Collaborative Filtering Model Research Articles

PGCF: Perception graph collaborative filtering for recommendation

Extensive studies have fully proved the effectiveness of collaborative filtering (CF) recommendation models based on graph convolutional networks (GCNs). As an advanced interaction encoder, however, GCN-based CF models do not differentiate neighboring nodes, which will lead to suboptimal recommendation performance. In addition, most GCN-based CF studies pay insufficient attention to the loss function and they simply select the Bayesian personalized ranking (BPR) loss function to train the model. However, we believe that the loss function is as important as the interaction encoder and deserves more attentions. To address the above issues, we propose a novel GCN-based CF model, named Perception Graph Collaborative Filtering (PGCF). Specifically, for the interaction encoder, we design a neighborhood-perception GCN to enhance the aggregation of interest-related information of the target node during the information aggregation process, while weakening the propagation of noise and irrelevant information to help the model learn better embedding representation. For the loss function, we design a margin-perception Bayesian personalized ranking (MBPR) loss function, which introduces a self-perception margin, requiring the predicted score of the user-positive sample to be greater than that of the user-negative sample, and also greater than the sum of the predicted score of the user-negative sample and the margin. The experimental results on five benchmark datasets show that PGCF is significantly superior to multiple existing CF models.

Average User-Side Counterfactual Fairness for Collaborative Filtering

Recently, the user-side fairness issue in Collaborative Filtering (CF) algorithms has gained considerable attention, arguing that results should not discriminate an individual or a sub-user group based on users’ sensitive attributes (e.g., gender). Researchers have proposed fairness-aware CF models by decreasing statistical associations between predictions and sensitive attributes. A more natural idea is to achieve model fairness from a causal perspective. The remaining challenge is that we have no access to interventions, i.e., the counterfactual world that produces recommendations when each user has changed the sensitive attribute value. To this end, we first borrow the Rubin-Neyman potential outcome framework to define average causal effects of sensitive attributes. Next, we show that removing causal effects of sensitive attributes is equal to average counterfactual fairness in CF. Then, we use the propensity re-weighting paradigm to estimate the average causal effects of sensitive attributes and formulate the estimated causal effects as an additional regularization term. To the best of our knowledge, we are one of the first few attempts to achieve counterfactual fairness from the causal effect estimation perspective in CF, which frees us from building sophisticated causal graphs. Finally, experiments on three real-world datasets show the superiority of our proposed model.

MD-GCCF: Multi-view deep graph contrastive learning for collaborative filtering

Collaborative Filtering (CF), a classical recommender system approach, learns users’ interests and behavioral preferences for items through a user–item interaction graph. CF based on graph neural network (GNN) and CF based on graph contrastive learning (GCL) show strong advantages in both modeling multi-layer signals and solving label sparsity, respectively. However, there are still two key problems to be solved: Most CF models based on (1) GNN suffer from the over-smoothing problem and are unable to aggregate deep collaborative signals and (2) GCL adopts a single aggregation paradigm, resulting in a lack of diversity in the feature representation of collaborative signals. To solve the above problems, a multi-view deep graph contrastive learning for collaborative filtering (MD-GCCF) has been proposed from two perspectives. First, a deep graph collaborative signal aggregation module is proposed to learn potential intention similarity representations for deep collaborative signal propagation within a few layers. Second, a novel multi-view contrastive learning module has been proposed, utilizing both local and global contrastive learning views from the collaborative signal aggregation module to enhance deep structures and semantic features in collaborative signals. MD-GCCF improves by 9.52%, 3.34%, and 2.49% compared to the rival models, respectively, in the Amazon book, Yelp2018, and Gowalla datasets.

AXCF: Aspect‐based collaborative filtering for explainable recommendations

AbstractWith the rapid growth of the e‐commerce market facilitated, users are often overwhelmed by the excessive online information, making item selection challenging. While recommendation services have significantly enhanced user experience and sales, these traditional models often overlook the complexity of user‐item interactions and user preferences based on various item aspects. The proposed AXCF framework innovatively combines graph‐based collaborative filtering (CF), which captures high‐order connectivity between users and items, with aspect‐based sentiment analysis (ABSA) to extract detailed user preferences from online reviews. This approach addresses the limitations of linear relationships in traditional CF models by incorporating deep neural networks and introduces a method to overcome the cold‐start problem using online reviews as auxiliary information. By focusing on main aspects such as food, ambiance, and service derived from restaurant reviews, AXCF provides personalized recommendations with improved accuracy and explanatory power, demonstrating its superiority over existing models through experimental results. This study not only presents a significant advancement in recommender systems but also highlights the importance of high‐order connectivity and aspect‐based preferences in understanding and catering to user needs in the e‐commerce platform.

A hybrid similarity model for mitigating the cold-start problem of collaborative filtering in sparse data

Similarity is a vital component for neighborhood-based collaborative filtering (CF). To improve the quality of recommendation, many similarity methods have been proposed and analyzed in recent decades. However, nearly all traditional similarity methods and many advanced similarity methods only utilize corated items among users to compute their similarity, which provides limited information in cold-start/sparse scenarios and yields misleading results. In addition, although a few advanced hybrid similarity models consider items beyond corated items, which can partly mitigate the above limitation, they still have drawbacks, such as disregarding penalizing noncorated items that have many disadvantages. In this paper, we explore a new robust hybrid similarity model, namely Wasserstein distance-based CF (WCF) model, for mitigating the cold-start problem of CF in sparse data. Specifically, we measure item similarity via the Wasserstein distance, which can help circumvent the drawbacks in the Bhattacharyya coefficient and KL divergence that are used in the literature, and is thus more robust in a cold-start/sparse scenario. Besides, we further design a new multiplicative user similarity formula which identifies all noncorated items as a whole to prioritize the importance of corated items and impair the negative effects of noncorated items, which will also play an important role in a cold-start/sparse scenario. In addition, we also propose two novel heuristic similarity factors to impair the negative effects of popular users and items as supplements. We conduct extensive experiments on five real-world benchmark recommendation datasets to test WCF. The experimental results show the superiority of WCF over other existing similarity methods in cold-start/sparse scenarios.

Increasing the Effectiveness of Prediction in Recommendation Engines Based on Collaborative Filtering

In the era of information abundance, the demand for personalized content recommendations has become paramount. Recommendation engines, particularly those employing collaborative filtering, play a pivotal role in delivering tailored suggestions based on user preferences. As technology evolves, the need to enhance the effectiveness of prediction algorithms within these engines becomes increasingly crucial. This research endeavors to contribute to this evolving landscape by delving into collaborative filtering methodologies, identifying challenges, and proposing novel strategies to elevate the accuracy and relevance of predictions in recommendation systems. Through this exploration, we aim to not only refine existing models but also pave the way for more sophisticated and reliable personalized content recommendations.
 This research aims to enhance prediction accuracy in recommendation engines utilizing collaborative filtering. Through an in-depth exploration of collaborative filtering techniques, we propose innovative approaches to improve the effectiveness of predictions. Our study addresses key challenges in collaborative filtering models, offering insights into refined algorithms and methodologies. By fine-tuning the collaborative filtering process, we anticipate a substantial boost in the overall performance of recommendation engines, ultimately advancing the field of personalized content suggestion. The simulation is performed using Java language and using two datasets Movie Lens 1M and Movie Lens 100K.The proposed model was evaluated using the Mean Absolute Error, Precision, and Recall.
 The proposed model achieved a mean absolute error value ranging between 0.78 and 0.84 using the Movie Lens 100K dataset, and a mean absolute error value ranging between 0.72 and 0.74 using the Movie Lens 1M dataset for different values of the number of user groups. As for precision and recall, the precision of the proposed model ranged between 0.97 and 0.985 using the Movie Lens 100K data set, and a precision value ranging between 0.944 and 0.954 using the Movie Lens 1M data set, also for different values of the number of user groups.
 As for the recall results, the proposed model achieved a recall value ranging between 0.755 and 0.85 using the Movie Lens 100K dataset, and a recall value ranging between 0.72 and 0.75 using the Movie Lens 100K dataset, also for different values of the number of user groups. These results were compared with the PMF, HPF, and NMF algorithms, where the proposed model proved its clear superiority over these algorithms. Using this analysis of the matrix allows us to obtain a good prediction accuracy of users' preferences and to find common groups of people with similar preferences.

Dynamic matrix factorization-based collaborative filtering in movie recommendation services

Movies are a primary source of entertainment, but finding specific content can be challenging given the exponentially increasing number of movies produced each year. Recommendation systems are extremely useful for solving this problem. While various approaches exist, Collaborative Filtering (CF) is the most straightforward. CF leverages user input and historical preferences to determine user similarity and suggest movies. Matrix Factorization (MF) is one of the most popular Collaborative Filtering (CF) techniques. It maps users and items into a joint latent space, using a vector of latent features to represent each user or item. However, traditional MF techniques are static, while user cognition and product variety are constantly evolving. As a result, traditional MF approaches struggle to accommodate the dynamic nature of user-item interactions. To address this challenge, we propose a Dynamic Matrix Factorization CF model for movie recommendation systems (DMF-CF) that considers the dynamic changes in user interactions. To validate our approach, we conducted evaluations using the standard MovieLens dataset and compared it to state-of-the-art models. Our preliminary findings highlight the substantial benefits of DMF-CF, which outperforms recent models on the MovieLens-100K and MovieLens-1M datasets in terms of Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) metrics.

Robust Collaborative Filtering to Popularity Distribution Shift

In leading collaborative filtering (CF) models, representations of users and items are prone to learn popularity bias in the training data as shortcuts. The popularity shortcut tricks are good for in-distribution (ID) performance but poorly generalized to out-of-distribution (OOD) data, i.e., when popularity distribution of test data shifts w.r.t. the training one. To close the gap, debiasing strategies try to assess the shortcut degrees and mitigate them from the representations. However, there exist two deficiencies: (1) when measuring the shortcut degrees, most strategies only use statistical metrics on a single aspect (i.e., item frequency on item and user frequency on user aspect), failing to accommodate the compositional degree of a user–item pair; (2) when mitigating shortcuts, many strategies assume that the test distribution is known in advance. This results in low-quality debiased representations. Worse still, these strategies achieve OOD generalizability with a sacrifice on ID performance. In this work, we present a simple yet effective debiasing strategy, PopGo , which quantifies and reduces the interaction-wise popularity shortcut without any assumptions on the test data. It first learns a shortcut model, which yields a shortcut degree of a user–item pair based on their popularity representations. Then, it trains the CF model by adjusting the predictions with the interaction-wise shortcut degrees. By taking both causal- and information-theoretical looks at PopGo, we can justify why it encourages the CF model to capture the critical popularity-agnostic features while leaving the spurious popularity-relevant patterns out. We use PopGo to debias two high-performing CF models (matrix factorization [ 28 ] and LightGCN [ 19 ]) on four benchmark datasets. On both ID and OOD test sets, PopGo achieves significant gains over the state-of-the-art debiasing strategies (e.g., DICE [ 71 ] and MACR [ 58 ]). Codes and datasets are available at https://github.com/anzhang314/PopGo .

Flexibly manipulating popularity bias for tackling trade-offs in recommendation

Latent factor models in collaborative filtering (CF) have been the state-of-the-art for over a decade and widely studied. Most models learn a representation of each user/item and calculate the inner product between them to perform a recommendation. Meanwhile, the inherent long-tailed data make the model training difficult and can cause popularity bias. Popularity bias mitigation has become one of the central themes of recent years. However, although the problems caused by popularity bias (e.g., the Matthew effect) should not be ignored, the popularity of an item implies its quality or trend. Popularity bias mitigation is rather undesirable for platforms and users who prefer popular items. In this study, we first focus on the inner product model and investigate the good property of the inner product for long-tailed data. The inner product is also employed in state-of-the-art CF models, and its effectiveness has been demonstrated empirically. We find that the inner product allows for modeling the long-tailed user–item interactions using the item vector magnitude while the representations remain identifiable. Based on this good property, we propose a method, DirectMag, that allows a platform and even the user to flexibly manipulate popularity bias. DirectMag determines the magnitude of the vectors directly after the training and can control the degree of popularity bias by adjusting only one parameter. In our experiments, we perform a detailed analysis that is not limited to the value of the average recommendation accuracy. We show that conventional methods inevitably suffer from trade-offs regarding the popularity of items, whereas our method is flexible enough to meet the diverse needs of a platform and the users.

KGFlex: Efficient Recommendation with Sparse Feature Factorization and Knowledge Graphs

Collaborative filtering models have undoubtedly dominated the scene of recommender systems in recent years. However, due to the little use of content information, they narrowly focus on accuracy, disregarding a higher degree of personalization. Meanwhile, knowledge graphs are arousing considerable interest in recommendation models thanks to their ability to enrich the system with content features that captures subtle user-item relations. Nevertheless, with many high-quality features, the models become more complex and challenging to train. We extend KGFlex [ 16 ], a hybrid model that analyzes historical data to understand the semantic features the user decisions depend on. KGFlex represents item features as embeddings, and it models user-item interactions as a factorized entropy-driven combination of the item attributes relevant to the user, thus reducing the complexity and raising the degree of personalization. The method does not neglect long tail items, reducing the popularity bias and ensuring a high level of fairness. The user-item prediction is mediated by the user’s personal views of the embeddings that grant a high degree of expressiveness. This extension analyzes different entropy measurement strategies, an enhanced user negative decision modeling, and assesses the fairness of KGFlex and the impact of its hyperparameters. KGFlex is available at https://split.to/kgflex .

Self-supervised Multimodal Graph Convolutional Network for collaborative filtering

Collaborative filtering (CF) is a central solution for capturing various user-item relationships in building recommender systems. However, when the relationships are sparsely observed, it is challenging to obtain enough signals to infer precise user preferences. Recent studies have attempted to address the sparsity issue by incorporating multimodal information (e.g., image and text) into CF models. However, existing methods mainly focus on capturing modal-specific user preference with multiple unimodal graphs, ignoring the complex nature of user behavior, which is determined by an intricate fusion of multimodal information. Therefore, we develop a Self-supervised Multimodal Graph Convolutional Network (SMGCN), which aims to learn the cross-modal user preferences over multiple modalities with an expressive multimodal fusion on a single graph.More importantly, to facilitate and enhance multimodal fusion in SMGCN, we devise two novel self-supervised learning techniques. 1) Collaborative Multimodal Alignment (CMA) uses contrastive learning to align the domain-specific multimodal semantics with the user-item relational semantics. 2) Multimodal Consistency Regularization (MCR) alleviates the sensitivity on a certain modality and increases model robustness. The experimental results demonstrate that our model consistently outperforms advanced multimodal models on three benchmark datasets.

A Customized Deep Sleep Recommender System Using Hybrid Deep Learning.

This paper proposes a recommendation system based on a hybrid learning approach for a personal deep sleep service, called the Customized Deep Sleep Recommender System (CDSRS). Sleep is one of the most important factors for human life in modern society. Optimal sleep contributes to increasing work efficiency and controlling overall well-being. Therefore, a sleep recommendation service is considered a necessary service for modern individuals. Accurate sleep analysis and data are required to provide such a personalized sleep service. However, given the variations in sleep patterns between individuals, there is currently no international standard for sleep. Additionally, service platforms face a cold start problem when dealing with new users. To address these challenges, this study utilizes K-means clustering analysis to define sleep patterns and employs a hybrid learning algorithm to evaluate recommendations by combining user-based and collaborative filtering methods. It also incorporates feedback top-N classification processing for user profile learning and recommendations. The behavior of the study model is as follows. Using personal information received through mobile devices and data, such as snoring, sleep time, movement, and noise collected through AI motion beds, we recommend sleep and receive user evaluations of recommended sleep. This assessment reconstructs the profile and, finally, makes recommendations using top-N classification. The experimental results were evaluated using two absolute error measurement methods: mean squared error (MSE) and mean absolute percentage error (MAPE). The research results regarding the hybrid learning methods show 13.2% fewer errors than collaborative filtering (CF) and 10.2% fewer errors than content-based filtering (CBF) on an MSE basis. According to the MAPE, the methods are 14.7% more accurate than the CF model and 9.2% better than the CBF model. These results demonstrate that CDSRS systems can provide more accurate recommendations and customized sleep services to users than CF, CBF, and combination models. As a result, CDSRS, a hybrid learning method, can better reflect a user's evaluation than traditional methods and can increase the accuracy of recommendations as the number of users increases.

Personalized recommendation with hybrid feedback by refining implicit data

In personalized recommender systems, the collaborative filtering (CF) recommendation approaches have been widely used to predict the preferences of users in real-world applications. Among them, Bayesian personalized ranking (BPR) attracts much attention as it can easily explore the binary form of implicit feedback. However, it still suffers from the absence problem of negative feedback. To address this issue, this paper proposes a hybrid-feedback collaborative filtering model by jointly exploiting the explicit and implicit feedback. Based on the assumption that users prefer items with high ratings, this work firstly introduces the definition of explicit rating data to the BPR model and further proposes an improved Bayesian personalized ranking (IBPR) model to jointly extract the implicit feedback features of users and items. The IBPR model alleviates the problem of lack of negative feedback and promotes the anti-noise performance of the recommender system. Then the IBPR and BiasSVD (Biased Singular Value Decomposition) models are combined to further extract explicit latent features of users as well as items and construct the hybrid-feedback CF model. In this model, the user–item ranking matrix is reconstructed based on the extracted implicit feedback features, and the rating matrix is constructed based on the extracted explicit feedback features. Our proposed method is evaluated on five public datasets and achieves the competitive performance.

SelfCF: A Simple Framework for Self-supervised Collaborative Filtering

Collaborative filtering (CF) is widely used to learn informative latent representations of users and items from observed interactions. Existing CF-based methods commonly adopt negative sampling to discriminate different items. That is, observed user-item pairs are treated as positive instances; unobserved pairs are considered as negative instances and are sampled under a defined distribution for training. Training with negative sampling on large datasets is computationally expensive. Further, negative items should be carefully sampled under the defined distribution, in order to avoid selecting an observed positive item in the training dataset. Unavoidably, some negative items sampled from the training dataset could be positive in the test set. Recently, self-supervised learning (SSL) , has emerged as a powerful tool to learn a model without negative samples. In this paper, we propose a self-supervised collaborative filtering framework (SelfCF) , that is specially designed for recommender scenario with implicit feedback. The proposed SelfCF framework simplifies Siamese networks and can be easily applied to existing deep-learning based CF models, which we refer to as backbone networks. The main idea of SelfCF is to augment the latent embeddings generated by backbone networks instead of the raw input of user/item ids. We propose and study three embedding perturbation techniques that can be applied to different types of backbone networks including both traditional CF models and graph-based models. The framework enables learning informative representations of users and items without negative samples, and is agnostic to the encapsulated backbones. We conduct experimental comparisons on four datasets, one self-supervised framework, and eight baselines to show that our framework may achieve even better recommendation accuracy than the encapsulated supervised counterpart with a 2×–4× faster training speed. The results also demonstrate that SelfCF can boost up the accuracy of a self-supervised framework BUIR by 17.79% on average and shows competitive performance with baselines.

White Box: On the Prediction of Collaborative Filtering Recommendation Systems’ Performance

Collaborative Filtering (CF) recommendation algorithms are a popular solution to the information overload problem, aiding users in the item selection process. Relevant research has long focused on refining and improving these models to produce better (more effective) recommendations, and has converged on a methodology to predict their effectiveness on target datasets by evaluating them on random samples of the latter. However, predicting the efficiency of the solutions—especially with regard to their time- and resource-hungry training phase, whose requirements dwarf those of the prediction/recommendation phase—has received little to no attention in the literature. This article addresses this gap for a number of representative and highly popular CF models, including algorithms based on matrix factorization, k-nearest neighbors, co-clustering, and slope one schemes. To this end, we first study the computational complexity of the training phase of said CF models and derive time and space complexity equations. Then, using characteristics of the input and the aforementioned equations, we contribute a methodology for predicting the processing time and memory usage of their training phase. Our contributions further include an adaptive sampling strategy, to address the tradeoff between resource usage costs and prediction accuracy, and a framework that quantifies both the efficiency and effectiveness of CF. Finally, a systematic experimental evaluation demonstrates that our method outperforms state-of-the-art regression schemes by a considerable margin, with an overhead that is a small fraction of the overall requirements of CF training.

Neural Reranking-Based Collaborative Filtering by Leveraging Listwise Relative Ranking Information

Reranking is a critical task used to refine the initial collaborative filtering (CF) recommendation by incorporating information from different viewpoints, such as the extra item side-information and user profile. In this article, a neural reranking-based CF (NRCF) model is proposed to leverage composite viewpoints from the basic CF model and user preference. More precisely, the predictive implicit user preference is first constructed from the initial top- <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$k$ </tex-math></inline-formula> items. The implicit user preference is then aggregated with the explicit user embedding to enrich the user intent representation. Moreover, the traditional listwise loss functions for reranking optimization are suboptimal, due to the fact that they neglect the relative ranking information (ReinRank) between the unobserved and positive items. To address this issue, a novel listwise loss function that leverages relative ranking information, referred to as ReinRank, is proposed for reranking optimization. ReinRank assigns different values to the unobserved items, according to their relative ranking distances between the positive items. Extensive experiments are performed on three public benchmarks and different CF models, in order to demonstrate the effectiveness of NRCF and ReinRank.

Content Based Movie Recommendation System

Abstract: Recommender System is a tool which helps users find the required content and overcome information overload. It predicts interests of users by using Machine Learning algorithms and makes recommendation according to the interest of users. The primary content-based recommender system is the continuation and development of collaborative filtering, which does not need the user’s appraisal for items. Instead, the similarity is calculated based on the data of items that are selected by users, and then make the recommendation appropriately. With the augmentation of machine learning, the current content-based recommender system can build profile for users and products respectively. Building or renewing the profile according to the perusal of items that are bought or seen by users. The system can differentiate the user and the profile of items and then recommend the most resembling products. So, this recommender method that compel user and product directly can’t be brought into collaborative filtering model. The groundwork of content-based algorithm is acquisition and quantitative analysis of the content. The research of acquisition and filtering of text information are fully fledged, many current modified content-based recommender systems make recommendations according to the analysis of text data. This paper introduces content-based recommendation system for the movie websites. There are a lot of factors extracted from the movie, they are diverse and unique, which is also different from other recommender systems. We use these aspects to construct movie model and calculate similarity. We introduce a new outlook for setting weight of features, which improvises the representation of movie recommendations. Finally, we evaluate the approach to illustrate the improvement

CoMix: Collaborative filtering with mixup for implicit datasets

Collaborative filtering (CF) is the prevalent solution to mitigate massive information overload in modern recommender systems. However, it usually suffers from data sparsity and popularity bias problems. Existing studies exploit auxiliary information or data augmentation, requiring additional data collection or expensive computational overheads. Inspired by Mixup used in the classification problem, we propose a simple-yet-effective data augmentation method for CF, namely Collaborative Mixup (CoMix), for implicit feedback datasets. The underlying idea of CoMix is to generate virtual users/items by logically combining random users/items. Unlike the original Mixup, we synthesize virtual users/items to complement weak collaborative signals by distinguishing the intersection and non-overlapping parts between two users/items. Despite its simplicity, extensive experimental results show that various CF models equipped with CoMix consistently improve the base models on four benchmark datasets.

Graph Collaborative Filtering Based on Dual-Message Propagation Mechanism.

The recommender system is a popular research topic in the past decades, and various models have been proposed. Among them, collaborative filtering (CF) is one of the most effective approaches. The underlying philosophy of CF is to capture and utilize two types of relationships among users/items, that is, the user-item preferences and the similarities among users/items, to make recommendations. In recent years, graph neural networks (GNNs) have gained popularity in many research fields, and in the recommendation field, GNN-based CF models have also been proposed, which are shown to have impressive performance. However, in our research, we observe a crucial drawback of these models, that is, while they can explicitly model and utilize the user-item preferences, the other necessary type of relationship, that is, the similarities among users/items, can only be implied and then utilized, which seems to hinder the performance of these models. Motivated by this, in this article, we first propose a novel dual-message propagation mechanism (DPM). The DPM can explicitly model and utilize both preferences and similarities to make recommendations; thus, it seems to be a better realization of CF's philosophy. Then, a dual-message graph CF (DGCF) model is proposed. Different from the existing models, in the DGCF, each user's/item's embedding is processed by two GNNs, with one handling the preferences and the other handling the similarities. Extensive experiments conducted on three real-world datasets demonstrate that DGCF substantially outperforms state-of-the-art CF models, and the small amount of sacrifice of time efficiency is tolerable considering the substantial improvement of model performance.

Cross-Domain Explicit–Implicit-Mixed Collaborative Filtering Neural Network

Collaborative filtering (CF) is a classical model for recommender systems. Though neural network-based CF models have been shown to perform well in some cases, they still suffer from the sparsity and cold-start issues. To address these issues, we propose a novel neural network-based CF model, termed the cross-domain explicit–implicit-mixed CF neural network (CEICFNet). The proposed model utilizes deep neural networks to learn latent factors not only from the explicit ratings and the implicit interactions but also in a cross-domain manner. In particular, domain-shared multilayer perception (MLP) networks are designed to learn the user rating latent factors and the user interaction latent factors from the explicit ratings and the implicit interactions, respectively, which act as bridges for transferring knowledge across domains. On the other hand, domain-specific MLP networks are designed to learn the item rating latent factors and the item interaction latent factors from the explicit ratings and the implicit interactions, respectively. Then, in each domain, based on the user rating latent factors and the item rating latent factors, the rating predictive representation for each user–item pair can be learned by an MLP. Similarly, the interaction predictive representation for each user–item pair can be learned. For integrating the explicit ratings and the implicit interactions, in each domain, a fully connected layer is used to automatically assign different weights to the rating predictive representations and the interaction predictive representations, based on which the final interaction probability can be generated. Extensive experiments are conducted on five datasets and the results have confirmed the effectiveness of our model.