An Aspect Performance-aware Hypergraph Neural Network for Review-based Recommendation

TL;DR

This paper comprehensively models the relationships among users, items, aspects, and sentiment polarity by systematically constructing an aspect hypergraph based on user reviews and aggregates aspects representing users and items by employing an aspect performance-aware hypergraph aggregation method.

Abstract

Online reviews allow consumers to provide detailed feedback on various aspects of items. Existing methods utilize these aspects to model users' fine-grained preferences for specific item features through graph neural networks. We argue that the performance of items on different aspects is important for making precise recommendations, which has not been taken into account by existing approaches, due to lack of data. In this paper, we propose an aspect performance-aware hypergraph neural network (APH) for the review-based recommendation, which learns the performance of items from the conflicting sentiment polarity of user reviews. Specifically, APH comprehensively models the relationships among users, items, aspects, and sentiment polarity by systematically constructing an aspect hypergraph based on user reviews. In addition, APH aggregates aspects representing users and items by employing an aspect performance-aware hypergraph aggregation method. It aggregates the sentiment polarities from multiple users by jointly considering user preferences and the semantics of their sentiments, determining the weights of sentiment polarities to infer the performance of items on various aspects. Such performances are then used as weights to aggregate neighboring aspects. Experiments on six real-world datasets demonstrate that APH improves MSE, Precision@5, and Recall@5 by an average of 2.30%, 4.89%, and 1.60% over the best baseline. The source code and data are available at https://github.com/dianziliu/APH.

Figures (7)

• Figure 1: Hypergraph vs. simple graph. There are three reviews written by three users for one headphone. Based on these reviews, we extract five relationships that record the users’ sentiment polarity towards various aspects of the item. Formally, an vertex set is $\mathcal{V}=\{u_1, u_2, u_3, i, a_1, a_2, a_3, Pos, Neg\}$ and a relationship set is $\mathcal{E}=\{e_1, e_2, e_3, e_4, e_5\}$. In the simple graph, two vertices are joined together by an edge if they commonly exist in any relationship. This graph cannot tell us much information, like whether a user has a positive sentiment for what. In the hypergraph, each hyperedge $e_n$ connects four vertices, and can completely illustrate one extracted relationship.
• Figure 2: Framework of APH. It first extracts aspects and user sentiments from reviews to construct a hypergraph. Then, to learn the true relationship between an item and an aspect from conflicting user sentiments, APH considers user preferences to identify the weight of their sentiments. Likewise, we use a similar way to calculate the aspect-based user representations. Finally, APH fuses items, neighbor aspect nodes, and their ID embeddings to make predictions.
• Figure 3: Aspect distribution in the Yelp dataset, which is similar to the other five datasets.
• Figure 4: A case study. We show the extracted aspect quadruples of an item "$B0000538AC$" and an aspect "pack"; then we build the subgraph; we also show the attention scores calculated by Equation \ref{['eq:weight_edge']}, and the final predicted rating.
• Figure 5: Sparsity analysis of learning rate $\gamma$ and the regularization parameter $\lambda$ on six datasets.