Consumer Taste Report — '향수'

Gutman (1982) argued that consumers choose a concrete product attribute because of the consequence it delivers, and want that consequence because of a personal value. Reynolds & Gutman's (1988) laddering interview repeatedly asks "why does that matter to you?" to recover A→C→V chains, then aggregates them across respondents into an implication matrix that is drawn as a Hierarchical Value Map. Taste Finder splits the consequence layer into perception and emotion, giving a four-level ladder: attribute → perception → emotion → value. In sensory categories "smooth" (a perceived impression) and "comforting" (an evoked emotion) are genuinely different levels.

Laddering requires one-on-one depth interviews, so samples are typically a few dozen people. Here, instead of interviews, attribute and perception terms are extracted by lexicon matching from queries, AI Overview text and organic snippets, and each node is sized by measured search volume. In other words, "how many people voiced this attribute" is measured as search demand rather than respondent count. The links between layers follow conventional mappings from sensory science and the fragrance industry; that part alone is not measured, so it is labelled as derived inference.

It gives you a ladder for deciding what to say in copy, packaging and product pages. If you only claim attributes, differentiation evaporates the moment a competitor claims the same one; knowing which emotion and value an attribute ladders up to lets you position at a higher level. Useful for concept boards, brand essence definition, and structuring an advertising message hierarchy.

The inter-layer links are convention-based and can differ by category. Search captures only tastes that have already been put into words, so preferences people feel but never type stay invisible. And volume is the size of attention, not the intensity of liking.

• Gutman, J. (1982) A Means-End Chain Model Based on Consumer Categorization Processes. Journal of Marketing
• Reynolds, T. & Gutman, J. (1988) Laddering Theory, Method, Analysis, and Interpretation. Journal of Advertising Research
• Olson, J. & Reynolds, T. (2001) Understanding Consumer Decision Making: The Means-End Approach

The classical requirements for segmentation (Wind 1978; Smith 1956) are identifiability, substantiality and accessibility. Demographic-only segments are easy to reach but weak at explaining behaviour; psychographic-only segments explain well but cannot be reached. Crossing taste with demographics bridges the two, and the preference index (lift) used here is the standard over-indexing measure: how many times more this demographic group leans into this taste versus the average.

Gender and age distributions from ListeningMind's Korean cluster data are aggregated as a weighted average per taste family. Lift is that taste's demographic share divided by the overall share; above 1.15 is over-indexed, below 0.85 under-indexed. Tastes with no demographic sample behind them (derived from SERP text only) are explicitly flagged so they never get mixed with sample-backed ones.

It lets you set targeting and creative casting from data. It is especially useful for checking whether a taste really belongs to the age group you assumed, and it gives you grounds for aligning the demographic target of media planning with the taste target of the message.

Demographics are cluster-level estimates, not individuals. US and Japan locales have no demographic data, so the filter is disabled there. Lift is approximated from the product of marginals, so genuine interactions (a spike specific to women in their 20s) need separate interpretation.

• Wind, Y. (1978) Issues and Advances in Segmentation Research. Journal of Marketing Research
• Smith, W. (1956) Product Differentiation and Market Segmentation. Journal of Marketing

Schwartz (1992) found that ten basic value types, replicated across cultures, form a circumplex on two axes: self-transcendence ↔ self-enhancement and conservation ↔ openness. Adjacent values are compatible; opposing values conflict. Kansei engineering (Nagamachi 1995) translates physical product attributes into consumers' affective vocabulary as a design language. Overlaying the two lets you see, in one space, which value a taste is a means of expressing. Aaker's (1997) brand personality scale is used as a secondary coordinate for translating each taste into brand language.

Taste families are defined as clusters of affective terms, then projected onto coordinates on the Schwartz circumplex. Bubble size is measured search volume; the coordinates themselves are derived inference from the value mapping. Opposing quadrants are conflicting values, so a brand claiming both ends of a diagonal can be read as cancelling its own message.

It shows the centre of gravity of your positioning and the empty space around it on a single page. Putting competitors on the same map reveals where differentiation is available, and it separates 'adjacent moves' (feasible) from 'diagonal moves' (expensive) when considering a taste extension.

Coordinates are theoretical mapping, not measurement, so read relative arrangement rather than absolute position. Schwartz values are an individual-level theory, so applying them to category tastes involves analogy. The map shows the size of a taste, not its profitability.

• Schwartz, S. (1992) Universals in the Content and Structure of Values. Advances in Experimental Social Psychology
• Nagamachi, M. (1995) Kansei Engineering. International Journal of Industrial Ergonomics
• Aaker, J. (1997) Dimensions of Brand Personality. Journal of Marketing Research

Brand Concept Maps (John, Loken, Kim & Monga 2006) reconstruct consumers' association networks as a map, and Keller's (1993) brand knowledge structure explains brand equity through the strength, favourability and uniqueness of associations. Aspect-based sentiment analysis separates what an evaluation is about, aspect by aspect. The operational definition of a brand having become a symbol of a taste is this: people use it as the yardstick when describing something else.

Two things are kept apart. First, description language: taste vocabulary extracted from AI Overview text and organic snippets, showing how the market describes this keyword. Second, reference patterns: real searches of the form '{keyword}-esque', 'like {keyword}', '{keyword} style', which are direct evidence that consumers use the brand as a benchmark. Where description exists but references do not, the report says the keyword is described but not yet symbolic.

It judges, with evidence, whether the brand has become shorthand for a taste inside its category. Used for assessing extension headroom, choosing licensing and collaboration partners, and diagnosing the starting point for repositioning.

Reference patterns live mostly in other categories' search corpora, so a single-keyword capture under-detects them — zero references is not proof that no symbolism exists. AI Overview text is a generative summary, so the market's language and the model's language may be mixed.

• Keller, K. L. (1993) Conceptualizing, Measuring, and Managing Customer-Based Brand Equity. Journal of Marketing
• John, D. R., Loken, B., Kim, K. & Monga, A. (2006) Brand Concept Maps. Journal of Marketing Research
• Pontiki, M. et al. (2016) SemEval Task 5: Aspect Based Sentiment Analysis

Cramér's V normalises the chi-square statistic by sample size and degrees of freedom to measure association strength between nominal variables. It ranges from 0 to 1 and does not inflate with sample size. Because a chi-square test alone will call trivial differences 'significant' once samples are large, reporting effect size alongside significance is standard practice (Cohen 1988).

Contingency tables of taste × gender and taste × age are built from sample-backed tastes only, and V is computed for each. Bands follow convention: weak <0.1, moderate 0.1–0.3, marked 0.3–0.5, strong >0.5. If fewer than two sample-backed tastes exist, the computation is skipped and the report says so.

It gives a quantitative answer to "is there any payoff in cutting the target by gender and age?" A low V means taste and context targeting will outperform demographic targeting; a high V justifies demographic media buying. It also helps decide where to spend research budget.

This is association strength, not causation. The contingency tables convert search-volume weights into cell frequencies rather than being a strict probability sample, which is why effect size is reported and p-values are not.

• Cramér, H. (1946) Mathematical Methods of Statistics
• Cohen, J. (1988) Statistical Power Analysis for the Behavioral Sciences