In the era of big data, the electric vehicle (EV) industry has witnessed rapid growth, particularly in the context of timeshare rental services. As a sustainable and innovative mobility solution, electric vehicle timeshare rental projects leverage internet-based platforms and GPS technologies to offer flexible, eco-friendly transportation options. However, despite governmental support and technological advancements, service quality issues such as inadequate charging infrastructure, vehicle availability problems, and delayed customer support persist, leading to user dissatisfaction. This study addresses these challenges by integrating text mining techniques with Quality Function Deployment (QFD) to identify customer demands and improve service quality for electric vehicle timeshare rental projects. Focusing on a representative case in China’s EV market, we demonstrate how user-generated online reviews can be harnessed to drive data-driven quality enhancements. By combining sentiment analysis, feature extraction, and QFD modeling, we provide a structured framework for translating customer feedback into actionable service improvements, thereby contributing to the sustainable development of the electric vehicle ecosystem in China and beyond.
The proliferation of electric vehicle timeshare rental services aligns with global trends toward green transportation and shared economies. In China, the electric vehicle sector has expanded significantly, driven by policies promoting new energy vehicles and smart city initiatives. However, the success of these services hinges on their ability to meet evolving customer expectations. Traditional methods for assessing service quality, such as surveys, often fall short in capturing real-time user sentiments. In contrast, online reviews offer a rich, dynamic source of data that reflects genuine user experiences. This study leverages text mining to analyze these reviews, extracting key customer demands and emotional cues. Subsequently, we employ QFD to map these demands onto specific service quality elements, enabling prioritized improvements. Our approach not only enhances the responsiveness of electric vehicle timeshare rental operators but also sets a precedent for data-informed quality management in the broader electric vehicle industry.
To begin, we collected a substantial dataset of user reviews from a prominent electric vehicle timeshare rental platform in China, focusing on a company analogous to EVCARD. The data comprised 995 reviews, which were preprocessed using tools like ROSTCM6 for stop-word removal and the Harbin Institute of Technology Language Technology Platform (LTP) for word segmentation and part-of-speech tagging. This preprocessing step was crucial for structuring the raw text and identifying salient features. A word cloud visualization highlighted frequent terms such as “vehicle,” “experience,” “cost,” and “service,” indicating primary areas of user concern. For instance, terms related to “electric vehicle performance” and “China EV infrastructure” appeared repeatedly, underscoring the importance of these aspects in the context of electric vehicle timeshare rentals.
Next, we applied regular expressions to extract feature-emotion word pairs based on predefined part-of-speech templates. These templates included patterns like “noun + adjective” and “verb + noun,” which effectively captured user opinions. For example, phrases like “battery life is poor” or “app is user-friendly” were identified as feature-emotion pairs. The extraction yielded 1,434 valid pairs, which were categorized into five broad themes: vehicle-related, software-related, cost-related, facility-related, and experience-related. The frequency distribution of these pairs is summarized in Table 1, illustrating the relative emphasis users place on different service aspects. Notably, experience-related pairs dominated, reflecting the holistic nature of user satisfaction in electric vehicle timeshare rentals.
| Category | Frequency | Percentage |
|---|---|---|
| Vehicle-Related | 212 | 23.4% |
| Software-Related | 75 | 8.3% |
| Cost-Related | 175 | 19.3% |
| Facility-Related | 55 | 6.1% |
| Experience-Related | 390 | 43.0% |
For sentiment analysis, we developed a customized lexicon incorporating positive and negative emotion words from established sources like the HowNet sentiment dictionary, supplemented with domain-specific terms relevant to electric vehicle services. Positive words were assigned a score of +1, while negative words received a score of -2 to account for their stronger impact on user perception. Adverbs and negation words were integrated to modulate sentiment intensity. The sentiment value \( F \) for a given feature-emotion pair was computed using the formula:
$$ F = f(x) \cdot g(x)^n \cdot \sum q(x) $$
where \( f(x) \) is the base sentiment value of the emotion word, \( g(x) = -1 \) for negation adverbs, \( n \) is the count of negation words, and \( q(x) \) is the weight of adverbs (e.g., “very” = 1.5). This approach allowed us to quantify user sentiments accurately. The comprehensive sentiment values for each category were calculated as the sum of positive and negative scores, normalized to percentages. For example, the negative sentiment for “chargingæ¡©” (charging piles) was notably high, indicating a critical pain point in China’s EV infrastructure. The results, depicted in Figure 1, reveal that aspects like “vehicle availability,” “additional fees,” and “customer service attitude” had higher negative sentiment, signaling urgent areas for improvement.
Building on these insights, we identified eight key customer demands for the QFD process: vehicle quantity, additional costs, deposit, network location, charging piles, parking spaces, customer service attitude, and special incident handling. The initial weight \( W_i(C_m) \) for each demand was determined by its frequency proportion:
$$ W_i(C_m) = \frac{\text{Number of feature-emotion pairs for } C_m}{\text{Total pairs}} \times 100\% $$
To refine these weights, we incorporated the average comprehensive sentiment value \( E(C_m) \), which represents the intensity of user dissatisfaction:
$$ E(C_m) = \left| \frac{E_p(C_m) + E_q(C_m)}{i + j} \right| $$
where \( E_p(C_m) \) and \( E_q(C_m) \) are the sums of positive and negative sentiment values for demand \( C_m \), and \( i \) and \( j \) are the counts of positive and negative pairs, respectively. The adjusted weight \( W_a(C_m) \) was then computed as:
$$ W_a(C_m) = W_i(C_m) \cdot E(C_m) \times 100\% $$
Finally, the relative weight \( W_r(C_m) \) was normalized to ensure the weights sum to 100%. The resulting customer demand weights are presented in Table 2, with “customer service attitude” emerging as the most critical demand, highlighting the human-centric nature of electric vehicle timeshare services in China.
| Customer Demand (\( C_m \)) | Initial Weight \( W_i(C_m) \) (%) | Average Sentiment \( E(C_m) \) | Adjusted Weight \( W_a(C_m) \) (%) | Relative Weight \( W_r(C_m) \) (%) |
|---|---|---|---|---|
| Vehicle Quantity | 1.985 | 1.358 | 2.696 | 18.02 |
| Additional Costs | 1.874 | 0.705 | 1.321 | 8.83 |
| Deposit | 1.544 | 1.107 | 1.709 | 11.43 |
| Network Location | 0.882 | 1.000 | 0.882 | 5.90 |
| Charging Piles | 0.992 | 2.000 | 1.984 | 13.26 |
| Parking Spaces | 0.992 | 0.666 | 0.661 | 4.42 |
| Customer Service Attitude | 2.867 | 1.711 | 4.905 | 32.80 |
| Special Incident Handling | 1.213 | 0.659 | 0.799 | 5.34 |
In the QFD phase, we constructed a House of Quality (HoQ) to translate customer demands into service quality elements. The left wall of the HoQ contained the eight customer demands with their relative weights, while the ceiling listed 17 service quality elements derived from literature review and expert consultations. These elements encompassed technological, operational, and strategic aspects, such as cloud server optimization, dedicated hotlines for incidents, dynamic pricing, and staff training programs. The relationship matrix \( R_{ij} \) in the room of the HoQ was populated by a QFD expert panel, which scored the correlation between each customer demand and service quality element on a scale from 0 (no correlation) to 9 (strong correlation). For instance, “customer service attitude” showed high correlations with “AI customer service hotlines” and “staff training systems.”
The self-correlation matrix \( P \) for the service quality elements was also established, capturing interdependencies among elements. Strong positive correlations (score 9) were noted between elements like “cloud server optimization” and “APP functionality upgrades,” while negative correlations (score -3) existed between “personalized service systems” and “standardized pricing.” The absolute importance \( S_{\text{abs}} \) of each service quality element was calculated using the formula:
$$ S_{\text{abs}} = R \cdot C $$
where \( R \) is the corrected relationship matrix incorporating self-correlations \( R = U \cdot P \), \( U \) is the original relationship matrix, and \( C \) is the vector of customer demand weights. The relative importance \( S_{\text{rel}} \) was then normalized. The rankings, shown in Table 3, indicate that “personalized service systems,” “staff training systems,” and “offline team enhancement” are the top-priority elements for improving electric vehicle timeshare rental services in China.
| Service Quality Element | Absolute Weight | Relative Weight | Rank | Importance Level |
|---|---|---|---|---|
| Personalized Service Systems | 102.64 | 0.0887 | 1 | 8 |
| Staff Training Systems | 97.55 | 0.0843 | 2 | 8 |
| Offline Team Enhancement | 93.15 | 0.0805 | 3 | 8 |
| Dedicated Incident Hotlines | 90.90 | 0.0785 | 4 | 7 |
| AI Customer Service Hotlines | 84.90 | 0.0733 | 5 | 7 |
| Differentiated Pricing Standards | 83.04 | 0.0717 | 6 | 7 |
| APP Data Accuracy | 69.11 | 0.0597 | 7 | 5 |
| Network Location Optimization | 68.65 | 0.0593 | 8 | 5 |
| Collaboration with Traffic Authorities | 66.41 | 0.0574 | 9 | 5 |
| Scheduling System Upgrades | 65.39 | 0.0565 | 10 | 5 |
| Government Policy Support | 58.74 | 0.0507 | 11 | 5 |
| Credit System Improvement | 56.95 | 0.0492 | 12 | 4 |
| Industry Alliance Development | 52.67 | 0.0455 | 13 | 4 |
| Network Facility Optimization | 50.35 | 0.0435 | 14 | 4 |
| Consumer Base Expansion | 43.18 | 0.0373 | 15 | 3 |
| APP Transparency on Fees | 39.60 | 0.0342 | 16 | 3 |
| Cloud Server Optimization | 34.49 | 0.0298 | 17 | 2 |
Based on these findings, we propose several strategic recommendations for electric vehicle timeshare rental operators, particularly in the context of China’s EV market. First, implement differentiated pricing models that account for vehicle type, usage patterns, and location. For example, dynamic pricing can incentivize users to pick up vehicles from low-demand areas, balancing fleet distribution. Second, enhance staff training programs to improve customer service attitudes and incident response capabilities. Regular workshops on electric vehicle maintenance and conflict resolution can reduce operational losses and boost user loyalty. Third, upgrade technological infrastructure, including AI-driven customer service hotlines and real-time APP updates, to address issues like vehicle availability and charging pile status. Fourth, foster collaborations with traffic management authorities to secure preferential policies, such as dedicated parking spaces and expedited violation processing, which can lower operational costs and improve service efficiency.
In conclusion, this study demonstrates the efficacy of combining text mining and QFD for service quality improvement in electric vehicle timeshare rental projects. By analyzing user reviews, we identified critical customer demands and quantified their importance through sentiment analysis. The QFD framework then enabled a systematic translation of these demands into prioritized service elements. This approach not only enhances customer satisfaction but also supports the sustainable growth of the electric vehicle industry in China. Future research could explore real-time data integration or cross-cultural comparisons to further refine service quality strategies for electric vehicle timeshare rentals globally.