| dc.contributor.author | Arthur, Esther | |
| dc.date.accessioned | 2026-11-13T19:34:01Z | |
| dc.date.graduationmonth | December | |
| dc.date.issued | 2026 | |
| dc.description.abstract | Plant-based meat alternatives (PBMAs) have gained attention as sustainable alternatives to animal-based products; however, their sensory quality remains a major barrier to consumer acceptance. Conventional sensory testing provides controlled and reliable data but can be time-consuming and resource intensive. AI-assisted analysis of online reviews may offer a rapid, complementary approach by extracting sensory information from consumers’ real-world product experiences. This study compared conventional consumer testing with AI-assisted online review analysis to identify drivers of liking for plant-based chicken products. In a controlled consumer study, 100 participants evaluated six plant-based chicken products using hedonic scales and Check-All-That-Apply (CATA) questions, followed by ANOVA, penalty-lift analysis, and Partial Least Squares Regression (PLSR). Attributes identified as drivers of liking across the penalty-lift and PLSR analyses included “meaty/savory,” “chicken-like,” “spice/herb,” “toasted/brown,” “crispy,” “tender,” “crunchy,” and “moist/juicy,” while “cardboard” was associated with disliking. For the AI-based approach, ChatGPT analyzed 631 Amazon reviews collected using Octoparse. After cleaning, 182 sensory-relevant reviews were retained. Qualitative analysis of review text identified positive drivers such as “juicy,” “moist,” “crispy,” “crunchy,” “soft,” “firm,” “savory,” “salty” and “seasoned,” while negative drivers included “dry,” “chewy,” “rubbery,” “tough,” “greasy,” “oily,” “bland,” and “spongy.” Quantitative analysis of attribute mentions and star ratings using PLSR showed that “crunchy,” “juicy,” “crispy,” and “fishy” were positively associated with higher star ratings, whereas “rubbery,” “tough,” and “salty” were negative predictors. In addition, based on product ingredients, AI inferred desirable attributes such as “fibrous,” “juicy,” “savory,” “brothy,” “toasted,” “allium/onion,” “moist,” and “cohesive” as potential drivers of liking, and undesirable attributes such as “rubbery,” “gummy,” “crumbly,” “chalky,” “pasty,” “dry,” “beany,” “green,” “artificial,” and “flat” as potential drivers of disliking. The controlled consumer test identified more attributes that significantly differentiated the products, (24 of 32) using Cochran’s Q test, compared with six of 12 attributes identified from the quantitative AI analysis of online reviews using Pearson’s chi-square test. Nevertheless, the methods aligned for several discriminating descriptors, including “chicken-like,” “chewy,” “crispy,” and “crunchy.” Multiple Factor Analysis further demonstrated substantial agreement between datasets, with an RV coefficient of 0.781. Comparison of the identified drivers of liking and disliking between the consumer study and quantitative AI approach showed that “juicy” and “crispy” were consistently associated with higher liking scores and star ratings, respectively, while “tough” was associated with lower liking scores and star ratings. In contrast, “fishy” was positively associated with star ratings in the online review analysis but negatively associated with liking in the consumer study. This discrepancy highlights a limitation of online review analysis, including potential ambiguity in the meaning of sensory terms across consumers and testing contexts. Sensory descriptors do not always indicate whether an attribute was perceived positively or negatively, and star ratings may reflect the overall product experience rather than the effect of an individual attribute. Overall, AI-assisted analysis of online reviews captured sensory patterns comparable to controlled consumer testing and may provide a rapid complementary tool for PBMA product development. However, because online reviews are uncontrolled and vary in sensory detail, AI-generated findings should be interpreted alongside findings obtained using conventional sensory methods before product optimization. | |
| dc.description.advisor | Martin Talavera | |
| dc.description.degree | Master of Science | |
| dc.description.department | Department of Food, Nutrition, Dietetics and Health | |
| dc.description.level | Masters | |
| dc.identifier.uri | https://hdl.handle.net/2097/47506 | |
| dc.language.iso | en_US | |
| dc.subject | ChatGPT | |
| dc.subject | Drivers of liking | |
| dc.subject | Check-All-That-Apply | |
| dc.subject | Sensory Science | |
| dc.subject | Plant-based meat alternatives | |
| dc.subject | Online reviews | |
| dc.title | AI-Based Online Review Analysis Versus Consumer Check-All-That Apply (CATA) for Drivers of Liking in Plant-Based Meat Alternatives | |
| dc.type | Thesis |
English
العربية
বাংলা
Català
Čeština
Deutsch
Ελληνικά
Español
فارسی
Suomi
Français
Gàidhlig
ગુજરાતી
हिंदी
Magyar
Italiano
Қазақ
Latviešu
मराठी
Nederlands
Polski
Português
Português do Brasil
Русский
Srpski (lat)
Српски
Svenska
தமிழ்
Türkçe
Yкраї́нська
Tiếng Việt
繁体中文