Importance-performance analysis (IPA) is widely used for needs analysis, product positioning, and strategic planning in product design. Previous research on IPA often employs single-source data such as customer surveys or online reviews with unavoidable subjective bias. In contrast, product maintenance records provide objective information on product quality and failure patterns, which can be cross-validated with customers’ personal experiences from surveys or online reviews. In this paper, we propose an integrated framework for conducting IPA from online reviews and product maintenance records jointly. An attribute-keyword dictionary is first established using keyword extraction and clustering methods. Then, semantic groups, including product attributes and associated descriptions, are extracted using dependency parsing analysis. The sentiment scores of identified product attributes are determined by a voting mechanism using two pre-trained sentiment analysis models. The importance of product attributes in IPA is estimated from the impact of sentiments of each product attribute on product ratings with the extreme gradient boosting (XGBoost) model, while the performance is estimated from the sentiment scores of online reviews or the quality statistics from product maintenance records. In addition, we propose two methods to validate the IPA results, in which the IPA results are compared with the actual product improvements on the market or compared with the analysis of customer reviews from different time periods, respectively. The validated IPA results from online reviews and maintenance records are then integrated to obtain a more comprehensive understanding of customer needs. A case study of passenger vehicles is used to demonstrate the framework. The proposed framework enables automatic data processing and can support companies in making efficient design decisions with more comprehensive perspectives from multisource data.

Customers make quick judgments when shopping online based on how they perceive product design features. These features can be visual such as material or can be descriptive like a ‘nice gift’. Relying on feature perceptions can save customers time but can also mislead them to make uninformed purchase decisions, for example, related to sustainability. In a previous study, we developed a method to extract product design features perceived as sustainable from Amazon reviews, identifying that customer perceptions of product sustainability may differ from engineered sustainability. We previously crowdsourced annotations of French press reviews and used a natural language processing algorithm to extract the features. While these features may not contribute to engineered sustainability, customers identify the features as sustainable enabling them to make informed purchase decisions. In this study, we validate how our previously developed method can be generalised by testing it with electric scooters and baby glass bottles. Features perceived as sustainable for both products are extracted and second, participants are tested on interpreting the features using a novel collage approach. Participants placed products on a set of two axes and selected features from a list. Our results confirm that the proposed method is effective for identifying features perceived as sustainable, and that it can generalise for different products with limitations. Positively biased Amazon reviews can limit the natural language processing performance. We recommend that designers use our method when designing products to capture feature perceptions and help inform customer-oriented design decisions.