback. However, the model introduced in this research offers a time and cost-saving solution, enabling manufacturers to efficiently predict customer satisfaction and thus, conserve resources.

Finally, the suggested ANN-based RMCS prediction model enables remanufacturers to determine the significance of individual features extracted from customer reviews in assessing satisfaction levels with refurbished mobile phones. This analysis of feature importance is crucial for developing ranking models for refurbished mobile phones based on detailed customer feedback, as emphasized by.²³

5.1 Theoretical implications

This research has significant theoretical implications for advancing the literature on customer satisfaction predictions in the remanufacturing electronic industry. Earlier investigations have presented various models to predict customer satisfaction, offering diverse insights. What sets this study apart is its unique approach. Initially, as per our understanding, this is the first study to utilize the characteristics found within customer reviews’ textual content to predict satisfaction levels regarding refurbished mobile phones. Secondly, in this study, we identifies the 50 most frequently used features regarding refurbished mobile phones. Thirdly, it utilizes these features as input in a novel feed-forward ANN-based model for predicting refurbished mobile customer satisfaction (RMCS). Fourthly, the study employs XG Boost, RF, NB, and KNN classifiers to validate the ANN-based model. The comparison revealed that ANN outperforms RF, KNN and DT, with XG Boost showing slightly comparable results. Additionally, the current study assesses the relative importance of 50 features in predicting customer satisfaction with refurbished mobile phones.

In recent times, there has been a notable increase in attention toward predicting the satisfaction of mobile customers, not only from researchers but also from practitioners.^40,41 The observations of this research reveal that customers place substantial importance on the quality of the battery, screen, and camera in refurbished mobile phones (ref. Figure 5). These factors contribute significantly to predicting RMCS, as highlighted in the Figure 5 on relative importance. Customers experience higher satisfaction when remanufacturers offer improved memory options, faster Bluetooth, the quickest motherboard, RAM, processor, and high resolution. Additionally, innovative features such as NFC, biometric authentication, facial recognition, sensors, and MCC play a vital role in predicting customer satisfaction with refurbished mobile phones.

Likewise, customers tend to give higher star ratings to refurbished mobiles that are lightweight, slim, visually appealing, and reasonably priced. This is because terms such as “compact,” “weight,” “design,” “slim,” “look,” “price,” “affordable,” “utilitarian,” “valuable,” and “cost” are commonly used by refurbished mobile customers in their reviews. These terms effectively contribute to predicting customer satisfaction. Furthermore, customers express satisfaction when they receive refurbished mobile phones with extended warranty, charger, data cable, earphones, case, and ejector, as these features also play a role in determining customer satisfaction with refurbished mobile phones.

Furthermore, in terms of methodology, particularly in the realm of text analytics, techniques like feature extraction, artificial neural networks (ANN), and XG Boost can certainly find applications in theoretical studies within the remanufacturing and online sectors.⁴² Employed neural networks and multiple regression analysis to identify key determinants of customer satisfaction for mobile phones. Conversely,⁴³ pinpointed important smartphone features affecting customer satisfaction using various machine learning algorithms. In this study, instead of traditional statistical methods like regression, SEM-PLS, or SEM, techniques such as ANN, XG Boost, RF, NB, DT, and CountVectorizer were utilized. A variety of software tools like Python, RapidMiner, MATLAB, etc., are easily accessible for the seamless and effortless use of ANN-based models on customer opinion datasets. The same approach is applicable to diverse sectors to anticipate satisfaction levels, including employee satisfaction, tourist satisfaction, dining experience satisfaction, job satisfaction, student satisfaction, and more.

6. Conclusions and discussions of results

Many Previous scholars underscored the importance of establishing a collaborative partnership between information technology (IT) departments and top management, a dynamic that holds increasing significance in the highly competitive landscape of online shopping.⁴⁴ Consequently, there is a growing necessity for adoption policies related to online customer reviews, particularly in the context of predicting satisfaction with refurbished mobile phones in online shopping environments.

This paper specifically focuses on the insights shared by past customers of refurbished mobile phones, recognizing it as a pivotal source for analyzing their emotions and forecasting their intent to repurchase specific products or services. The objective of this research is to fill the void in existing studies on customer satisfaction within the realm of refurbished mobile phones by retrieving the features that appear most often in the text of online reviews.

This research provides a methodical examination for determining Refurbished Mobile Customer Satisfaction (RMCS) using an online reviews dataset. The primary outcome of this study involves identifying the 50 most frequently mentioned features within reviews from previous customers of refurbished mobile phones. These features include terms such as “nano,” “lithium,” “light,” “mAh,” “backup,” “price,” “cost,” “cheapen,” “utilitarian,” “worthy,” “Ram,” “processor,” “motherboard,” “bluetooth,” “resolution,” “Brightness,” “FHD,” “LCD,” “AMOLED,” “inch,” “Internal,” “external,” “GB,” “memory,” “space,” “lens,” “camera,” “video,” “images,” “visibility,” “compact,” “slim,” “weight,” “design,” “look,” “biometric,” “Facialrecognition,” “NFC,” “MCC,” “sensor,” “warranty,” “guarantee,” “quality,” “period,” “extended,” “charger,” “datacable,” “Ejector,” “earphone,” and “Case.” These features are deemed crucial in influencing the satisfaction level of refurbished mobile customers.

The findings of this study align with previous research, which has also suggested that features related to battery, screen, camera, memory, speed, and performance significantly impact the satisfaction levels of refurbished mobile customers.^26,31,33

The research framework and the utilization of neural networks to predict Refurbished Mobile Customer Satisfaction (RMCS) using an online reviews dataset have successfully achieved the primary objective of demonstrating the significance of artificial neural networks (ANN) in customer satisfaction. Performance measures were calculated for the ANN-based RMCS prediction model, including Mean Squared Error (MSE) at 0.24, Root Mean Squared Error (RMSE) at 0.49, Residual Sum of Squares (RSS) at 0.00095 and adj R² is 85%. All these metrics fall within an acceptable range, explaining 85% of the variability in the ANN-based RMCS prediction model. This underscores the effectiveness of employing the feed-forward neural network to predict satisfaction levels of customers with refurbished mobile phones.

While acknowledging the significance of the outcomes, it is crucial to recognize that relying solely on a particular model is insufficient to assert the utility of ANN in RMCS prediction. The robustness and reliability of the model proposed in the study is validated by applying four alternative techniques (KNN, RF, DT, and XGB) to predict RMCS using a similar set of features. The comparison revealed that ANN, with an accuracy of 85.4%, outperformed KNN (83.9%), RF (84.4%), DT (76.7%), and XGB (85.0%), providing a compelling indication of its superior predictive performance.

A sequence of inquiries and findings from this study substantiate its practical contributions. The research proposes an innovative approach to identify the most prevalent dimensions within the extensive dataset of online reviews for refurbished mobile customers. These identified dimensions are subsequently utilized as inputs for the Artificial Neural Network (ANN)-based Refurbished Mobile Customer Satisfaction (RMCS) prediction model. Furthermore, the study determines the significance of each extracted feature by evaluating its contribution in determining customer satisfaction levels with refurbished mobiles.

7 Limitations and future research directions

The outcomes of this paper are constrained by a few drawbacks. Initially, the study exclusively employs a feed-forward Artificial Neural Network (ANN) for predicting Refurbished Mobile Customer Satisfaction (RMCS) and evaluates its accuracy against KNN, RF, DT, and XGB. However, future researchers might explore alternative models like Gaussian Naïve Bayes, Genomic classifier, Adaptive boosting, etc. Secondly, the satisfaction prediction model is tested solely on one online customer reviews dataset. Future studies could enhance the scope by incorporating additional datasets from diverse sources. Additionally, this research analyzes Amazon's refurbished mobile customer reviews dataset without filtering out fake reviews. In subsequent studies, practitioners may consider applying the same model on refined reviews datasets.

Footnotes

ORCID iDs

Shikha Tomar

Aakash Aakash

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

MacArthur

. Towards the circular economy, economic and business rationale for an accelerated transition. Cowes, UK: Ellen MacArthur Foundation, 2013; 1.

Atasu

Sarvary

Van Wassenhove

. Remanufacturing as a marketing strategy. Manage Sci 2008; 54: 1731–1746.

Anderson

Fornell

Lehmann

. Customer satisfaction, market share, and profitability: findings from Sweden. J Mark 1994; 58: 53–66.

Wang

Y-S

Liao

Y-W

. The conceptualization and measurement of m-commerce user satisfaction. Comput Human Behav 2007; 23: 381–398.

Farhadloo

Patterson

Rolland

. Modeling customer satisfaction from unstructured data using a Bayesian approach. Decis Support Syst 2016; 90: 1–11.

Guo

Barnes

Jia

. Mining meaning from online ratings and reviews: tourist satisfaction analysis using latent dirichlet allocation. Tourism Manage 2017; 59: 467–483.

Xiao

Wei

C-P

Dong

. Crowd intelligence: analyzing online product reviews for preference measurement. Inf Manage 2016; 53: 169–182.

Danaher

Mattsson

. Customer satisfaction during the service delivery process. Eur J Mark 1994; 28: 5–16.

Liébana-Cabanillas

Marinković

Kalinić

. A SEM-neural network approach for predicting antecedents of m-commerce acceptance. Int J Inf Manage 2017; 37: 14–24.

10.

Brown

Swartz

. A gap analysis of professional service quality. J Mark 1989; 53: 92–98.

11.

Nasiri

Shokouhyar

. Actual consumers’ response to purchase refurbished smartphones: exploring perceived value from product reviews in online retailing. J Retailing Consum Serv 2021; 62: 102652.

12.

Jelodar

Wang

Yuan

, et al. Latent Dirichlet allocation (LDA) and topic modeling: models, applications, a survey. Multimed Tools Appl 2019; 78: 15169–15211.

13.

Pradhan

Wuehr

Akrami

, et al. Automated classification of neurological disorders of gait using spatio-temporal gait parameters. J Electromyogr Kinesiol 2015; 25: 413–422.

14.

Fix

Hodges

. Discriminatory analysis. Nonparametric discrimination: consistency properties. Int Stat Rev/Revue Internationale de Statistique 1989; 57: 238–247.

15.

Breiman

. Random forests. Mach Learn 2001; 45: 5–32.

16.

Breiman

Friedman

Olshen

, et al. Classification and Regression Trees. Monterey, CA: Wadsworth and Brooks/Cole, 1984; Links.

17.

Suhanda

Nurlaela

Kurniati

, et al. Predictive analysis of customer retention using the Random Forest algorithm. TIERS Inf Technol J 2022; 3: 35–47.

18.

Tsami

Adamos

Nathanail

, et al. A decision tree approach for achieving high customer satisfaction at urban interchanges. Transp Telecommun 2018; 19: 194.

19.

AL-Shatnwai

Faris

. Predicting customer retention using XGBoost and balancing methods. Int J Adv Comput Sci Appl 2020; 11.

20.

Hennig-Thurau

Gwinner

Walsh

, et al.

Electronic word-of-mouth via consumer-opinion platforms: what motivates consumers to articulate themselves on the internet?

J Interact Mark 2004; 18: 38–52.

21.

Hankin

. The effects of user reviews on online purchasing behavior across multiple product categories. Master’s final project report, UC Berkeley School of Information, http://www ischool berkeley edu/files/lhankin report pdf. 2007.

22.

Balducci

Marinova

. Unstructured data in marketing. J Acad Mark Sci 2018; 46: 557–590.

23.

Sharifi

Shokouhyar

. Promoting consumer's attitude toward refurbished mobile phones: A social media analytics approach. Resour Conserv Recycl 2021; 167: 105398.

24.

Suchánek

Sobota

. Modelling Customer Satisfaction of Smartphones in the Czech Republic. Qual Innovation Prosperity/Kvalita Inovácia Prosperita 2023; 27.

25.

J-W

Liu

Fan

Z-P

,, et al. Modelling customer satisfaction from online reviews using ensemble neural network and effect-based Kano model. Int J Prod Res 2019; 57: 7068–7088.

26.

Seifian

Shokouhyar

Bahrami

. Exploring customers’ purchasing behavior toward refurbished mobile phones: a cross-cultural opinion mining of Amazon reviews. Environment. Dev Sustainability 2023; 26: 1–29.

27.

Choi

Leon

. When trust cues help helpfulness: investigating the effect of trust cues on online review helpfulness using big data survey based on the Amazon platform. Electron Commer Res 2023: 1–28.

28.

Porter

. An algorithm for suffix stripping. Program 1980; 14: 130–137.

29.

Mugge

Jockin

Bocken

. How to sell refurbished smartphones? An investigation of different customer groups and appropriate incentives. J Cleaner Prod 2017; 147: 284–296.

30.

Moriasi

Arnold

Van Liew

, et al. Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Trans ASABE 2007; 50: 885–900.

31.

Bag

Tiwari

Chan

. Predicting the consumer's purchase intention of durable goods: an attribute-level analysis. J Bus Res 2019; 94: 408–419.

32.

Haverila

. Mobile phone feature preferences, customer satisfaction and repurchase intent among male users. Australas Mark J 2011; 19: 238–246.

33.

Cheng

Qian

, et al. The effects of mobile applications’ passive and interactive loading screen types on waiting experience. Behav Inf Technol 2023; 43: 1–12.

34.

Liu

Zhong

, et al. The effects of customer online reviews on sales performance: the role of mobile phone’s quality characteristics. Electron Commer Res Appl 2023; 57: 101229.

35.

Salo

Olsson

Makkonen

, et al. Consumer value of camera-based mobile interaction with the real world. Pervasive Mob Comput 2013; 9: 258–268.

36.

Zhong

Liang

Cui

, et al. Using online reviews to explore consumer purchasing behaviour in different cultural settings. Kybernetes 2019; 48: 1242–1263.

37.

Belhadi

Kamble

Benkhati

, et al. Does strategic management of digital technologies influence electronic word-of-mouth (eWOM) and customer loyalty? Empirical insights from B2B platform economy. J Bus Res 2023; 156: 113548.

38.

Kurani

Doshi

Vakharia

, et al. A comprehensive comparative study of artificial neural network (ANN) and support vector machines (SVM) on stock forecasting. Ann Data Sci 2023; 10: 183–208.

39.

Khang

Gupta

Dixit

, et al. Data-driven application of human capital management databases, big data, and data mining. In: Designing workforce management systems for industry 40. Boca Raton: CRC Press, 2023, pp.105–120.

40.

Iluno

. Predicting mobile phone attributes that influence customer satisfaction using the Kano model. Niger J Manage Sci 2023; 24.

41.

Shaikh

Watto

Brahmi

, et al. Customer satisfaction and customer loyalty: an empirical case study on the impact of benefits generated through smartphone applications. Int J Public Sect Perform Manage 2023; 12: 119–139.

42.

Goode

Davies

Moutinho

, et al. Determining customer satisfaction from mobile phones: a neural network approach. J Mark Manage 2005; 21: 755–778.

43.

Imtiaz

Ben Islam

. Identifying significance of product features on customer satisfaction recognizing public sentiment polarity: analysis of smart phone industry using machine-learning approaches. Appl Artif Intell 2020; 34: 832–848.

44.

Chun

Y-Y

Matsumoto

Chinen

, et al. What will lead Asian consumers into circular consumption? An empirical study of purchasing refurbished smartphones in Japan and Indonesia. Sustainable Prod Consumption 2022; 33: 158–167.

Analyzing the impact of features embedded in customers’ reviews of refurbished mobile phones on customer satisfaction

Abstract

Keywords

1 Introduction

2 Literature review

3.2 Data pre-processing

3.3 Extraction of features

3.4 Refurbished mobile customer satisfaction (RMCS) prediction model

4.1 Goodness of fit measure for ANN based prediction

Table 2. Goodness of fit results. Measure Value MSE 0.2469908 RMSE 0.4969817 RSS 0.0009531 R 2 0.8505486 Adjusted R 2 0.8505486

5.1 Theoretical implications

6. Conclusions and discussions of results

7 Limitations and future research directions

Footnotes

ORCID iDs

Funding

Declaration of conflicting interests

References

Table 2.
Goodness of fit results.

Measure Value

MSE 0.2469908

RMSE 0.4969817

RSS 0.0009531

R ² 0.8505486

Adjusted R ² 0.8505486