KG E Recommendation Systems

Enhancing recommendations with contrastive learning from collaborative knowledge graph

https://doi.org/10.1016/j.neucom.2022.12.032

Abstract

There have been excellent results using knowledge graphs in recommender systems. Knowledge graphs can be used as auxiliary information to alleviate data sparsity and strengthen the modeling of item sets and the representation of user preferences. However, users as the Core subject in the recommendation process, should be taken seriously. We believe that the user's choice of items will be affected by internal and external factors. Internal factors refer to the users’ fuzzy interest sets, which initially affect the users' choices. External factors refer to the influence of similar users and similar items in the users' selection of items.

[Os itens recomendados ainda passam por um processo de decisão do usuário]

Introduction

There is a lot of knowledge on today's Internet. Generally speaking, this knowledge is not isolated but interrelated; the same is true in recommender systems. The traditional collaborative filtering algorithms [14], [29], [50] are the cornerstone of recommendation system research. However, collaborative filtering algorithms usually face data sparsity and cold start problems. Knowledge graphs (KGs) provide rich item-side information [30], [34], [35], [1], which alleviates the problem of data sparsity in the recommendation process and provides a new idea for the interpretability and accuracy research of recommender systems.

[KGs sobre os itens podem enriquecer o conhecimento dos sistemas de recomendação do lado dos itens mas não do lado dos usuários. Os interesses dos usuários podem mudar com o tempo e com a influência destes fatores externos.]

Although the current research has achieved good results, we find that there are still some problems: (1) The user is an important role in the recommendation process and the quality of user modeling will greatly affects the performance of the recommender systems. Although KGs can be viewed as graph structure of attribute knowledge about items, it is more or less important to users.

[Usuários podem ser similares em relação a gostos/interesses mas isto tem influência temporal e espacial / cultural também]

[Contexto do item para a noção de similaridade entre itens ... Rio de Janeiro e Buenos Aires são mais similares que Nova Iorque e Londres? Em qual contexto? Na Localização, distância (NY e Londres tem um oceano no meio)? Na Cultura/Idioma, os ingleses colonizam os EUA enquanto que a colonização da Argentina e Brasil foi de Espanha e Portugal na mesma época. O conceito de TWIN CITIES. ]

Categorization of knowledge graph based recommendation methods and benchmark datasets from the perspectives of application scenarios: A comprehensive survey

https://doi.org/10.1016/j.eswa.2022.117737

Abstract

Recently, exploitation of Knowledge Graph (KG)-based data as Side Information in recommendation methods has revealed as a sign of resolution to the corresponding challenges; and thus, acquired incredible focus, applicability, and popularity. The incorporation of KG in recommendation has not only effectively alleviated the contrasting challenges, but also has provided specific, accurate, personalized and explainable recommendations about the target items to the end users. In this paper, we explore well-known RSs, popular knowledge repositories, benchmark datasets, recommendation methods, and future research dimensions about the current research.

[KG dos itens permite explicar recomendações]

Introduction

With the current expansion of big data and paced augmentation of internet technology, the storage capacity has exponentially increased with respect to the volume of online dispersed data by providing enormous aid to the information overload. The “information overload”, as a problem, is defined by the Business-Dictionary5 as “The stress persuaded by reception of excessive amount of information to make a decision and dealing with this information without knowing about the validity of its timespan”.

[Todas as respostas possíveis causa Sobrecarga de Informação?]

In 1990 s, online assistants – application programs used to filter out irrelevant data – are developed to recommend selectable choices to the end users based on their previous interactions’ record. Later on, these application programs are designated as Recommender Systems (RSs).

[Nem sempre atividade passada permite antecipar necessidade futura, interesesses / gostos mudam]

Scenario-aware recommendation is currently an emerging research-trend in KG-based RSs. There are innumerable application frameworks that exploit KG-based RSs to automatically provide suggestive responses to the end users. For instance, acquiring recommendations about Books or Research-Articles (Yang et al., 2020, Song et al., 2019a, Wang et al., 2019b, Wang et al., 2019f), e-commerce (Ding et al., 2021, Fu et al., 2020, Ma et al., 2019, Zhang and Chen, 2018), News (Liu et al., 2019, Wang et al., 2019b, Wang et al., 2019d, Zhang et al., 2018a), Question-answering interaction (Zhang et al., 2020a, Park et al., 2021, Zheng and Zhang, 2019, Qin et al., 2019), Entertainment7 (Palumbo et al., 2020, Song et al., 2019a, Hu et al., 2018), MoPI8 (Burke, 2022, Wang et al., 2020b, Mauro et al., 2020, Dadoun et al., 2019, Sha et al., 2019, Sun et al., 2018b) and Social Connections (Song et al., 2019b, Wang et al., 2018a), etc., are clearly expressive application scenarios.

[Bastante pesquisa ....]

The objective of this paper is to categorize current and previous research work on KG-based recommendation with respect to the implementation techniques, methods, datasets, frameworks and applied approaches to streamline the available knowledge and clear the future research dimensions. In this work, we analyze and compare literature of about ten years (i.e., 2011–2020) related to KG-embedding-based methods, KG-path-based methods, and their hybrid frameworks (i.e., unification of KG-embedding and path-based methods) exploited to provide recommendations.

[Os métodos de embeddings consideram contexto? E os de caminho?]

Y. Ge, J. Ma, L. Zhang et al., Trustworthiness-aware knowledge graph representation for recommendation, Knowledge-Based Systems (2023), doi: https://doi.org/10.1016/j.knosys.2023.110865.

[gerar um indicador de trustworthiness-aware das informações de um KG para uso em Sistemas de Recomendação (RS). ]

[Achei interessante por abordar o aspecto da confiança nas informações do KG para tomar a decisão de recomendar ou não um item a um usuário e pq não é sobre busca. KGs em abordagens de filtragem colaborativa de RS poderia atenuar o problema de partida a frio e esparsidade. Mas o cálculo da confiabilidade de cada tripla usa informações da estrutura do grafo como motif (padrão de sub-grafo, comunidades e caminhos) e não metadados de contexto. ]

".... we study the effects of the trustworthiness of the KG on RS and propose a novel method trustworthiness-aware knowledge graph representation (KGR) for recommendation (TrustRec). TrustRec introduces a trustworthiness estimator into noise-tolerant KGR methods for collaborative filtering. Specifically, to assign trustworthiness, we leverage internal structures of KGs from microscopic to macroscopic levels: motifs, communities and global information, to reflect the true degree of triple expression."