Self-Supervised Dynamic Hypergraph Recommendation based on Hyper-Relational Knowledge Graph - Leitura de Artigo

ABSTRACT

... However, the long-tail distribution of entities leads to sparsity in supervision signals, which weakens the quality of item representation when utilizing KG enhancement. Additionally, the binary relation representation of KGs simplifies hyper-relational facts, making it challenging to model complex real-world information. Furthermore, the over-smoothing phenomenon results in indistinguishable representations and information loss. To address these challenges, we propose the SDK (Self-Supervised Dynamic Hypergraph Recommendation based on Hyper-Relational Knowledge Graph) framework. This framework establishes a cross-view hypergraph self-supervised learning mechanism for KG enhancement. Specifically, we model hyper-relational facts in KGs to capture interdependencies between entities under complete semantic conditions. With the refined representation, a hypergraph is dynamically constructed to preserve features in the deep vector space, thereby alleviating the over-smoothing problem. ... 

[Dois problemas que afetam o uso de KGs Sistemas de Recomendação e também LLM: tail entities (poucos dados) e relação binária]

1 INTRODUCTION

Specifically, we transform the KG into a hyper-relational format and model facts within the context of
𝑁 -ary relations. This approach enables us to capture more nuanced relationships among entities. For example, a hyper-relational fact is depicted in Figure 1. 

The form of hyper-relational facts, which consists of a basic triplet (ℎ, 𝑟, 𝑡) and several qualifiers (𝑞𝑟, 𝑞𝑣) ...  Unlike triplet-based facts that model each piece of semantic information independently before aggregation, hyper-relational facts represent intrinsic semantic associations by directly modeling the basic triplet and qualifiers as a whole. By employing the SDK framework, we aim to enhance the representation ability and generalization performance of KG-based recommendation systems by leveraging hyper-relational facts and their qualifiers. 

2 RELATED WORK

Hyper-relational KG: Since the triplets in the traditional KG over-simplify the complexity of the data, recent studies have begun to model hyper-relational facts. m-TransH [32] is a method based on TransH [30] to transform hyper-relational facts through star-to-clique conversion. RAE [46] builds upon m-TransH and further transforms hyper-relational facts into 𝑁 -ary facts with abstract relations. NaLP [6] proposes a link prediction method that models 𝑁 -ary facts as role-value pairs and utilizes a convolution-based framework to compute the similarity of each pair. StarE [3] specifically designed an encoder for 𝑁-ary facts to be compatible with indefinite-length qualifiers and emphasize the interaction of the basic triplets to qualifiers.

[Embeddings em KG Hiper Relacionais]

3 PRELIMINARIES

Knowledge Graph: A KG provides auxiliary information for the recommender system to alleviate the problem of data sparsity. The KG G𝑘 utilizes the triplet set {(ℎ, 𝑟, 𝑡)|ℎ, 𝑡 ∈ 𝐸, 𝑟 ∈ 𝑅} to describe facts, where 𝐸 and 𝑅 are respectively the sets of entities and relations, and (ℎ, 𝑟, 𝑡) indicates there is a relation 𝑟 from head entity ℎ to tail entity 𝑡. In KG-based recommendation where 𝑉 ∈ 𝐸, an item 𝑣 ∈ 𝑉 may form of triplets with several different entities in the given KG G𝑘 .

Hyper-relational Knowledge Graph: HKG is an extension of the standard KG, which describes the 𝑁 -ary facts in the real world by supplementing the basic triple semantics with qualifier pairs. A hyper-relational fact is represented by a tuple (ℎ, 𝑟, 𝑡, Qℎ𝑟𝑡 ), where (ℎ, 𝑟, 𝑡) is the knowledge triplet, and Qℎ𝑟𝑡 is the set of qualifier pairs {(𝑞𝑣𝑖 , 𝑞𝑟𝑖 )}| Qℎ𝑟𝑡 | 𝑖=1 with qualifier relations 𝑞𝑟 ∈ 𝑅 and qualifier entities 𝑞𝑣 ∈ 𝐸. In this way, these entities can be seen as connected by a hyperedge in the HKG, representing an 𝑁 -ary fact, also known as a statement.

[Para apresentar a formalização dos modelos de KG]

4 METHODOLOGY

5 EXPERIMENTS