方法对比

	Aspect-Based Review Mining	Implicit Reaction-Time Brand Measures
领域	市场营销	市场营销
方法族	Machine learning	Process / pipeline
起源年份	2004	1986
提出者	Minqing Hu & Bing Liu	Russell Fazio (affective priming); Anthony Greenwald, Brian Nosek & Mahzarin Banaji (D-score scoring)
类型	NLP pipeline for feature-level opinion mining of consumer reviews	Response-latency measurement pipeline for implicit brand associations
开创性文献	Hu, M., & Liu, B. (2004). Mining and Summarizing Customer Reviews. Proceedings of the Tenth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD '04), 168-177.	Fazio, R. H., Sanbonmatsu, D. M., Powell, M. C., & Kardes, F. R. (1986). On the automatic activation of attitudes. Journal of Personality and Social Psychology, 50(2), 229-238.
别名	Feature-Based Opinion Mining, Product-Feature Sentiment Analysis, Review Opinion Mining, Feature-Level Sentiment Summarization	Implicit Brand Association Measures, Response-Latency Brand Testing, Affective Priming for Brands, Implicit Brand Attitude Measurement
相关	3	3
摘要	Aspect-based review mining is a natural-language-processing technique that turns large volumes of consumer reviews into feature-level opinion summaries useful for product and brand insight. Rather than scoring a review as merely positive or negative overall, it identifies the specific product features, or aspects, that customers comment on, the battery life, screen, price, customer service, and so on, and determines the sentiment expressed toward each. Minqing Hu and Bing Liu's 2004 KDD paper, Mining and Summarizing Customer Reviews, defined the canonical pipeline: extract the frequently mentioned features, find the opinion words associated with them, decide each opinion's polarity, and produce a feature-by-feature summary of how many reviewers praised or criticized each aspect. This granularity is what makes the method valuable to marketers, because a four-star product can hide a beloved design and a hated battery, and only feature-level analysis reveals it. Applied across a brand's reviews, it yields a structured map of product strengths and weaknesses straight from the voice of the customer. It scales qualitative listening to thousands or millions of reviews that no team could read by hand.	Implicit reaction-time brand measures use how fast people respond, rather than what they say, to gauge the associations a brand automatically triggers. The logic comes from Russell Fazio's demonstration that strong attitudes are activated automatically: when a brand acts as a prime, it speeds responses to evaluatively congruent targets and slows responses to incongruent ones, and the size of that facilitation indexes the brand's implicit evaluation. Building on this, response-latency tasks pair brands with positive or negative words, with attribute categories, or with competing brands, and read off implicit associations from millisecond differences in reaction time. Anthony Greenwald, Brian Nosek, and Mahzarin Banaji's improved scoring algorithm turns these latency differences into a standardized D-score that is comparable across people and tasks. Because the measures tap associations that operate before deliberate editing, they capture brand equity that consumers may be unwilling or unable to report. The result is a behaviorally grounded, hard-to-fake complement to survey-based brand tracking.