A rose-like scent. A dry scent. A scent that might suit a composed person. Even when discussing the same perfume, we invoke not only ingredient names but also textures, personalities, and situations in which it might be worn. How can these words be organized?
In 2026, Han Zhou and colleagues analyzed 29,372 Chinese-language perfume reviews. They aimed to investigate the vocabulary and patterns of expression used to describe scents in Chinese, along with the emotional evaluations conveyed in the writing. Their study sought to add a large Chinese-language corpus to scent-language research that had primarily focused on English.
Analyzing reviews of 36 perfumes
The material consisted of reviews posted on the perfume-review website NoseTime from 2016 through September 20, 2023. The researchers selected 36 perfumes from four fragrance families, floral, fresh, woody, and amber, requiring a verifiable family classification and more than 300 reviews. The figure 29,372 refers to comments, not to that many people taking part in an experiment.
The study used three distinct methods. First, the researchers read 300 randomly selected reviews sentence by sentence and manually classified scent-related expressions. Next, they randomly selected 500 reviews per fragrance family and used a computational method to identify topic groupings from words frequently used together. Finally, all 29,372 comments were automatically classified as positive or negative.
Thus, the three analyses did not all examine approximately 30,000 reviews in the same way. The study combined close reading of a smaller set of texts with computation of patterns across a larger collection.
Three broad ways of describing scent
Reading and organizing the texts revealed three broad types of scent expression. The first linked scents to concrete objects or situations. Alongside naming sources such as rose or coffee, it included expressions evoking landscapes or scenes from past experience.
Naming odor-source objects was the most common approach in this material. Descriptions of scenes also appeared repeatedly, however, so the researchers organized them separately from expressions naming objects. Describing where a scent makes someone feel they are, as well as what it smells like, was a way of communicating scent.
The second type expressed scent qualities in abstract terms. Alongside intensity words, it included terms also used for taste and touch, such as sweet, bitter, and dry. There were also expressions projecting human personalities or impressions onto scents, such as passionate or mature.
The third type involved evaluation. This included not only preferences, such as liking or disliking a scent, but also judgments of suitability for a season, occasion, or type of person. A single review could describe an object, a scent impression, and a preference at the same time. Each review was not assigned to only one category.
After developing the classification, the researchers repeated the process with another 50 randomly selected reviews. No new categories emerged from these additional texts, leading them to regard the three types as a useful framework for organizing this material.
Different fragrance families prompted different topics
The computational topic analysis sought groupings based on combinations of words in the texts. The researchers compared results with different numbers of topics and chose a four-topic model based on interpretability and overlap in content. They named the topics by examining the words that appeared prominently in each.
For example, woody fragrances featured topics related to refreshing bitterness and a composed impression. Fresh fragrances prompted discussion of freshness, fruit, and citrus sweet-and-sour qualities; floral fragrances of gentle impressions and preferences; and amber fragrances of sweetness, richness, and spiciness.
Expressions evoking personality or femininity appeared across several families. These represent social images that reviewers associated with the scents, not a determination of the sex or personality of people who wear those perfumes.
An analysis combining reviews identified four topics: olfactory experience, personality and atmosphere, fresh impressions, and fruity fragrance. Such results indicate which things tend to be discussed together in writing. They are not measurements of perfume chemistry, nor do they confirm that materials mentioned in reviews are actually present in the formulations.
About 73% of comments were classified as positive
When all comments were assessed using Baidu’s text-analysis service, 21,520 were classified as positive and 7,852 as negative, approximately 73% and 27%. The positive proportion within each fragrance family was around 71–75%, and a statistically significant association with family was found.
However, these figures are not proportions of users answering a satisfaction question. They are classifications inferred from text. Moreover, the researchers selected perfumes with many reviews and examined only posted writing. The figures do not represent the preferences of people in China as a whole or demonstrate that a fragrance family causes a positive mood.
Automatic text classification can also misinterpret metaphor or context. The study did not report a separate validation comparing its judgments with human assessments of perfume reviews.
Scent language also contains memory and self-expression
The authors emphasize that scent descriptions were not limited to properties perceived by the nose. Expressions depicting scenes may involve memories and emotions, while references to personality or suitable occasions may reflect awareness of how wearing perfume presents oneself to others.
From this perspective, perfume, like clothing or jewelry, is discussed in relation to situations, moods, and personal images. However, the study measured writing. It did not experimentally verify reviewers’ memories or test whether perfume actually changed interpersonal impressions.
The varied expressions found within the same fragrance family may also reflect differences in experience and scent memory. However, this analysis alone cannot separate differences between individuals from changes in perception within an individual.
Chinese and English have not yet been compared directly
The three broad expression types overlap with earlier studies of English and other languages. At the same time, the authors argue that expressions referring to concrete objects and scenes, personal images, and social suitability were prominent in this Chinese-language material.
However, this was not a study that collected Chinese and English reviews of the same perfumes and compared them using identical methods. Earlier studies also differed methodologically, for example in classifying sentences versus examining relationships between words. The observed differences therefore cannot simply be established as differences between languages or cultures.
The automatic topic analysis also involved preprocessing decisions such as excluding single-character words. The structure that emerges may change according to which words are retained and how many topics are chosen. Rather than discovering the one correct classification of scent language, the study is better understood as providing a map for making sense of this material.
The research was supported by the National Natural Science Foundation of China, and the authors declared no competing interests.
Summary
Chinese-language perfume reviews included expressions comparing scents to objects or scenes, describing scent qualities or personal images, and evaluating preference or suitability. Topic patterns differed among fragrance families, and automatic classification identified approximately 73% of the overall comments as positive.
The study’s significance lies in treating scent language as more than a list of material names. Words about scent encompass not only sensation but also experiences associated with the scent and the image of oneself wearing it. The authors suggest that organizing this structure can provide a foundation for communicating scents, designing products, and developing computational models that handle scent-related text. The next challenge is to compare these findings with writing in other languages about the same perfumes and with experiments in which people actually smell and respond, to establish what differences in language reflect.
References
Zhou, H., Zhou, K., Zhu, Z., & Zou, L. (2026). The language of scents: a corpus-based study of Chinese perfume descriptions. Humanities and Social Sciences Communications, 13, 1149. DOI: 10.1057/s41599-026-07494-4. Full text. © The Author(s) 2026. The original article license is CC BY-NC-ND 4.0.
