Research report: From consumer concerns to market returns: research on agricultural product quality credit based on online rice reviews

Every online review of a bag of rice is more than a complaint or a compliment. According to a study published on 5 October 2026 in the journal Agricultural and Food Economics, those reviews can be mined systematically to reveal what consumers actually care about, and the resulting signals can be linked to objective corporate records to explain why some rice sellers thrive in the e-commerce marketplace while others struggle. The research, led by Shuai Zhan and Xinyue Xu, analyzed 123,722 consumer reviews of rice products collected from the Chinese e-commerce platforms JD.com, Tmall and Taobao between 2022 and 2025, covering 92 enterprises.

The study addresses a persistent problem in food safety oversight. Traditional supervision of agricultural products in China has relied on a government-led model of daily sampling, on-site inspection and complaint handling, an approach the authors describe as costly, slow to respond to sudden quality incidents and difficult to extend to comprehensive coverage. Meanwhile, most academic work on e-commerce reviews has treated text mining and economic analysis as separate steps, limiting itself to sentiment classification and descriptive topic summaries without connecting micro-level consumer perceptions to macro-level corporate credit indicators. The new study attempts to bridge that gap by treating consumer-generated text as a dynamic, demand-side verification layer on top of official regulatory data.

Methodologically, the researchers applied BERTopic, a topic-modeling technique built on pre-trained Transformer language models that combines semantic embedding, dimensionality reduction and clustering. Unlike older bag-of-words models such as LDA, which struggle with the short, colloquial character of e-commerce reviews, BERTopic can identify latent themes without a preset number of topics. Before modeling, the team cleaned the raw corpus extensively: 48,532 system-generated default positive reviews and duplicates were removed, 25,410 reviews containing only emojis, punctuation or fewer than four characters were discarded, and 14,111 irrelevant or garbled entries were screened out. Word segmentation and stop-word filtering using combined Chinese lexicons yielded a final corpus of 35,669 valid reviews.

The topic modeling identified eight core themes of consumer concern. Three of them mapped directly onto measurable firm-level variables: a comprehensive product quality perception topic corresponded to whether the enterprise held quality management system certifications, a product traceability topic corresponded to whether the firm used traceability identifiers, and a packaging integrity topic corresponded to whether products used preservation packaging technology. The remaining topics largely reflected basic quality evaluations. A similarity heatmap of the topics showed that consumers frequently raise interrelated issues at once; the authors note that complaints such as rice weevils may be external manifestations of systematic credit risks in warehousing and other supply chain links, and could serve as dynamic early-warning signals rather than isolated problems.

To quantify market outcomes, the study used the number of new reviews within the observation period as a proxy for product sales, a standard approach in e-commerce research, since platforms enforce purchase-before-review rules so that each valid review typically reflects a real transaction. The authors acknowledge the limitation that review volume may capture consumer engagement bias rather than a one-to-one mapping of actual sales, but argue it provides a reliable relative measure given the sample size. The key mediating variable, agricultural product quality credit, was defined as an objective record of a firm’s compliance with statutory quality and safety standards, measured by each firm’s product sampling pass rate over the past year as announced by the State Administration for Market Regulation and provincial market supervision authorities. This construct is deliberately distinguished from actual product quality, from firm reputation and from subjective consumer perception.

Using ordinary least squares regression with heteroskedasticity-robust standard errors, the analysis found that information transparency and enterprise quality certification were each positively associated with review-based market activity, and both were also positively associated with agricultural product quality credit. Quality credit itself was positively associated with the sales proxy, and after controlling for it, the direct effects of transparency and certification became non-significant, a pattern the authors interpret as supporting a substantial mediating role. A Bootstrap test with 5,000 resamples supported the significant indirect effects. Packaging technology positively moderated the association between quality credit and market activity, meaning the link between credit and sales was stronger for firms using advanced freshness-keeping packaging, which the authors attribute to reduced quality loss during circulation and enhanced perceived credibility.

Robustness checks reinforced these patterns. Removing the control variables of enterprise age, scale and region did not alter the significance or magnitude of the core coefficients, and a natural logarithmic transformation of the right-skewed dependent variable, whose skewness was 0.601, preserved the significant positive associations at p < 0.01. To address potential selection bias, in which better-resourced firms might simply afford more certifications and packaging, the researchers used propensity score matching with 1:1 nearest-neighbor matching on enterprise scale, age and region, producing a balanced subsample of 88 firms in which both quality certification and information transparency retained significant positive associations with the sales proxy. A heterogeneity analysis showed that quality signals were associated with more pronounced market acceptance for smaller enterprises, with negative interaction terms for large firms.

Because regression assumes symmetric relationships, the study supplemented it with fuzzy-set qualitative comparative analysis, a configurational method that identifies combinations of conditions linked to an outcome. For high market acceptance, the solution consistency was 0.874 with coverage of 0.764, and two paths emerged. One configuration combined enterprise quality certification, agricultural product quality credit and advanced packaging technology. A second, more striking path paired the absence of information transparency and the absence of quality credit with advanced packaging, achieving a consistency of 0.935, suggesting that packaging can temporarily compensate for weak credit signals through physical protection and better display of labels and origin information. The authors caution that such marketing-driven acceptance may lack long-term sustainability without an underlying foundation of credit and trust. For non-high market acceptance, both identified paths involved the absence of agricultural product quality credit, often alongside missing certification, packaging technology or scale.

The authors frame the findings as illustrative of core correlation patterns among agricultural enterprises rather than proof of causal effects, noting that the cross-sectional design cannot fully exclude endogeneity despite the matching procedure. They also flag broader limitations: the sample covers a single product category, binary measures of transparency, certification and packaging simplify continuous organizational realities, and applying BERTopic to colloquial Chinese text introduces segmentation noise. Review-based credit evaluation may also disadvantage small producers and rural cooperatives with few online reviews, and commenting consumers skew toward digitally savvy users whose preferences may not represent rural or low-income groups.

Even with those caveats, the study sketches a practical vision for data-driven regulation. Integrating unstructured online reviews into formal credit evaluation frameworks could, the authors suggest, help supervision shift from reactive sampling toward proactive digital governance, shortening risk identification cycles and broadening coverage. For enterprises, the results reframe transparency and certification not merely as compliance costs but as strategic digital credit assets whose market value is amplified by logistics technology. The authors also note that consumer review-based monitoring and laboratory testing have complementary strengths: shoppers readily detect sensory defects such as aged rice but cannot perceive chemical hazards, so the review-based approach is positioned as a low-cost early-warning layer working alongside, not replacing, traditional quality control.

Subject of Research: Text mining of online rice reviews to link consumer quality concerns and corporate quality credit with e-commerce market performance

Article Title: From consumer concerns to market returns: research on agricultural product quality credit based on online rice reviews

Article References: Zhan, S., & Xu, X. (2026). From consumer concerns to market returns: research on agricultural product quality credit based on online rice reviews. Agricultural and Food Economics, 14(1), Article 80. https://doi.org/10.1186/s40100-026-00516-4

Image Credits: AI Generated

DOI: 10.1186/s40100-026-00516-4

Keywords: agricultural product quality, e-commerce reviews, BERTopic, food safety regulation, quality credit, rice, OLS regression, fsQCA, packaging technology, information transparency, quality certification, China