Neural News Recommendation with Multi-Head Self-Attention

Wu, Chuhan; Wu, Fangzhao; Ge, Suyu; Qi, Tao; Huang, Yongfeng; Xie, Xing

doi:10.18653/v1/d19-1671

Cited by 258 publications

(325 citation statements)

References 19 publications

Supporting

Mentioning

325

Contrasting

Order By: Relevance

“…All models mentioned above represent users and items following the same principle, and the representations are derived from the same data source. Contrary to this, NPA [43] and NeuHash-CF [12] represent users and items in different ways. While they model the items using content-based information, the users are represented by one-hot coded userID.…”

Section: Review-based User/item Modelingmentioning

confidence: 99%

Leave No User Behind: Towards Improving the Utility of Recommender Systems for Non-mainstream Users

Urbano

Hanjalic

2021

Proceedings of the 14th ACM International Conference on Web Search and Data Mining

View full text Add to dashboard Cite

In a collaborative-filtering recommendation scenario, biases in the data will likely propagate in the learned recommendations. In this paper we focus on the so-called mainstream bias: the tendency of a recommender system to provide better recommendations to users who have a mainstream taste, as opposed to non-mainstream users. We propose NAECF, a conceptually simple but effective idea to address this bias. The idea consists of adding an autoencoder (AE) layer when learning user and item representations with text-based Convolutional Neural Networks. The AEs, one for the users and one for the items, serve as adversaries to the process of minimizing the rating prediction error when learning how to recommend. They enforce that the specific unique properties of all users and items are sufficiently well incorporated and preserved in the learned representations. These representations, extracted as the bottlenecks of the corresponding AEs, are expected to be less biased towards mainstream users, and to provide more balanced recommendation utility across all users. Our experimental results confirm these expectations, significantly improving the recommendations for nonmainstream users while maintaining the recommendation quality for mainstream users. Our results emphasize the importance of deploying extensive content-based features, such as online reviews, in order to better represent users and items to maximize the debiasing effect.

show abstract

Section: Review-based User/item Modelingmentioning

confidence: 99%

Leave No User Behind: Towards Improving the Utility of Recommender Systems for Non-mainstream Users

Urbano

Hanjalic

2021

Proceedings of the 14th ACM International Conference on Web Search and Data Mining

View full text Add to dashboard Cite

show abstract

“…Similarly, in the other studies a whole document or article is embedded and then its similarity with others is found [4], [53]- [57]. Moreover, It is used as base line model of feature extraction before feeding data to neural network [13], [58]- [61]. Furthermore a knowledge graph of entities are also converted in to embeddings to further use it in deep-learning settings.…”

Section: F Embeddingmentioning

confidence: 99%

“…An LDA is applied to text to get several topics, that are made input to the CNN model, which outputs the corresponding latent factor model L1. In another study [58] a convolution neural network-based news recommender system is proposed which mainly constitutes three modules, i.e., news encoder, user encoder, and click predictor. The purpose of the news encoder is to learn news representation.…”

Section: ) Cnnmentioning

confidence: 99%

A Review of Text-Based Recommendation Systems

et al. 2021

View full text Add to dashboard Cite

Many websites over the Internet are producing a variety of textual data; such as news, research articles, ebooks, personal blogs, and user reviews. In these websites, the textual data is so large that the process of finding pertinent information by a user often becomes cumbersome. To overcome this issue, "Text-based Recommendation Systems (RS)" are being developed. They are the systems with the capability to find the relevant information in a minimal time using text as the primary feature. There exist several techniques to build and evaluate such systems. And though a good number of surveys compile the general attributes of recommendation systems, there is still a lack of comprehensive literature review about the text-based recommendation systems. In this paper, we present a review of the latest studies on text-based RS. We have conducted this survey by collecting literature from preeminent digital repositories, that was published during the period 2010-2020. This survey mainly covers the four major aspects of the textual based recommendation systems used in the reviewed literature. The aspects are datasets, feature extraction techniques, computational approaches, and evaluation metrics. As benchmark datasets carry a vital role in any research, publicly available datasets are extensively reviewed in this paper. Moreover, for text-based RS many proprietary datasets are also used, which are not available in the public. But we have consolidated all the attributes of these publically available and proprietary datasets to familiarize these attributes to new researchers. Furthermore, the feature extraction methods from the text are briefed and their usage in the construction of text-based RS are discussed. Later, various computational approaches that use these features are also discussed. To evaluate these systems, some evaluation metrics are adopted. We have presented an overview of these evaluation metrics and diagramed them according to their popularity. The survey concludes that Word Embedding is the widely used feature selection technique in the latest research. The survey also deduces that hybridization of text features with other features enhance the recommendation accuracy. The study highlights the fact that most of the work is on English textual data, and News recommendation is the most popular domain.

show abstract

“…Zhou et al [40] utilized a self-attention based sequential framework to project user representation into multiple latent spaces and models user behavior for personalized recommendation. Wu et al [41] facilitate news recommendation by modeling the contextual interaction between words and news with multi-head self-attention. Cong et al [42] distinguish the importance of reviews at both word level and sentence level using a hierarchical attention-based network for e-commerce recommendation.…”

Section: Related Workmentioning

confidence: 99%

Hierarchical Attentional Factorization Machines for Expert Recommendation in Community Question Answering

Tang

et al. 2020

IEEE Access

View full text Add to dashboard Cite

The most challenging task of Community Question Answering (CQA) is to provide high-quality answers to users' questions. Currently, a variety of expert recommendation methods have been proposed and greatly improved the effective matching between questions and potential good answerers. However, the performance of existing methods can be adversely affected by many common factors such as data sparsity and noise problem, which cause less precise user modeling. Moreover, existing methods often model user-question interactions through simple ways, failing to capture the multiple scale interactions of question and answerers, which make it difficult to find answerers who are able to provide the best answers. In this paper, we propose an attention-based variant of Factorization Machines (FM) called Hierarchical Attentional Factorization Machines (HaFMRank) for answerer recommendation in CQA, which not only models the interactions between pairs of individual features but emphasizes the roles of crucial features and pairwise interactions. Specifically, we introduce the within-field attention layer to capture the inner structure of features belonging to the same field, while a feature-interaction attention layer is adopted to examine the importance of each pairwise interaction. A pre-training procedure is designed to generate latent FM feature embedding that encode question context and user history into the training process of HaFMRank. The performance of the proposed HaFMRank is evaluated by using real-world datasets of Stack Exchange and experimental results demonstrate that it outperforms several state-of-the-art methods in best answerer recommendation.

show abstract

Neural News Recommendation with Multi-Head Self-Attention

Cited by 258 publications

References 19 publications

Leave No User Behind: Towards Improving the Utility of Recommender Systems for Non-mainstream Users

Leave No User Behind: Towards Improving the Utility of Recommender Systems for Non-mainstream Users

A Review of Text-Based Recommendation Systems

Hierarchical Attentional Factorization Machines for Expert Recommendation in Community Question Answering

Contact Info

Product

Resources

About