Sequential Recommender Systems Research Articles

Ranking on user–item heterogeneous graph for Ecommerce next basket recommendations

Recommender systems have become an indispensable engine for online applications, which can help users locate the items they need among numerous other candidate items. Despite the success of general models, sequential recommender systems based on sequential patterns have garnered attention, especially in e-commerce, because customers do not necessarily purchase the same items over time. Markov chains and neural networks are the most adopted models in previous studies, but few integrated user sequential order information and user numeric rating feedback. To simultaneously capture the time-aware demands and time-invariant interests of e-commerce customers, this study proposes a unique heterogeneous graph structure to model sequential buying order and rating feedback and develops a query-based recommendation approach to predict future purchase behavior. Experiments on four different datasets show that the approach demonstrates improved performance for different metrics.

O‐SCDR: Optimal cluster with attention based shared‐account cross‐domain sequential recommendation using deep reinforcement learning technique

AbstractSequential recommendation involves suggesting subsequent items in a series of user activities. When recommending relevant items to users within the same account, the challenge lies in discerning diverse user behaviours to provide tailored recommendations based on individual preferences and timing. Cross‐domain sequential recommendation (CDSR) focuses on accurately extracting cross‐domain user preferences from both within‐sequence and between‐sequence interactions among items. Current approaches typically concentrate on learning preferences within a single domain using intra‐sequence item interactions, followed by a transfer module for cross‐domain preferences. However, this sequential process and implicit method are constrained by the effectiveness of the transfer module and may overlook inter‐sequence item associations. In this study, we propose an optimal cluster with attention‐based shared‐account cross‐domain sequential recommendation (O‐SCSR) system using deep reinforcement learning techniques. Our approach commences by formulating a modified hummingbird optimization (MHO) algorithm for clustering, effectively identifying latent users who share the same account to enhance the understanding of user interactions within shared‐account scenarios. Additionally, we design a domain filter based on quantum classic deep reinforcement learning (QCDRL), intelligently selecting interactions contributing to O‐SCSR. By quantifying rewards from transferred domain knowledge, the QCDRL‐based filter retains only valuable interactions for the task of SCDR. Finally, we validate the efficacy of our proposed O‐SCDR method using real‐world datasets, namely HVIDEO and HAMAZON. Through simulation results comparing the O‐SCDR system with existing state‐of‐the‐art systems, we demonstrate its effectiveness and legitimacy.

SMLP4Rec: An Efficient All-MLP Architecture for Sequential Recommendations

Self-attention models have achieved the state-of-the-art performance in sequential recommender systems by capturing the sequential dependencies among user–item interactions. However, they rely on adding positional embeddings to the item sequence to retain the sequential information, which may break the semantics of item embeddings due to the heterogeneity between these two types of embeddings. In addition, most existing works assume that such dependencies exist solely in the item embeddings, but neglect their existence among the item features. In our previous study, we proposed a novel sequential recommendation model, i.e., MLP4Rec, based on the recent advances of MLP-Mixer architectures, which is naturally sensitive to the order of items in a sequence because matrix elements related to different positions of a sequence will be given different weights in training. We developed a tri-directional fusion scheme to coherently capture sequential, cross-channel, and cross-feature correlations with linear computational complexity as well as much fewer model parameters than existing self-attention methods. However, the cascading mixer structure, the large number of normalization layers between different mixer layers, and the noise generated by these operations limit the efficiency of information extraction and the effectiveness of MLP4Rec. In this extended version, we propose a novel framework – SMLP4Rec for sequential recommendation to address the aforementioned issues. The new framework changes the flawed cascading structure to a parallel mode, and integrates normalization layers to minimize their impact on the model’s efficiency while maximizing their effectiveness. As a result, the training speed and prediction accuracy of SMLP4Rec are vastly improved in comparison to MLP4Rec. Extensive experimental results demonstrate that the proposed method is significantly superior to the state-of-the-art approaches. The implementation code is available online to ease reproducibility.

Implementing sharing platform based on ontology using a sequential recommender system

<span lang="EN-US">While recommender systems have shown success in many fields, accurate recommendations in industrial settings remain challenging. In maintenance, existing techniques often struggle with the “cold start” problem and fail to consider differences in the target population's characteristics. To address this, additional user information can be incorporated into the recommendation process. This paper proposes a recommender system for recommending repair actions to technicians based on an ontology (knowledge base) and a sequential model. The approach utilizes two ontologies, one representing failure knowledge and the other representing asset attributes. The proposed method involves two steps: i) calculating score similarity based on ontology domain knowledge to make predictions for targeted failures and ii) generating Top-N repair actions through collaborative filtering recommendations for targeted failures. An additional module was implemented to evaluate the recommender system, and results showed improved performance.</span>

Graph convolutional network and self-attentive for sequential recommendation

Sequential recommender systems (SRS) aim to provide personalized recommendations to users in the context of large-scale datasets and complex user behavior sequences. However, the effectiveness of most existing embedding techniques in capturing the intricate relationships between items remains suboptimal, with a significant concentration of item embedding vectors that hinder the improvement of final prediction performance. Nevertheless, our study reveals that the distribution of item embeddings can be effectively dispersed through graph interaction networks and contrastive learning. In this article, we propose a graph convolutional neural network to capture the complex relationships between users and items, leveraging the learned embedding vectors of nodes to represent items. Additionally, we employ a self-attentive sequential model to predict outcomes based on the item embedding sequences of individual users. Furthermore, we incorporate instance-wise contrastive learning (ICL) and prototype contrastive learning (PCL) during the training process to enhance the effectiveness of representation learning. Broad comparative experiments and ablation studies were conducted across four distinct datasets. The experimental outcomes clearly demonstrate the superior performance of our proposed GSASRec model.

T3SRS: Tensor Train Transformer for compressing sequential recommender systems

In recent years, attention mechanisms have gained popularity in sequential recommender systems (SRSs) due to obtaining dynamic user preferences efficiently. However, over-parameterization of these models often increases the risk of overfitting. To address this challenge, we propose a Transformer model based on tensor train networks. Initially, we propose a tensor train layer (TTL) to accommodate the original weight matrix, thus reducing the space complexity of the mapping layer. Based on the TTL, we reconfigure the multi-head attention module and the position-wise feed-forward network. Finally, a tensor train layer replaces the output layer to complete the overall compression. According to the experimental results, the proposed model compresses SRSs parameters effectively, achieving compression rates of 76.2%−85.0%, while maintaining or enhancing sequence recommendation performance. To our knowledge, the Tensor Train Transformer is the first model compression approach for Transformer-based SRSs, and the model is broadly applicable.

Multi-scale filter-enhanced transformer for sequential recommendation

Recent research has shown that self-attention mechanisms improve Sequential Recommender Systems (SRS) by capturing sequential associations with the interactions. Nevertheless, existing work still needs to address two critical limitations. Firstly, the behavior of users in the original sequences contains various preference signals that are implicit and noisy and hard to reflect the users intentions fully. As a result, it would deteriorate the representation of their true intentions to model all interactions. Secondly, most models only model single-scale interaction sequences and ignore the multi-scale feature relationships of the sequences. In order to address these limitations, the paper proposes MFTSRec (Multi-scale Filter Enhanced Transformer Sequential Recommender), which can weaken those interactions irrelevant to the users intentions from their implicit feedback and adaptively focus on the users multi-scale intentions. Besides, this paper also does extensive experiments on four benchmark datasets and further demonstrates the effectiveness and robustness of MFTSRec compared to the state-of-the-art model.

A Survey and Taxonomy of Sequential Recommender Systems for E-commerce Product Recommendation

A Survey and Taxonomy of Sequential Recommender Systems for E-commerce Product Recommendation

Popularity-aware sequential recommendation with user desire

In recent years, sequential recommendation system has attracted more and more attention from both academy and industry. Sequential recommendation treats the user’s historical behaviors as a sequence and captures the sequential dependency from it. Though the works on sequential recommendation have made great progress, sequential recommendation models still suffer from popularity bias. The existence of popularity bias is common in recommendation systems: the popular items are recommended overly while the less popular items which users may be interested in get fewer chances to be recommended than they should, which may negatively affect the recommendation accuracy and induce other problems such as Matthew Effect and Echo Chambers. Although there have been some works on alleviating popularity bias in traditional recommendation, there is still a lack of research on popularity bias in sequential recommendation. Because there exists complicated sequential dependency in sequences, the debiasing models designed for traditional recommendation cannot be applied directly, it is necessary to propose a debiasing model for sequential recommendation to take the complicated sequential dependency into account. To alleviate the popularity bias in sequential recommendation, we first propose a causal graph for sequential recommendation in which we propose to consider the dynamic user desire which would affect the effect of popularity. Then, we conduct intervention analysis and counterfactual reasoning to quantify the effect of popularity and reason about the user interaction probability in a hypothetical situation that the popularity is set to a certain level. Based on the results of obtained, we propose a new popularity-aware sequential recommendation model with user desire (PAUDRec) which consists of a Transformer-based module, a user desire module and a popularity effect module. Extensive experiments on four widely used benchmark datasets demonstrate that the proposed PAUDRec model outperforms the state-of-the-art sequential recommendation models while alleviating the popularity bias in sequential recommendation.

Dynamic Multi-Objective Optimization Framework With Interactive Evolution for Sequential Recommendation

In contrast to traditional recommender systems which usually pay attention to users' general and long-term preferences, sequential recommendation (SR) can model users' dynamic intents based on their behaviour sequences and suggest the next item(s) to them. However, most of existing sequential models learn the ranking score of an item only based on its relevance property, and the personalized user demands in terms of different learning objectives, such as diversity, tail novelty or recency, which have been played essential roles in multi-objective recommendation (MOR), are often neglected in SR. In this paper, we first discuss the importance of considering multiple different objectives within a learning model for recommender system. Next, to capture users' objective-level preferences by utilizing interactive information in the sequential context, we propose a novel Dynamic Multi-objective Recommendation (DMORec) framework with interactive evolution for SR. In particular, DMORec formulates a dynamic multi-objective optimization task to simultaneously optimize more than two varying objectives in an interactive recommendation process. Moreover, to resolve this optimization task in SR, we develop an evolutionary algorithm with supervised learning approach to obtain the Pareto-optimal solutions of the formulated problem. Comprehensive experiments on four real-world datasets demonstrate the effectiveness of the proposed DMORec for dynamic multi-objective recommendation in sequential recommender systems.

Node representation learning with graph augmentation for sequential recommendation

Current sequential recommender systems primarily focus on modeling users' dynamic interests over time by leveraging GNN-based approaches. However, most studies tend to overlook the impact of high-order data outside the sequence and its corresponding position information on users' interest, especially fail to maximize the collaborative signals between users or items. To alleviate the aforementioned problems, we propose a novel sequential recommendation model, called GNSR, which utilizes graph structures at various levels to enhance the node representation and ultimately improve recommendation performance. Specifically, we construct a high-dimensional continuous-time bipartite graph centered on target users and their historical sequences while preserving the interaction order. Subsequently, we create meta-path graphs through meta-path exploration. Additionally, to capture both effective high-dimensional collaborative signals and dynamic changes in users' interests throughout the sequence, we design a High-Dimensional Information Collaboration Module within GNSR, which enables a comprehensive understanding of user behavior. Finally, we conduct extensive experiments on three public datasets and demonstrate that the performance of GNSR improves by approximately 0.1% ∼ 3.2% over selected baselines.

Edge-enabled federated sequential recommendation with knowledge-aware Transformer

The increasing ubiquity of recommendation systems in modern applications has brought about significant changes in various aspects of our daily lives. However, the emergence of privacy concerns has highlighted the need for innovative solutions to address the challenges faced by traditional recommender systems. In this regard, federated learning provides an approach to address privacy concerns by training global models based on edge data. However, the integration of the existing federated learning framework with the latest recommendation system method remains insufficiently explored. To address this issue, we present edge-enabled federated Sequential Recommendation with Knowledge-aware Transformer (KG-FedTrans4Rec) model that incorporates knowledge graph information into sequential recommendation tasks while applying federated learning for privacy preservation. The KG extraction module and sequential processing module in KG-FedTrans4Rec are designed to capture item-related information and sequence inner relationships simultaneously. We leverage Graph Convolutional Networks (GCNs) to aggregate item-related information and user preferences in the KG extraction module. Furthermore, sequential recommender systems model users’ dynamic preferences based on their behaviors to make customized recommendations. In the sequential processing module, we employ replaced token detection and two-stream self-attention strategies to enhance the Transformer-based model. In summary, our proposed KG-FedTrans4Rec model presents a approach to sequential recommendation tasks by incorporating knowledge graph information while preserving user privacy through federated learning. Our experimental results demonstrate the superior performance of our model compared to various recommendation models. This research is expected to make significant contributions to the development of privacy-preserving recommendation systems, which are increasingly essential in modern applications.

Dynamic Multi-Behavior Sequence Modeling for Next Item Recommendation

Sequential Recommender Systems (SRSs) aim to predict the next item that users will consume, by modeling the user interests within their item sequences. While most existing SRSs focus on a single type of user behavior, only a few pay attention to multi-behavior sequences, although they are very common in real-world scenarios. It is challenging to effectively capture the user interests within multi-behavior sequences, because the information about user interests is entangled throughout the sequences in complex relationships. To this end, we first address the characteristics of multi-behavior sequences that should be considered in SRSs, and then propose novel methods for Dynamic Multi-behavior Sequence modeling named DyMuS, which is a light version, and DyMuS+, which is an improved version, considering the characteristics. DyMuS first encodes each behavior sequence independently, and then combines the encoded sequences using dynamic routing, which dynamically integrates information required in the final result from among many candidates, based on correlations between the sequences. DyMuS+, furthermore, applies the dynamic routing even to encoding each behavior sequence to further capture the correlations at item-level. Moreover, we release a new, large and up-to-date dataset for multi-behavior recommendation. Our experiments on DyMuS and DyMuS+ show their superiority and the significance of capturing the characteristics of multi-behavior sequences.

RSS: Effective and Efficient Training for Sequential Recommendation using Recency Sampling

Many modern sequential recommender systems use deep neural networks, which can effectively estimate the relevance of items, but require a lot of time to train. Slow training increases the costs of training, hinders product development timescales and prevents the model from being regularly updated to adapt to changing user preferences. The training of such sequential models involves appropriately sampling past user interactions to create a realistic training objective. The existing training objectives have limitations. For instance, next item prediction never uses the beginning of the sequence as a learning target, thereby potentially discarding valuable data. On the other hand, the item masking used by the state-of-the-art BERT4Rec recommender model is only weakly related to the goal of the sequential recommendation; therefore, it requires much more time to obtain an effective model. Hence, we propose a novel Recency-based Sampling of Sequences (RSS) training objective (which is parameterized by a choice of recency importance function) that addresses both limitations. We apply our method to various recent and state-of-the-art model architectures – such as GRU4Rec, Caser, and SASRec. We show that the models enhanced with our method can achieve performances exceeding or very close to the effective BERT4Rec, but with much less training time. For example, on the MovieLens-20M dataset, RSS applied to the SASRec model can result in a 60% improvement in NDCG over a vanilla SASRec, and a 16% improvement over a fully-trained BERT4Rec model, despite taking 93% less training time than BERT4Rec. We also experiment with two families of recency importance functions and show that they perform similarly. We further empirically demonstrate that RSS-enhanced SASRec successfully learns to distinguish differences between recent and older interactions – a property that the original SASRec model does not exhibit. Overall, we show that RSS is a viable (and frequently better) alternative to the existing training objectives, which is both effective and efficient for training sequential recommender model when the computational resources for training are limited.

Live streaming recommendations based on dynamic representation learning

As an emerging form of social media, live streaming services (e.g., Twitch and Clubhouse) allow users to interact with hosts and peers in real time while enjoying shows or participating in discussions. These platforms are also dynamic, with shows or discussions changing quickly inside a room and users frequently switching between rooms. To improve user engagement and experience on such platforms, we design a new recommendation model named Dynamic Representations for Live Streaming Rooms (DRIVER) to provide room recommendations. Guided by the Integrated Framework for Consumer Path Modeling and the social affordance theory, DRIVER infers dynamic representations of live streaming rooms by leveraging users’ behavior paths in entering, staying in, and leaving rooms. One contribution of our model is a new and efficient dynamic learning framework to model instantaneous and ever-changing inter-room relationships by considering individual users’ behavior paths after leaving a room. Also supported by social affordance theory, another methodological novelty of our model is to capture dynamic characteristics of a room by incorporating features of the current audience inside the room. Experiments on real-world datasets from two different types of live streaming platforms demonstrate that DRIVER outperforms state-of-the-art representation learning methods and sequential recommender systems. The proposed method also has implications for recommender system design in other contexts, in which items are characterized by users’ dynamic behavior paths and ongoing social interactions.

Mining Implicit Behavioral Patterns via Attention Networks for Sequential Recommendation

Sequential recommender systems (SRSs) seek to model users’ dynamic preferences based on their interaction sequences to suggest items they may be interested in. Existing methods suffer from two major flaws in modeling user interest representations. First, most methods only encode the absolute positions of items in an interaction sequence, but ignore the relative position information between items, impeding the modeling of temporal patterns of user behaviors. At the same time, the multi-layer stacked network structure will cause the dilution of position information. Second, historical interactions usually contain many items that are irrelevant to the user's subsequent choices, while existing work lacks approaches to explicitly distinguish them from interest-related items. In this work, we alleviate the foregoing issues by proposing a user behavioral patterns enhanced attention network (BPERec), where a novel position encoding module enables the enhancement of temporal patterns implicit in interaction sequences. Additionally, an explicit denoising attention mechanism is proposed to enhance user interest representations. Experimental results on three benchmark datasets reveal that our method outperforms state-of-the-art baseline methods.

Evolution of Deep Learning-Based Sequential Recommender Systems: From Current Trends to New Perspectives

The recommender system which gets higher in practical use in applying the Apriori algorithm in the early 2000s has revolutionized our daily life as it currently is widely used by big-tech platform companies. In the early stages of the development of recommender systems, services that can be provided to users were simply derived to the extent that only related products were recommended. However, the new research wave like deep learning-based recommender systems due to the development of information technology and the complexity of users’ online behavior extensively grabs researchers’ and academia’s attention in the field of recommender systems. This paper describes the algorithms and characteristics of the recent popular deep learning-based representative models such as recurrent neural networks (RNNs), convolutional neural networks (CNNs), generative adversarial networks (GANs), and graph neural networks (GNNs) in the view of the sequential recommendation. The sequential recommendation of understanding user preferences in chronological order is useful for analyzing user-item interaction more accurately and flexibly. Therefore, the models specialized in sequential recommendation take advantage of understanding user behavior through temporal factors and improving recommendation quality by easily realizing the correlation between user and items. Also, the transformer-based model was developed to improve the problem of long-term dependency between users and items through factors, such as points, lines, and nodes, experienced in the early models of RNN and CNN and self-supervised learning (SSL)-based models, which are originally purposed to solve the data sparsity issues of recommender systems, will be discussed in this paper.

Modeling and predicting user preferences with multiple item attributes for sequential recommendations

Sequential recommendations have become a focus of attention across the deep learning community owing to their fitness to the actual application scenario. Although recently we have witnessed a surge of work on sequential recommender systems, they are still insufficient in exploring and exploiting item-attribute relations to enhance prediction accuracy. In this work, we propose a novel technological framework, MIA-SR, for sequential recommendation (SR) by modeling and predicting user preferences with multiple item attributes (MIA). When modeling the dynamic behavior of a user, not only the item sequence but also the attribute sequence is used to generate the fused representation of users. Further, we propose using a graph convolution network on the item-attribute bipartite graph to enhance the representations of items and attribute entities. Moreover, MIA-SR is naturally empowered with a multi-tasking strategy to exploit inductive bias among different preference signals and enhance item recommendation. Extensive experiments on public benchmark datasets have verified the merits of MIA-SR. The source code and data are available at: https://github.com/619496775/MIA-SR.

Offline Evaluation for Reinforcement Learning-Based Recommendation

In this paper, we argue that the paradigm commonly adopted for offline evaluation of sequential recommender systems is unsuitable for evaluating reinforcement learning-based recommenders. We find that most of the existing offline evaluation practices for reinforcement learning-based recommendation are based on a next-item prediction protocol, and detail three shortcomings of such an evaluation protocol. Notably, it cannot reflect the potential benefits that reinforcement learning (RL) is expected to bring while it hides critical deficiencies of certain offline RL agents. Our suggestions for alternative ways to evaluate RL-based recommender systems aim to shed light on the existing possibilities and inspire future research on reliable evaluation protocols.

Double Attention Convolutional Neural Network for Sequential Recommendation

The explosive growth of e-commerce and online service has led to the development of recommender system. Aiming to provide a list of items to meet a user’s personalized need by analyzing his/her interaction 1 history, recommender system has been widely studied in academic and industrial communities. Different from conventional recommender systems, sequential recommender systems attempt to capture the pattern of users’ sequential behaviors and the evolution of users’ preferences. Most of the existing sequential recommendation models only focus on user interaction sequence, but neglect item interaction sequence. An item interaction sequence also contains rich contextual information for capturing the item’s dynamic characteristic, since an item’s dynamic characteristic can be reflected by the users who interact with it in a period. Furthermore, existing dual sequential models use the same method to handle the user interaction sequence and item interaction sequence, and do not consider their different characteristics. Hence, we propose a novel D ouble A ttention C onvolution N eural N etwork (DACNN) , which incorporates user interaction sequence and item interaction sequence into an integrated neural network framework. DACNN leverages the strength of attention mechanism to capture the temporary suitability and adopts CNN to extract local sequential features. Experimental evaluations on the real datasets show that DACNN outperforms the baseline approaches.