Learning Attitudes and Attributes from Multi-aspect Reviews

By Julian McAuley, Jure Leskovec, Dan Jurafsky. 2012

327 citations recorded by the source corpus.

Most online reviews consist of plain-text feedback together with a single numeric score. However, understanding the multiple `aspects' that contribute to users' ratings may help us to better understand their individual preferences. For example, a user's impression of an audio book presumably depends on aspects such as the story and the narrator, and knowing their opinions on these aspects may help us to recommend better products. In this paper, we build models for rating systems in which such dimensions are explicit, in the sense that users leave separate ratings for each aspect of a product. By introducing new corpora consisting of five million reviews, rated with between three and six aspects, we evaluate our models on three prediction tasks: First, we uncover which parts of a review discuss which of the rated aspects. Second, we summarize reviews by finding the sentences that best explain a user's rating. Finally, since aspect ratings are optional in many of the datasets we consider, we recover ratings that are missing from a user's evaluation. Our model matches state-of-the-art approaches on existing small-scale datasets, while scaling to the real-world datasets we introduce. Moreover, our model is able to `disentangle' content and sentiment words: we automatically learn content words that are indicative of a particular aspect as well as the aspect-specific sentiment words that are indicative of a particular rating.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

Measuring Association Between Labels and Free-Text Rationales (2020, background): “…Inference (Lehman et al., 2019), Movie Reviews (Zaidan and Eisner, 2008), MultiRC (Khashabi et al., 2018), LGD (Linzen et al., 2016), 20 News (Lang, 1995), Amazon Re-views (McAuley and Leskovec, 2013), Beer Re-views (McAuley et al., 2012), BoolQ (Clark et al., 2019), FEVER (Thorne et al., 2018).”
Rationalizing Neural Predictions (2016, methodology): “A smaller test set in this corpus identiﬁes, for each aspect, the sentence(s) that relate to this aspect.”
From amateurs to connoisseurs: modeling the evolution of user expertise through online reviews (2013, methodology): “We previously considered BeerAdvocate and RateBeer data in [28], though not in the context of recommendation.”
EDUCE: Explaining model Decisions through Unsupervised Concepts Extraction (2019, methodology): “hile our paradigm is different than Lei et al. (2016); Bastings et al. (2019), we also perform a multi-aspect sentiment regression experiment on the pre-processed subset of the BeerAdvocate3 dataset (McAuley et al., 2012). It consists of 260;000 beer reviews where ratings 2We do not show the amount ”
Interpretable Neural Predictions with Differentiable Binary Variables (2019, methodology): “A pre-processed subset of the BeerAdvocate5 data set is used (McAuley et al., 2012).”
Joint Deep Modeling of Users and Items Using Reviews for Recommendation (2017, background): “The data span a period of more than 10 years, including almost 3 million reviews up to November 2011 [18].”
DynaSent: A Dynamic Benchmark for Sentiment Analysis (2020, methodology): “The Yelp and Amazon datasets are those used in Zhang et al. 2015; the ﬁrst is derived from an earlier version of the Yelp Academic Dataset, and the second is derived from the dataset used by McAuley et al. (2012).”
FLAME: A Probabilistic Model Combining Aspect Based Opinion Mining and Collaborative Filtering (2015, background): “For the process of generating words, we follow the assumption in [20, 13, 25] that the words in one sentence of a review refer to the same aspect.”
Description Based Text Classification with Reinforcement Learning (2020, methodology): “Widely used datasets include (1) the BeerAd-vocate review dataset (McAuley et al., 2012).”
Learning Credible Deep Neural Networks with Rationale Regularization (2019, background): “Product Review Dataset (PR): It is a multi-aspect beer review dataset [37] with data derived from BeerAdvocate 2.”
SemEval-2015 Task 12: Aspect Based Sentiment Analysis (2015, background): “For example, the datasets of McAuley et al. (2012) provide aspects and respective ratings at the review level (i.e., aspects and ratings associated with entire reviews, not particular sentences) 1 about Beers, Pubs, Toys and Games, and Audiobooks.”
Gradient Estimation with Stochastic Softmax Tricks (2020, methodology): “We used the BeerAdvocate dataset [56], which contains reviews comprised of free-text feedback and ratings for multiple aspects (appearance, aroma, palate, and taste; Fig.”
Jointly modeling aspects, ratings and sentiments for movie recommendation (JMARS) (2014, background): “Similarly, [13] uses multi-aspect ratings to infer sentiments for predefined aspects.”
A Context-Aware User-Item Representation Learning for Item Recommendation (2017, methodology): “The other two datasets are collected from the RateBeer website(6) and the Yelp challenge website,(7) named Beer [25] and Yelp.”
SemEval-2016 Task 5: Aspect Based Sentiment Analysis (2016, background): “Such ratings can be used to estimate the mean sentiment per aspect from multiple reviews (McAuley et al., 2012).”
No country for old members: user lifecycle and linguistic change in online communities (2013, methodology): “We crawled a complete set of reviews for BeerAdvocate and RateBeer all the way back to the inception of the site [33], spanning a period of more than 10 years—from 2001 until 2011.”
An Unsupervised Neural Attention Model for Aspect Extraction (2017, background): “(2) BeerAdvocate: This is a beer review corpus introduced in (McAuley et al., 2012), containing over 1.5 million reviews.”
An Information Bottleneck Approach for Controlling Conciseness in Rationale Extraction (2020, methodology): “, 2019) and the BeerAdvocate regression task (McAuley et al., 2012) used in Lei et al.”