Hidden factors and hidden topics: understanding rating dimensions with review text

By Julian McAuley, Jure Leskovec. 2013

1,985 citations recorded by the source corpus.

In order to recommend products to users we must ultimately predict how a user will respond to a new product. To do so we must uncover the implicit tastes of each user as well as the properties of each product. For example, in order to predict whether a user will enjoy Harry Potter, it helps to identify that the book is about wizards, as well as the user's level of interest in wizardry. User feedback is required to discover these latent product and user dimensions. Such feedback often comes in the form of a numeric rating accompanied by review text. However, traditional methods often discard review text, which makes user and product latent dimensions difficult to interpret, since they ignore the very text that justifies a user's rating. In this paper, we aim to combine latent rating dimensions (such as those of latent-factor recommender systems) with latent review topics (such as those learned by topic models like LDA). Our approach has several advantages. Firstly, we obtain highly interpretable textual labels for latent rating dimensions, which helps us to `justify' ratings with text. Secondly, our approach more accurately predicts product ratings by harnessing the information present in review text; this is especially true for new products and users, who may have too few ratings to model their latent factors, yet may still provide substantial information from the text of even a single review. Thirdly, our discovered topics can be used to facilitate other tasks such as automated genre discovery, and to identify useful and representative reviews.

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

Semi-supervised Sequence Learning (2015, methodology): “To better the performance, we add unlabeled data from the IMD B dataset in the previous experiment and Amazon movie reviews [22] to the autoencoder training st a e.”
Revisiting LSTM Networks for Semi-Supervised Text Classification via Mixed Objective Function (2019, neutral): “ACL-IMDB [15] and Elec [8] datasets are for sentiment classification of movie reviews and Amazon product reviews [16] respectively while AG-News [27] and DBpedia [27] are for topic classification of news articles and Wikipedia articles respectively.”
Adversarial Training Methods for Semi-Supervised Text Classification (2016, methodology): “For the Rotten Tomatoes dataset, we also collected unlabeled examples using movie reviews from the Amazon Reviews dataset [19] 4.”
Character-level Convolutional Networks for Text Classification (2015, methodology): “We obtained an Amazon review dataset from the Stanford Network Analysis Project (SNAP), which spans 18 years with 34,686,770 reviews from 6,643,669 users on 2,441,053 products [22].”
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).”
Learning Distributed Representations from Reviews for Collaborative Filtering (2015, methodology): “as Topics” (HFT) recently proposed in [13], and discuss it with respect to the proposed approaches.”
Text Understanding from Scratch (2015, methodology): “We obtained an Amazon review dataset from the Stanford Network Analysis Project (SNAP), which spans 18 years with 34,686,770 reviews from 6,643,669 users on 2,441,053 products(McAuley & Leskovec, 2013).”
Explainable Recommendation: A Survey and New Perspectives (2018, background): “In these approaches, users can usually be provided with intuitive explanations in the form of topical word clouds (McAuley and Leskovec [2013], Wu and Ester [2015], Zhao et al. [2015]). In this section, we review the related work that can be categories into this approach. McAuley and Leskovec [2013]”
Averaging Gone Wrong: Using Time-Aware Analyses to Better Understand Behavior (2016, neutral): “ACM 978-1-4503-4143-1/16/04. http://dx.doi.org/10.1145/2872427.2883083. aim at explaining these temporal aspects of evolution.”
Ask the GRU: Multi-task Learning for Deep Text Recommendations (2016, methodology): “Existing hybrid methods have used unsupervised learning objectives on text content to regularize the parameters of the recommendation model [4, 23, 24].”
Effective Use of Word Order for Text Categorization with Convolutional Neural Networks (2014, background): “It is part of a large Amazon review dataset (McAuley and Leskovec, 2013).”
MLlib: Machine Learning in Apache Spark (2015, methodology): “For this benchmark, we worked with scaled copies of the Amazon Reviews dataset (McAuley and Leskovec, 2013), where we duplicated user information as necessary to increase the size of the data.”
Understanding Convolutional Neural Networks for Text Classification (2018, methodology): “b) Elec: electronic product reviews for sentiment classification introduced by Johnson and Zhang (2015), assembled from the Amazon review dataset (McAuley and Leskovec, 2013; McAuley et al., 2015), containing 200k train and 25k test evenly split reviews.”
Explaining Explanations in AI (2018, neutral): “According to Lipton (2016), approaches to post-hoc interpretability include verbal (natural language) explanations (e.g. McAuley and Leskovec (2013)), visualisations and interactive interfaces (e.g. Simonyan et al. (2013); Tam-agnini et al. (2017)), local explanations or approximations (e.g. Fong…”
Noisy Channel Language Model Prompting for Few-Shot Text Classification (2021, methodology): “We use the batch size of 32 and the sequence length of 128 for datasets with short input text (SST-2, SST-5, MR, TREC) and the batch size of 16 and the sequence length of 256 for datasets with long input text (AGNews, Amazon, Yelp, DBPedia, Ya-hoo, Subj).”