Improve Recommendation Performance Research Articles

A feature-aware long-short interest evolution network for sequential recommendation

Recommendation systems are an effective solution to deal with information overload, particularly in the e-commerce sector, in which sequential recommendation is extensively utilized. Sequential recommendations aim to acquire users’ interests and provide accurate recommendations by analyzing users’ historical interaction sequences. To improve recommendation performance, it is vital to take into account the long- and short-term interests of users. Despite significant advancements in this domain, some issues need to be addressed. Conventional sequential recommendation models typically express each item with a uniform embedding, ignoring evolutionary patterns among item attributes, such as category, brand, and price. Moreover, these models often model users’ long- and short-term interests independently, failing to adequately address the issues of interest drift and short-term interest evolution. This study proposes a new model, the Feature-aware Long-Short Interest Evolution Network (FLSIE), to address the above-mentioned issues. Specifically, the model uses explicit feature embedding to represent item attribute information and employs a two-dimensional (2D) attention mechanism to distinguish the significance of individual features in a specific item and the relevance of each item in the interaction sequence. Furthermore, to avoid the issue of interest drift, the model employs a long-term interest guidance mechanism to enhance the representation of short-term interest and adopts a gated recurrent unit with attentional update gate to model the dynamic evolution of users’ short-term interest. Experimental results indicate that our presented model outperforms existing methods on three real-world datasets.

Mitigating Recommendation Biases via Group-Alignment and Global-Uniformity in Representation Learning

Collaborative Filtering (CF) plays a crucial role in modern recommender systems, leveraging historical user-item interactions to provide personalized suggestions. However, CF-based methods often encounter biases due to imbalances in training data. This phenomenon makes CF-based methods tend to prioritize recommending popular items and performing unsatisfactorily on inactive users. Existing works address this issue by rebalancing training samples, reranking recommendation results, or making the modeling process robust to the bias. Despite their effectiveness, these approaches can compromise accuracy or be sensitive to weighting strategies, making them challenging to train. Therefore, exploring how to mitigate these biases remains in urgent demand. In this paper, we deeply analyze the causes and effects of the biases and propose a framework to alleviate biases in recommendation from the perspective of representation distribution, namely Group- A lignment and Global- U niformity Enhanced R epresentation L earning for Debiasing Recommendation (AURL) . Specifically, we identify two significant problems in the representation distribution of users and items, namely group-discrepancy and global-collapse. These two problems directly lead to biases in the recommendation results. To this end, we propose two simple but effective regularizers in the representation space, respectively named group-alignment and global-uniformity. The goal of group-alignment is to bring the representation distribution of long-tail entities closer to that of popular entities, while global-uniformity aims to preserve the information of entities as much as possible by evenly distributing representations. Our method directly optimizes both the group-alignment and global-uniformity regularization terms to mitigate recommendation biases. Please note that AURL applies to arbitrary CF-based recommendation backbones. Extensive experiments on three real datasets and various recommendation backbones verify the superiority of our proposed framework. The results show that AURL not only outperforms existing debiasing models in mitigating biases but also improves recommendation performance to some extent.

Privacy-preserving Cross-domain Recommendation with Federated Graph Learning

As people inevitably interact with items across multiple domains or various platforms, cross-domain recommendation (CDR) has gained increasing attention. However, the rising privacy concerns limit the practical applications of existing CDR models, since they assume that full or partial data are accessible among different domains. Recent studies on privacy-aware CDR models neglect the heterogeneity from multiple-domain data and fail to achieve consistent improvements in cross-domain recommendation; thus, it remains a challenging task to conduct effective CDR in a privacy-preserving way. In this article, we propose a novel, as far as we know, federated graph learning approach for Privacy-Preserving Cross-Domain Recommendation (PPCDR) to capture users’ preferences based on distributed multi-domain data and improve recommendation performance for all domains without privacy leakage. The main idea of PPCDR is to model both global preference among multiple domains and local preference at a specific domain for a given user, which characterizes the user’s shared and domain-specific tastes toward the items for interaction. Specifically, in the private update process of PPCDR, we design a graph transfer module for each domain to fuse global and local user preferences and update them based on local domain data. In the federated update process, through applying the local differential privacy technique for privacy-preserving, we collaboratively learn global user preferences based on multi-domain data and adapt these global preferences to heterogeneous domain data via personalized aggregation. In this way, PPCDR can effectively approximate the multi-domain training process that directly shares local interaction data in a privacy-preserving way. Extensive experiments on three CDR datasets demonstrate that PPCDR consistently outperforms competitive single- and cross-domain baselines and effectively protects domain privacy.

The Differentiation of Residents’ Cultural Consumption Tendency and Consumption Recommendation System Based on Network Inference Algorithm

Abstract To address the issue of insufficient accuracy in consumer recommendation systems, a new biased network inference algorithm is proposed based on traditional network inference algorithms. This new network inference algorithm can significantly improve the resource allocation ability of the original one, thereby improving recommendation performance. Then, the performance of this algorithm is verified through comparative experiments with network-based inference algorithms, network inference algorithms with initial resource optimization, and heterogeneous network inference algorithms. The results showed that the accuracy of the new network inference algorithm was 24.5%, which was superior to traditional one. In terms of system performance testing, the recommendation hit rate of the new network inference algorithm increased by 13.97%, which was superior to the other three comparative algorithms. The experimental results indicated that a novel network inference algorithm with bias can improve the performance of consumer recommendation systems, providing new ideas for improving the performance of consumer recommendation systems.

A graph attention network with contrastive learning for temporal review-based recommendations

In recent years, review-based recommendations have recently attracted extensive attention from researchers. However, applying reviews to improve recommendation performance still faces some problems that remain unsolved: (1) Most existing works cannot capture the time-varying user preferences and high-order collaborative features. (2) Most of the existing works often use supervised learning to train models with a limited interaction between users and items, which is not sufficient to learn the more accurate recommendation models. To overcome these problems, we propose a new Graph Attention Network with Contrastive Learning for temporary review-based recommendation, named GANCL. Specifically, to capture dynamic user preferences and high-order collaborative features, we design a user–item bipartite graph with time-series review information and ratings as its edges, and then use the graph attention and different gating mechanisms to extract the corresponding features. To make full use of the limited interaction between the users and items, we use the contrastive learning paradigm for the nodes and edges in the bipartite graph to more effectively model the user–item interaction. A large number of experiments on five public data from Amazon prove that the performance of the GANCL is improved by 2.76% and 2.83% respectively compared with the state-of-the-art algorithms in MSE and MAE.

Asymmetric multilevel interactive attention network integrating reviews for item recommendation

Recently, most studies in the field have focused on integrating reviews behind ratings to improve recommendation performance. However, two main problems remain (1) Most works use a unified data form and the same processing method to address the user and the item reviews, regardless of their essential differences. (2) Most works only adopt simple concatenation operation when constructing user-item interaction, thus ignoring the multilevel relationship between the user and the item, which may lead to suboptimal recommendation performance. In this paper, we propose a novel Asymmetric Multi-Level Interactive Attention Network (AMLIAN) integrating reviews for item recommendation. AMLIAN can predict precise ratings to help the user make better and faster decisions. Specifically, to address the essential difference between the user and the item reviews, AMLIAN uses the asymmetric network to construct user and item features using different data forms (document-level and review-level). To learn more personalized user-item interaction, the user ID and item ID and some processed features of user reviews and item reviews are respectively used for multilevel relationships. Experiments on five real-world datasets show that AMLIAN significantly outperforms state-of-the-art methods.

Robust social recommendation based on contrastive learning and dual-stage graph neural network

GNN-based social recommendation aims to use social network information to improve recommendation performance of traditional user–item interaction network (U–I network). However, in graph neural network information aggregation, both social networks and U–I networks inevitably have noise, which affects accuracy of recommendation results. To reduce the noise impact of network data, we propose Robust Social Recommendation based on Contrastive Learning and Dual-Stage Graph Neural Network (CLDS). First, considering instability of social networks, we propose the social preference network. It is robust and retains only social friend relationships with common preferences. Based on it and U–I network, we construct a social recommendation pre-training model. Next, we propose self-contrastive learning method. The method initializes multiple social network node representations through Gaussian distribution, pre-training and random disturbance, respectively. Then, it uses contrastive learning on the generated multiple node representations to enhance the robustness of node representation. Finally, CLDS avoids directly capturing potentially user–user and item–item information in U–I networks which is incomplete and untrusted. And instead, it only extracts user–item information to reduce the noise generated by GNN-based U–I network information aggregation. We conduct experiments under the open-source real network dataset. The experimental results show that CLDS outperforms state-of-art methods in social recommendation. The code is available at: https://github.com/Andrewsama/CLDS-master.

Matrix factorization recommender based on adaptive Gaussian differential privacy for implicit feedback

Matrix factorization (MF) is an effective technique in recommendation systems. Since MF needs to utilize and analyze large amounts of user data during the recommendation process, this may lead to the leakage of personal data. Most of the current privacy-preserving MF research aims to protect explicit feedback, but ignores the protection of implicit feedback. In response to this limitation, we propose an adaptive differentially private MF (ADPMF) for implicit feedback. The proposed model is trained under the framework of Bayesian personalized ranking and uses gradient perturbation to achieve the (ϵ,δ)-differential privacy. In our model, we design two effective methods, adaptive clipping and adaptive noise scale, to improve recommendation performance while maintaining privacy. We use Gaussian Differential Privacy (GDP) to accommodate privacy analysis for dynamically changing clipping thresholds and noise scale. Theoretical analysis and experimental results demonstrate that ADPMF not only achieves highly accurate recommendations but also provides differential privacy protection for implicit feedback. The results show that ADPMF can improve the recommended performance substantially by 10% to 20% compared to the current privacy-preserving recommendation methods and has promising application prospects in various fields.

A multi-task learning model for recommendation based on fusion of dynamic and static neighbors

To improve recommendation performance, this study introduces self-supervised learning into recommendation systems and proposes a multi-task learning recommendation framework that combines static neighbor and dynamic neighbor contrastive learning. Specifically, this study considers node relationships at both the graph and embedding levels, which can be defined in two aspects: (1) Static neighbors, which are positive nodes obtained by integrating information from user-item interaction graphs and social graphs. (2) Dynamic neighbors refer to nodes at the encoding level that are similar, but not necessarily the same, at each iteration; hence, they are termed dynamic neighbors. By employing static neighbor contrastive learning and dynamic neighbor contrastive learning as auxiliary tasks for the main recommendation task, we optimized the user and item embeddings. Through extensive experiments using multiple real datasets, we validate the effectiveness of the proposed approach and its components.

Learning Neighbor User Intention on User–Item Interaction Graphs for Better Sequential Recommendation

The task of sequential recommendation aims to predict a user’s preference by analyzing the user’s historical behaviours. Existing methods model item transitions through leveraging sequential patterns. However, they mainly consider the target user’s behaviours and dynamic characteristics, while often ignoring high-order collaborative connections when modelling user preferences. Some recent works try to use graph-based methods to introduce high-order collaborative signals for sequential recommendation. However, these methods are flawed by two problems: the sequential patterns cannot be effectively mined and their way of introducing high-order collaborative signals is not suitable for sequential recommendation. To address these problems, we propose to fully exploit sequence features and model high-order collaborative signals for sequential recommendation. We propose a N eighbor user I ntention-based S equential Rec ommender (NISRec), which utilizes the intentions of high-order connected neighbor users as high-order collaborative signals in order to improve recommendation performance for the target user. The NISRec contains two main modules: the neighbor user intention embedding module (NIE) and the fusion module. The NIE module describes both the long-term and short-term intentions of neighbor users and aggregates them separately. The fusion module uses these two types of aggregated intentions to model high-order collaborative signals in both the embedding process and user preference modelling phase for recommendations of the target user. Experimental results show that our new approach outperforms the state-of-the-art methods on both sparse and dense datasets. Extensive studies further show the effectiveness of the diverse neighbor intentions introduced by the NISRec.

A Survey on Federated Recommendation Systems.

Federated learning has recently been applied to recommendation systems to protect user privacy. In federated learning settings, recommendation systems can train recommendation models by collecting the intermediate parameters instead of the real user data, which greatly enhances user privacy. In addition, federated recommendation systems (FedRSs) can cooperate with other data platforms to improve recommendation performance while meeting the regulation and privacy constraints. However, FedRSs face many new challenges such as privacy, security, heterogeneity, and communication costs. While significant research has been conducted in these areas, gaps in the surveying literature still exist. In this article, we: 1) summarize some common privacy mechanisms used in FedRSs and discuss the advantages and limitations of each mechanism; 2) review several novel attacks and defenses against security; 3) summarize some approaches to address heterogeneity and communication costs problems; 4) introduce some realistic applications and public benchmark datasets for FedRSs; and 5) present some prospective research directions in the future. This article can guide researchers and practitioners understand the research progress in these areas.

Pre-train, Prompt, and Recommendation: A Comprehensive Survey of Language Modeling Paradigm Adaptations in Recommender Systems

Abstract The emergence of Pre-trained Language Models (PLMs) has achieved tremendous success in the field of Natural Language Processing (NLP) by learning universal representations on large corpora in a self-supervised manner. The pre-trained models and the learned representations can be beneficial to a series of downstream NLP tasks. This training paradigm has recently been adapted to the recommendation domain and is considered a promising approach by both academia and industry. In this paper, we systematically investigate how to extract and transfer knowledge from pre-trained models learned by different PLM-related training paradigms to improve recommendation performance from various perspectives, such as generality, sparsity, efficiency and effectiveness. Specifically, we propose a comprehensive taxonomy to divide existing PLM-based recommender systems w.r.t. their training strategies and objectives. Then, we analyze and summarize the connection between PLM-based training paradigms and different input data types for recommender systems. Finally, we elaborate on open issues and future research directions in this vibrant field.

Attention-based multi attribute matrix factorization for enhanced recommendation performance

In E-commerce platforms, auxiliary information containing several attributes (e.g., price, quality, and brand) can improve recommendation performance. However, previous studies used a simple combined embedding approach that did not consider the importance of each attribute embedded in the auxiliary information or only used some attributes of the auxiliary information. However, user purchasing behavior can vary significantly depending on the attributes. Thus, we propose multi attribute-based matrix factorization (MAMF), which considers the importance of each attribute embedded in various auxiliary information. MAMF obtains more representative and specific attention features of the user and item using a self-attention mechanism. By acquiring attentive representation, MAMF learns a high-level interaction precisely between users and items. To evaluate the performance of the proposed MAMF, we conducted extensive experiments using three real-world datasets from amazon.com. The experimental results show that MAMF exhibits excellent recommendation performance compared with various baseline models.

Knowledge-Aware Graph Self-Supervised Learning for Recommendation

Collaborative filtering (CF) based on graph neural networks (GNN) can capture higher-order relationships between nodes, which in turn improves recommendation performance. Although effective, GNN-based methods still face the challenges of sparsity and noise in real scenarios. In recent years, researchers have introduced graph self-supervised learning (SSL) techniques into CF to alleviate the sparse supervision problem. The technique first augments the data to obtain contrastive views and then utilizes the mutual information maximization to provide self-supervised signals for the contrastive views. However, the existing approaches based on graph self-supervised signals still face the following challenges: (i) Most of the works fail to effectively mine and exploit the supervised information from the item knowledge graph, resulting in suboptimal performance. (ii) Existing data augmentation methods are unable to fully exploit the potential of contrastive learning, because they primarily focus on the contrastive view of data structure changes and neglect the adjacent relationship among users and items. To address these issues, we propose a novel self-supervised learning approach, namely Knowledge-aware Graph Self-supervised Learning (KGSL). Specifically, we calculate node similarity based on semantic relations between items in the knowledge graph to generate a semantic-based item similarity graph. Then, the self-supervised learning contrast views are generated from both the user–item interaction graph and the item similarity graph, respectively. Maximization of the information from these contrastive views provides additional self-supervised signals to enhance the node representation capacity. Finally, we establish a joint training strategy for the self-supervised learning task and the recommendation task to further optimize the learning process of KGSL. Extensive comparative experiments as well as ablation experiments are conducted on three real-world datasets to verify the effectiveness of KGSL.

Distinguishing latent interaction types from implicit feedbacks for recommendation

For privacy considerations, many recommendation algorithms are only based on a kind of implicit feedbacks, where various user behaviors, like view, purchase and etc., are simplified as binary interactions. As side information is assumed not available, these algorithms mainly focus on how to encode users and items via a bipartite graph (viz. the user-item interaction matrix), while ignoring to encode edges for distinguishing the interaction types. In this paper, we argue that implicit feedbacks can be classified into a few user-item relations (viz., latent interaction types) via encoding the edges of a bipartite graph. In particular, we design an edge distinguishment module (EDM) into our neural recommendation model, called Relation-Aware Neural Model (RANM). Based on the latent interaction types, we divide the bipartite graph into a few subgraphs, each consisting of only edges of the same relation and their connected user and item nodes. We propose a Relation-Aware Graph Neural Network (RAGNN) for learning user and item representations. For encoding items, we apply the RAGNN on the relation-aware bipartite graph; While for encoding a user, we first encode several latent interests each on one subgraph and then fuse these interest encodings as the user representation. Experiments on three public datasets validate that our approach of edge classification and representation learning help improving recommendation performance compared with the state-of-the-art competitors. The implementations are available at https://github.com/lulu0913/RAGNN.

Deep recommendation system based on knowledge graph and review text

With the explosive increase of information, recommendation system is applied in a variety of areas. However, the performance of recommendation system is limited due to issues such as data sparsity, cold starts and poor semantic understanding. In order to make full use of external information to assist recommendation, deeply mine the semantic information of review text and further improve the performance of recommendation system, a deep recommendation system based on knowledge graph and review text (Drs-kgrt) is proposed in this paper. In Drs-kgrt, knowledge graph, review text and the social records between users are used as auxiliary information to improve recommendation performance. Firstly, the review text is divided into user review text and item review text. BERT (Bidirectional Encoder Representation from Transformers) is used to accurately understand semantic information in user review text and the social records between users. The trust relationship between users and user preferences are fully mined to form user feature vectors. Secondly, BERT and knowledge graph entity recognition link technology are combined to extract item attribute feature entities and their associated entities. The fine-grained features of the items are analyzed to form item feature vectors. Thirdly, based on the scoring matrix, latent vectors of users and items are obtained by matrix decomposition. The deep features of users and items are generated based on user feature vectors, item feature vectors, latent vectors of users and items, the deep recommendation system is established to predict user scores for items. Finally, experiments are conducted on the Douban dataset and Amazon Movie Review dataset, the results show that the proposed algorithm can achieve better performance compared with other benchmark recommendation algorithms.

IUG-CF: Neural collaborative filtering with ideal user group labels

Demographics are crucial information for recommender systems (RSs). Most existing demographic-based RSs focus on similarity between user profiles. However, they rarely incorporate demographic data to describe an item and establish the connection between items and users. In this paper, we propose the concept of the ideal user group (IUG) as a dynamic label for items. This label indicates the users who are most suitable for an item, based on the demographics of its historical customers. Unlike a general label (such as genre or language), the IUG is dynamically changing with the distribution of historical user demographics and is built based on demographic information that undergoes a split-combine process. To validate our method’s effectiveness, we propose an IUG-based neural collaborative filtering (IUG-CF) model. Experimental results on three real-world datasets show that the IUG is an effective approach for improving recommendation performance.

Attention-guided graph convolutional network for multi-behavior recommendation

Multi-behavior recommendation system aims to improve recommendation performance by using the interaction data of users′ multiple behaviors. Although some methods have explored the dependencies between different behaviors, there are also existing challenges: (1) user–item interactions have complex dependencies; (2) the dependencies between multiple behaviors vary due to users′ personalized preferences. To address these challenges, we propose a new model MB-AGCN (Attention-Guided Graph Convolutional Network for Multi-Behavior Recommendation), which considers personalized interaction patterns and cross-typed behavioral interdependencies. In the MB-AGCN framework, we take the different effects of multi-behavior information on predicting user preferences into account. We first model the user multi-behavior relationships with the attention mechanism to capture the personalized multi-behavior characteristics. Then, we explore the knowledge learned from the multi-behavior relationship modeling to generate a weight matrix that guides the graph neural network to learn the complex dependencies in different types of user–item interactions and capture the relationships between different types of behaviors. A comprehensive evaluation on three real-world datasets shows that MB-AGCN consistently outperforms state-of-the-art methods. Our codes will be available at https://github.com/3endurance/MB-AGCN.

A Social Recommendation Model Based on Basic Spatial Mapping and Bilateral Generative Adversarial Networks.

Social recommender systems are expected to improve recommendation quality by incorporating social information when there is little user-item interaction data. Therefore, how to effectively fuse interaction information and social information becomes a hot research topic in social recommendation, and how to mine and exploit the heterogeneous information in the interaction and social space becomes the key to improving recommendation performance. In this paper, we propose a social recommendation model based on basic spatial mapping and bilateral generative adversarial networks (MBSGAN). First, we propose to map the base space to the interaction and social space, respectively, in order to overcome the issue of heterogeneous information fusion in two spaces. Then, we construct bilateral generative adversarial networks in both interaction space and social space. Specifically, two generators are used to select candidate samples that are most similar to user feature vectors, and two discriminators are adopted to distinguish candidate samples from high-quality positive and negative examples obtained from popularity sampling, so as to learn complex information in the two spaces. Finally, the effectiveness of the proposed MBSGAN model is verified by comparing it with both eight social recommendation models and six models based on generative adversarial networks on four public datasets, Douban, FilmTrust, Ciao, and Epinions.

Enhancing Multi-Behavior Recommendations Through Capturing Dynamic Preferences

Multi-behavior recommendation has gained significant attention in recent years for its ability to outperform single-behavior models. Current research related to multi-behavior models leaves room for improvement in the following two areas. First, the noise carried by individual behaviors and the additional noise generated during behavior processing is often overlooked, and these can ultimately degrade recommendation performance. Second, the specific time period of behavioral interactions and the frequency of interactions within that time period are also not taken into account. To address the above limitations, we propose a multi-behavior recommendation model integrating dynamic preferences (MB-DP) that captures dynamic interests while smoothing and denoising multi-behavior information. MB-DP extracts low and high-order semantics from various behaviors and unifies the measurements to generate interaction predictions. Additionally, it analyzes the interaction time and frequency of each behavior using gated recurrent units to capture the dynamic preferences of users and improve the prediction values. Extensive experimental results on two real-world datasets show that MB-DP significantly improves recommendation performance compared to the state-of-the-art baselines.