Recommend for a Reason: Unlocking the Power of Unsupervised Aspect-Sentiment Co-Extraction

TL;DR

This paper introduces a two-stage, unsupervised approach for extracting aspect-sentiment pairs from reviews and using them to improve rating predictions, enhancing interpretability and accuracy in recommendation systems.

Contribution

It presents a novel unsupervised method combining aspect-sentiment extraction with an attention-aware rating estimator, improving interpretability and performance over existing models.

Findings

ASPE effectively extracts aspect-sentiment pairs.

APRE achieves superior rating prediction accuracy.

Method outperforms leading baselines on Amazon datasets.

Abstract

Compliments and concerns in reviews are valuable for understanding users' shopping interests and their opinions with respect to specific aspects of certain items. Existing review-based recommenders favor large and complex language encoders that can only learn latent and uninterpretable text representations. They lack explicit user attention and item property modeling, which however could provide valuable information beyond the ability to recommend items. Therefore, we propose a tightly coupled two-stage approach, including an Aspect-Sentiment Pair Extractor (ASPE) and an Attention-Property-aware Rating Estimator (APRE). Unsupervised ASPE mines Aspect-Sentiment pairs (AS-pairs) and APRE predicts ratings using AS-pairs as concrete aspect-level evidence. Extensive experiments on seven real-world Amazon Review Datasets demonstrate that ASPE can effectively extract AS-pairs which enable APRE to deliver superior accuracy over the leading baselines.