[Paper Review] Aspect-Aware Latent Factor Model: Rating Prediction with Ratings and Reviews

Zhiyong Cheng, Ying Ding+2 more|arXiv (Cornell University)|Feb 22, 2018

Recommender Systems and Techniques43 references141 citations

TL;DR

The paper proposes ALFM, which combines an aspect-aware topic model (ATM) with an aspect-aware latent factor model to predict ratings using ratings and reviews, improving local and cold-start performance and interpretability.

ABSTRACT

Although latent factor models (e.g., matrix factorization) achieve good accuracy in rating prediction, they suffer from several problems including cold-start, non-transparency, and suboptimal recommendation for local users or items. In this paper, we employ textual review information with ratings to tackle these limitations. Firstly, we apply a proposed aspect-aware topic model (ATM) on the review text to model user preferences and item features from different aspects, and estimate the aspect importance of a user towards an item. The aspect importance is then integrated into a novel aspect-aware latent factor model (ALFM), which learns user's and item's latent factors based on ratings. In particular, ALFM introduces a weighted matrix to associate those latent factors with the same set of aspects discovered by ATM, such that the latent factors could be used to estimate aspect ratings. Finally, the overall rating is computed via a linear combination of the aspect ratings, which are weighted by the corresponding aspect importance. To this end, our model could alleviate the data sparsity problem and gain good interpretability for recommendation. Besides, an aspect rating is weighted by an aspect importance, which is dependent on the targeted user's preferences and targeted item's features. Therefore, it is expected that the proposed method can model a user's preferences on an item more accurately for each user-item pair locally. Comprehensive experimental studies have been conducted on 19 datasets from Amazon and Yelp 2017 Challenge dataset. Results show that our method achieves significant improvement compared with strong baseline methods, especially for users with only few ratings. Moreover, our model could interpret the recommendation results in depth.

Motivation & Objective

• Leverage review text to model user preferences and item features across multiple aspects.
• Learn aspect importance for each user-item pair to improve local rating prediction.
• Associate latent factors with aspects via a weight matrix to estimate aspect ratings.
• Compute overall rating as a weighted sum of aspect ratings with aspect importance.
• Improve handling of data sparsity and enhance interpretability of recommendations.

Proposed method

• Introduce an aspect-aware topic model (ATM) to extract user and item aspect-topic distributions from reviews.
• Develop an aspect-aware latent factor model (ALFM) with a weight matrix W mapping latent factors to aspects, enabling aspect-specific ratings.
• Define aspect rating r_{u,i,a} as r_{u,i,a}=s_{u,i,a} * (w_a ⊙ p_u)^T (w_a ⊙ q_i).
• Compute aspect importance ρ_{u,i,a} from user and item priors via π_u, λ_u,a, λ_i,a.
• Estimate s_{u,i,a} as 1 - JSD(θ_{u,a}, ψ_{i,a}) using ATM outputs.
• Optimize the rating prediction with SGD on the objective combining MF regularization and L1/L2 penalties.

Experimental results

Research questions

• RQ1 Can ALFM outperform state-of-the-art methods that use both ratings and reviews for rating prediction?
• RQ2 Does incorporating aspect-aware representations improve predictions for cold-start or sparsely rated users/items?
• RQ3 Can the model provide interpretable explanations for its recommendations by identifying influential aspects?
• RQ4 How do key hyperparameters (e.g., number of topics K and latent factors f) affect performance?

Key findings

• ALFM significantly outperforms strong baselines on 19 real-world Amazon and Yelp datasets.
• The model shows improved performance for users with few ratings, addressing cold-start effects.
• ASPECT-aware weights enable local, per user-item adaptation improving rating prediction accuracy.
• ATM provides interpretable aspect extraction and, when coupled with ALFM, yields explainable recommendations.