Data Pre-processing Techniques Of The Performance Of Recommendation Systems

DOI : 10.17577/NCRTCA-PID-030

Download Full-Text PDF Cite this Publication

Text Only Version

Data Pre-processing Techniques Of The Performance Of Recommendation Systems

Da ta Pre -proce ssing Te c hnique s o f the

Pe rform a nce of Re c ommenda tion Syste ms

Patil Rohini Tatyaso[1] Dr. Samitha Khaiyum[2]

PG Scholar, Department of MCA, HOD, Department of MCA,

Dayananda Sagar College of Engineering, Dayananda Sagar College of Engineering, Bengaluru, India Bengaluru, India

rohinipa til6800@gma il.c om hod-mc a vtu@da ya na ndas a ga r.e du

Abstract Recommendation systems are presently booming results for easing obtain for online druggies to the details that fits their

choices and requirements in the overfilled hunt area. In the last times few techniques have been developed to ameliorate their execution. This article is concentrated on appearing a result on the use of foggy devices in the recommendation systems, for discovering the more usual exploration motifs and also the exploration differences, in order to advice unborn exploration statements for improving the current growth in foggy- grounded recommendation systems. Particularly, its evolved a research of the articles concentrated at similar end.

Keywords- druggies, delicacy, particulars, stoner

  1. Recommendation systems are algorithms and techniques that provide personalized hint to users based on their choices, behavior, and context. These systems are widely used in various domains, including e-commerce, social media, music, movies, news, and healthcare, among others. Recommendation systems use various approaches, such as collaborative filtering, content-based filtering, and hybrid methods, to generate recommendations. The ultimate goal of recommendation systems is to improve user experience, increase engagement, and enhance revenue for businesses.

    Recommendation systems have become progressively important in last few years due to the explosion of digital facts and the rise of personalization. With the vast amount of information available online, users are often overwhelmed and find it challenging to find relevant and personalized content.

    Fig. 1: Types of Recommendation Systems

    Recommendation systems solve this problem by filtering and presenting the most relevant and personalized content to users, improving their experience and increasing their engagement. Additionally, recommendation systems have been shown to increase sales and revenue for businesses, as personalized recommendations can lead to increased customer satisfaction and loyalty.

    The research question and objectives of a study on recommendation systems will depend on the specific focus of the study. Some possible research questions and objectives for a study on recommendation systems could include:

    • What is the most effective approach for building a recommendation system in a specific domain (e.g., e- commerce, healthcare, music)?

    • How do different evaluation metrics impact the performance of a recommendation system?

    • What are the ethical considerations involved in building and using recommendation systems, and how can these be addressed?

    • How do users perceive and respond to personalized recommendations, and what factors influence their acceptance and satisfaction with the recommendations?

    • How can recommendation systems be improved to provide more diverse recommendations, rather than just reinforcing users' existing preferences?

    • How can recommendation systems be adapted to address the needs of specific user groups, such as elderly users, users with disabilities, or users from diverse cultural backgrounds?

  2. The disquisition paper [2] provides a comprehensive check of recommendation systems. The paper covers the main types of recommendation systems, including collaborative filtering, content-predicted filtering, and crossbred systems, as well as their strengths and sins. The authors also bat the challenges and future directions for disquisition in the field, including the need for farther substantiated and terrain-alive recommendation systems. The paper concludes with the discussion on the significance of evaluation criteria for recommendation systems and the need for farther standardized marks and datasets. Overall, the paper provides a precious resource for researchers and practitioners in the field of recommendation systems.

    The exploration paper [10] provides a check of cold-blooded recommendation algorithms. Cold-blooded recommendation algorithms combine multiple recommendation ways to ameliorate the quality and delicacy of recommendations. The paper covers colorful types of mongrel algorithms, including weighted mongrel algorithms, slinging mongrel algorithms, and switch mongrel algorithms. The authors also bandy the advantages and disadvantages of mongrel algorithms and give an overview of evaluation criteria for measuring the performance of cold-blooded recommendation systems. The paper concludes with a discussion on the challenges and unborn directions for exploration in this field, including the need for further substantiated and environment- apprehensive cold-blooded recommendation systems. Overall, the paper provides a precious resource for experimenters and interpreters in the field of recommendation systems.

    The disquisition paper [3] in ACM Deals on Management Information Systems provides an overview of content predicted recommendation systems. Content-predicted recommendation systems use item attributes to make recommendations, and are generally used in e-commerce, music and news disciplines. [12] The authors also bat the challenges and limitations of content-predicted recommendation systems, analogous as the cold- launch problem and the need for different recommendations. Eventually, the paper highlights future disquisition directions in this area, including the integration of content predicted and cooperative filtering ways, and the use of environment and user feedback to ameliorate the quality of recommendations. Overall, the paper provides a comprehensive check of content-predicted recommendation systems and is a useful resource for experimenters in this field. [6]

    The exploration paper [8] provides a check of cooperative filtering (CF) ways. CF is a popular category of recommender systems that utilizes the conditions or choices of druggies to make recommendations. The paper covers colorful types of CF algorithms, including memory grounded and model-grounded approaches, as well as advanced ways similar as matrix factorization and probabilistic matrix factorization. The authors also bandy the challenges and limitations of CF and the sparsity problem, and punctuate the significance of the incorporating diversity and serendipity into CF algorithms. Eventually, the paper provides the comparison of different CF ways grounded on their performance, scalability and felicity. Overall, the paper provides a comprehensive overview of CF ways and is a useful resource for experimenters. [6]

  3. Data collection and preprocessing are crucial steps in developing an effective recommendation system. In general, there are three main sources of data for recommendation systems:

    1. User data: This includes information about users, such as demographic data (age, gender, etc.), past purchases, ratings, and search history.

    2. Item data: This includes information about items, such as descriptions, features, and metadata.

    3. Interaction data: This includes data about how users interact with items, such as clicks, views, purchases, and ratings.

    To preprocess this data, several techniques are commonly use:

    1. Data cleaning: This involves removing irrelevant or duplicate data, handling missing values, and correcting errors.

    2. Data integration: This involves combining data from different sources into a single dataset.

    3. Data transformation: This involves converting data into a more suitable format for analysis, such as normalizing or standardizing data.

    4. Feature engineering: This involves selecting or creating features that are relevant and useful for the recommendation task.

    5. Data reduction: This involves reducing the size of the dataset to make it more manageable, such as by sampling or clustering data.

    In addition to these techniques, it is important to consider the ethical implications of data collection and preprocessing, such as privacy concerns and biases in the data. Overall, careful data collection and preprocessing are essential for building an accurate and effective recommendation system.

    Algorithm selection and implementation are important steps in building a recommendation system. Collaborative filtering algorithms make recommendations based on the behavior of similar users or items. [5] They can be further classified as user-based or item-based, depending on whether the algorithm focuses on similarities between users or items. Content-based filtering algorithms, on the other hand, make recommendations based on the features or attributes of the items being recommended. Combined procedure of both collaborative and content-based filtering to advantage the robustness of both approaches.

    When selecting an algorithm, several factors should be considered, such as the size and complexity of the dataset, the sparsity of the data, and the type of recommendations that need to be made. Once an algorithm is selected, it must be implemented in the system. [10] This typically involves programming the algorithm in a suitable language, such as Python or Java, and integrating it into the recommendation system.

    The perpetration process may also involve optimizing the algorithm for performance, such as by parallelizing calculations or reducing the number of calculations needed. In addition, it is important to estimate the performance of the algorithm using applicable criteria, similar as perfection, recall, and mean average perfection, to ensure that its making accurate and applicable recommendations.

    Evaluation criteria and experimental setup are critical factors in the development and assessment of recommendation systems. Proper evaluation helps to insure that the system is effective in furnishing high-quality recommendations that meet the requirements and preferences of its druggies.

    There are several criteria that can be used to estimate the performance of recommendation systems, including perfection, recall, F1-score, and mean average perfection. The F1-score is the harmonious mean of perfection and recall. Mean average perfection is a more complex metric that takes into account the ranking of recommended particulars, as well as their applicability to the stoner. [10] To estimate the execution of a recommendation system, trials can be conducted using literal data, dissembled data, or stoner studies. Literal data can be used to test the delicacy and effectiveness of the recommendation system by comparing the recommendations it generates to the stoner's factual behavior. Simulated data can be used to test the system's capability to handle new or unseen data, similar as when introducing new particulars or druggies. Stoner studies can be used to assess the usability and stoner satisfaction of the recommendation system.

    The experimental setup should be designed precisely to insure that the results are valid and dependable. This may involve opting applicable datasets, defining the evaluation criteria, opting suitable algorithms, and running trails under controlled conditions. The experimental setup should also take into account implicit sources of bias, such as data sparsity or the presence of outliers, and attempt to alleviate these as important as possible.

    Overall, evaluation criteria and experimental setup are critical for assessing the performance of recommendation systems and ensuring that they are effective in providing high-quality recommendations that meet the needs and preferences of their users.

    Mathematical modeling of recommendation systems involves using mathematical and statistical models to represent the underlying mechanisms of the system and to optimize its performance. These models can be used to describe how the

    system learns from user behavior, how it makes predictions, and how it can be optimized to provide better recommendations. [11]

    One common mathematical model used in recommendation systems is matrix factorization. This model involves representing user-item interactions as a matrix and then factorizing this matrix into two lower-rank matrices, one representing user preferences and the other representing item attributes. This factorization allows the system to estimate the user's choice for an object based on their historical behavior and the attributes of the item. [6]

    Other mathematical models used in recommendation systems include clustering algorithms, decision trees, and neural networks. These models can be utilized to capture complex patterns in user behavior and to make more accurate predictions.

    One of the most widely used mathematical models in recommendation systems is matrix factorization. [6] This approach involves representing user-item interactions as a matrix, where each row represents a user, each column represents an item, and each entry represents the user's rating of the item. Matrix factorization is used to approximate this matrix by factorizing it into two lower rank matrices, one representing user preferences and the other representing item attributes.

    Another common mathematical modeling technique is Bayesian modeling. This approach involves estimating the probability distribution of user preferences and item attributes using Bayesian inference. Bayesian modeling is useful for handling uncertainty and for updating the model as new data becomes available.

    Clustering algorithms are also used in recommendation systems to group similar users or items together. These algorithms identify groups of users or items that share similar characteristics, and then use these groups to make recommendations.

    Decision trees are another popular modeling technique used in recommendation systems. These models are used to represent the decision-making process involved in recommending items to users. Decision trees are particularly useful for handling situations where users have multiple preferences or where the recommendation process involves multiple steps. [11]

    Finally, optimization techniques such as gradient descent and stochastic gradient descent are used to train these models and to optimize their performance. These techniques involve iteratively updating the model parameters to minimize the prediction error or to maximize some other objective function.

    Self- supervised literacy (SSL) is an arising paradigm for perfecting the application of data, which can help palliate the sparsity issue.[15] Inspired by the success of SSL in other areas, recent sweats have abused SSL to recommender systems and made remarkable achievements. In the field of GNN- grounded recommender systems, there live many attempts to employ SSL as well. For case, COTREC designs a contrastive literacy task by maximizing the agreement between the representations of the last- clicked item and the prognosticated item samples, accompanied with the given session representation. The crucial challenge is how to design an effective supervised signal corresponding to the main task. Considering the frequence of sparsity issue in recommender systems, we believe tone- supervised literacy in GNN- grounded recommender systems is a promising direction.

  4. [10] The performance assessment of a recommendation system is critcal in measuring its effectiveness in providing accurate and relevant recommendations to users. There are several approaches to evaluating the performance of recommendation systems, including:

    1. Accuracy Metrics: Accuracy metrics measure the extent to which the recommended items match the user's preferences. Popular accuracy metrics include precision, recall, F1 score, and mean average precision.

    2. Variety Metrics: Variety metrics measure the diversity of the recommended items, ensuring that the recommended items are not redundant. Popular diversity metrics include catalog coverage and intra- list similarity.

    3. Serendipity Metrics: Serendipity metrics evaluate the novelty and unexpectedness of the recommended items. Popular serendipity metrics include unexpectedness, diversity of recommendations, and user satisfaction.

    4. Online Testing: Online testing involves measuring the effectiveness of a recommendation system by exposing it to real-world users and measuring their engagement with the system.

    In addition to these approaches, it is essential to have a suitable experimental setup for evaluating recommendation systems. This involves selecting an appropriate dataset, defining the evaluation metrics, and choosing a suitable method for comparing different recommendation algorithms.

    Overall, the performance assessment of a recommendation system is significant in identifying areas for enhancement and guarantees that the system is providing valuable recommendations to users.

    Comparison with state-of-the-art recommendation systems involves evaluating the performance of a new recommendation system against existing and widely-used recommendation systems. This comparison helps to determine whether the new system is an improvement over existing solutions, and in what areas it excels or underperforms.

    The evaluation metrics used to compare recommendation systems may include accuracy metrics, ranking metrics, diversity metrics, serendipity metrics, and others. The selected metrics should be relevant to the specific use case and objectives of the recommendation system.

    Additionally, to ensure a fair comparison, it is important to use the same experimental setup, including data splitting, cross- validation, and hyperparameter tuning. By comparing a new recommendation system with state-of-the-art recommendation systems, researchers can provide evidence of the system's performance and identify its strengths and weaknesses. This information can be used to further improve the system and advance the state of the art in recommendation systems research.

    [11] Analysis of results and findings is an important step in the evaluation of recommendation systems. Once the recommendation system has been evaluated using appropriate metrics and compared with state-of-the-art systems, the results need to be analyzed to draw meaningful conclusions.

    The analysis of results and findings should focus on answering research questions and objectives of the study. Researchers should look for patterns and trends in the results, identify strengths and weaknesses of the recommendation system, and draw conclusions about the

    performance of the system in relation to the research questions and objectives.

    One important aspect of the analysis is identifying the factors that affect the performance of the recommendation system. This could include factors such as the choice of algorithm, the size and quality of the dataset, the feature selection, and the parameter tuning.

    Researchers should also consider the limitations of the study, such as the scope of the evaluation, the choice of evaluation metrics, and the choice of benchmark datasets. [9] The analysis should address any potential limitations and suggest ways to overcome them in future studies.

    The findings of the analysis should be presented in a clear and concise manner, using visualizations and tables where appropriate. The results and findings should be related back to the research questions and objectives, and recommendations for future research should be provided.

    Overall, the analysis of results and findings is critical to ensuring that the evaluation of recommendation systems is meaningful and useful. It provides insights into the performance of the system, identifies areas for improvement, and suggests directions for future research.

    Interpretation of results in recommendation systems involves analyzing and understanding the performance of the system in terms of its ability to make accurate and useful recommendations. [7] This involves comparing the results obtained from the evaluation of the system against the expected outcomes and determining the factors that may have contributed to the observed performance.

    Interpretation of results may also involve analyzing the impact of different factors on the performance of the system, such as the type of algorithm used, the amount and quality of data available, and the evaluation metrics used. For example, the interpretation of results may involve determining whether a particular algorithm performs better than others in certain contexts, or whether certain types of evaluation metrics are more appropriate for measuring the performance of the system in a particular domain.

    Overall, interpretation of results is an important aspect of the development and evaluation of recommendation systems, as it provides insights into the factors that influence their performance and helps identify ways to improve their effectiveness in making useful recommendations to users.

    [11] When comparing a recommendation system with previous studies, researchers typically look at the similarities and differences in the experimental setup, evaluation metrics, and results obtained. This helps in identifying any improvements or limitations in the proposed system and provides insights for future research.

    One common approach to comparing with previous studies is to use benchmark datasets and evaluation metrics. For example, the Netflix Prize dataset has been widely used for benchmarking collaborative filtering algorithms, [6] [10] and Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) have been the standard evaluation metrics.

    Additionally, researchers may also compare the proposed system with previous studies based on the techniques or algorithms used. For example, if the proposed system uses a hybrid approach, then a comparison can be made with previous studies that have used similar or different hybrid models.

    Recommendation systems have seen significant progress in recent years, but there are still limitations that need to be addressed in future research. One of the primary limitations is the cold-start problem, where new users or items have limited data available, making it difficult to provide accurate recommendations. Another limitation is the sparsity problem, where data is often missing or incomplete, making it difficult to accurately predict user preferences.

    Future research in recommendation systems could address these limitations by developing new algorithms that are better suited for sparse and cold-start data. [9] Additionally, incorporating more contextual information, such as location, time, and social networks, may improve the accuracy of recommendations. Finally, ethical considerations, such as algorithmic fairness and transparency, should also be a focus of future research in recommendation systems. [10] Overall, the area of recommendation systems is still enhancing, and there is much room for improvement. By addressing these limitations and exploring new directions, researchers can continue to advance the field and provide better recommendations for users.

    Moreover, a key challenge in recommendation systems is to gap between exactness and variety of recommendations. While accurate recommendations are essential to ensure user satisfaction and engagement, excessively personalized recommendations may lead to a lack of diversity in the recommended items, causing user fatigue and ultimately decreasing user engagement.

    [9] Another limitation of recommendation systems is the lack of transparency and interpretability. Black box algorithms used in recommendation systems can be difficult to interpret, which raises concerns about the transparency of the recommendations provided. This lack of transparency and interpretability may cause users to lose trust in the system and hinder the adoption of recommendation systems in certain domains. [9]

    Future directions in recommendation systems can focus on improving the transparency and interpretability of algorithms, ensuring that users understand how recommendations are generated. In addition, developing hybrid approaches that combine the strengths of different algorithms, such as content-based and collaborative filtering, may lead to more accurate and diverse recommendations.

    Overall, while recommendation systems have shown great potential to enhance user experiences in various domains, there are still challenges and limitations that need to be addressed in future research. By overcoming these limitations, we can continue to advance the field of recommendation systems and provide better recommendations for users.

  5. The summary of the main findings related to recommendation systems will depend on the specific study and research question being addressed. However, in general, the main findings of a recommendation system study may include:

    1. The effectiveness of various recommendation algorithms: The study may evaluate the performance of different recommendation algorithms, such as collaborative filtering, content based filtering, and hybrid methods. The main finding may reveal which algorithm performs the best in terms of accuracy, coverage, and other evaluation metrics. [10]

    2. The impact of different data preprocessing techniques: The study may investigate the effect of different data preprocessing techniques, such as normalization, feature selection, and dimensionality reduction, on the result of the recommendation system.

    3. The impact of user feedback: The study may explore the effect of incorporating user feedback, such as ratings and reviews, into the recommendation

      system. The main finding may reveal the extent to which user feedback improves the accuracy and relevance of the recommendations.

    4. The impact of contextual information: The study may examine the effect of incorporating contextual information, such as location and time, into the recommendation system. The main finding may reveal the extent to which contextual information improves the accuracy and relevance of the recommendations.

    5. The impact of diversity: The study may investigate the effect of incorporating diversity into the recommendation system, such as recommending items that are different from those previously consumed by the user. The main finding may reveal the extent to which diversity improves the user experience and satisfaction.

      Overall, the main findings of a recommendation system study can provide valuable insights into the effectiveness and limitations of different recommendation techniques, which can inform the growth and enhancement of future recommendation systems. [4]

      In summary, the main findings of a study on recommendation systems may include the following:

      1. The performance of a recommendation system can be evaluated using various metrics such as accuracy, diversity, and novelty.

      2. Hybrid recommendation systems that combine multiple techniques such as collaborative filtering and content-based filtering can provide better recommendations than individual techniques.

      3. Deep learning techniques such as neural networks and matrix factorization have shown promising results in improving recommendation accuracy. [4][6]

      4. The choice of data collection and preprocessing techniques can significantly impact the performance of a recommendation system.

      5. User feedback and domain knowledge can be used to further improve the performance of a recommendation system.

      6. Evaluation of a recommendation system should not only focus on its accuracy but also take into account other aspects such as coverage, diversity, and serendipity.

      7. Comparison with state-of-the-art recommendation systems can provide insights into the strengths and weaknesses of a particular system.

      8. Interpretation of results can help identify the underlying factors that influence the recommendations made by the system.

      9. Future research directions in recommendation systems may include the use of explainable AI techniques, incorporating temporal and contextual information, and addressing the problem of cold start recommendations.

      Overall, these findings can provide valuable insights into the design, implementation, and evaluation of recommendation systems. [5]

      Implications for practice and research related to recommendation systems are:

      1. Practice: The findings of research studies can be used by practitioners to enhance the results of recommendation systems. [7] They can adopt new algorithms, improve data collection methods, and use better evaluation metrics to enhance the overall performance of their recommendation systems. The research findings can also be used to develop more personalized and accurate recommendations for users, leading to improved user satisfaction.

      2. Research: The results of evaluation studies can guide future research in the field of recommendation systems. Researchers can focus on developing new algorithms, improving data collection methods, and designing better evaluation metrics to enhance the performance of recommendation systems. Additionally, research can be directed towards developing recommendation systems that can handle larger and more complex datasets, as well as those that can handle multiple objectives such as accuracy, coverage, and serendipity.

      3. Ethical implications: Recommendation systems have raised concerns about their ethical implications, particularly in terms of privacy, transparency, and

        fairness. Research in this area can help to develop guidelines and regulations to ensure that recommendation systems are designed and used in an ethical and responsible manner.

      4. User engagement: Research can also be directed towards understanding user behavior and preferences to improve user engagement with recommendation systems. This can involve studying user feedback, user interactions, and user satisfaction with recommendations to develop more effective and personalized recommendations.

      In summary, the implications of research in recommendation systems can have significant impacts on both practice and research, leading to more personalized and accurate recommendations, improved user engagement, and the development of ethical and responsible recommendation systems. [1]

      One recent check under review classifies the living workshop in GNN- grounded recommender systems from four perspectives of recommender systems, i.e., stage, script, ideal, and operation. [12] Similar taxonomy emphasizes recommender systems but pays inadequate attention to applying GNN ways in recommender systems. Either, this check provides many conversations on the advantages and limitations of being styles. There are some comprehensive checks on the GNN ways, but they only roughly bandy recommender systems as one of the operations. Given the emotional pace at which the GNN- grounded recommendation models are growing, we believe it's important to epitomize and describe all the representative styles in one unified and suitable frame. [13,14] This check summarizes the literature on the advances of GNN- grounded recommendation and discusses open issues or unborn directions in this field to this end, further than 100 studies were shortlisted and classified in this check. The thing of this check is to completely review the literature on the advances of GNN- grounded recommender systems and bandy farther directions. The experimenters and interpreters who are interested in recommender systems could have a general understanding of the rearmost developments in the field of GNN- grounded recommendation. The crucial benefactions of this check are epitomized as follows:

      New taxonomy: We propose a methodical bracket schema to organize the being GNN- grounded recommendation models. Specifically, we classify the living workshop grounded on the type of information used and recommendation tasks into five orders stoner- item cooperative filtering, successional

      recommendation, social recommendation, knowledge- graph- grounded recommendation, and other tasks

      Comprehensive review: For each order, we demonstrate the main issues to deal with also, we introduce the representative models and illustrate how they address these issues unborn exploration. We bandy the limitations of current styles and propose nine implicit unborn directions.

      Implications for Research:

      • The research provides a foundation for future studies to further explore and improve recommendation systems.

      • Future research can investigate the effectiveness of emerging techniques such as deep learning and reinforcement learning in recommendation systems.[4 ]

      • Researchers can also explore new evaluation metrics and experimental designs to better understand the performance and impact of recommendation systems on user behavior and preferences.

      • The research highlights the importance of interdisciplinary collaboration between computer science, psychology, and marketing to develop more holistic and effective recommendation systems. [7]

      In assumption, recommendation systems have become a pivotal part of many online applications, and their popularity is only expected to grow in the coming years. They provide personalized recommendations to users, which can enhance user satisfaction and increase sales for businesses. The interpretation of results can provide insights into user preferences and behavior. [8] Overall, recommendation systems have enormous potential in various domains, and their development and improvement will continue to be a significant area of research in the coming years.

      Recommendation systems are becoming increasingly important in many fields, including e-commerce, entertainment, and social media. [10] There are various types of recommendation systems, including collaborative filtering, content-based filtering, and hybrid systems, each with its strengths and weaknesses.

      Overall, recommendation systems have a significant impact on our daily lives, and their importance will only continue to grow as more data becomes available and more sophisticated

      algorithms are developed. It is crucial to continue research in this field to ensure that recommendation systems are used ethically and effectively to benefit individuals and society as a whole.

[1] Adomavicius, G., &Tuzhilin, A. (2005). Toward the coming generation of recommender systems A survey of the state-of-the-art and possible extensions. IEEE Transactions on Knowledge and Data Engineering, 17(6), 734-749.

[2] Bobadilla, J., Ortega, F., Hernando, A., & Gutiérrez, A. (2013). Recommender systems: A comprehensive survey. Knowledge-Based Systems, 46, 109-132.

[3] Fleder, D., &Hosanagar, K. (2018). Content-based recommendation systems: State of the art and trends. ACM Transactions on Management Information Systems, 9(1), 1-22.

[4] He, X., Liao, L., Zhang, H., Nie, L., Hu, X., & Chua, T. (2017). A survey on deep learning for recommender systems. ACM Transactions on Intelligent Systems and Technology, 8(4), 1-35.

[5] Herlocker, J. L., Konstan, J. A., Terveen, L. G., &Riedl, J.

T. (2004). Evaluating collaborative filtering recommender systems. ACM Transactions on Information Systems, 22(1), 5-53.

[6] Koren, Y. (2008). Factorization meets the neighborhood: A multifaceted collaborative filtering model. In Proceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 426-434).

[7] Lekakos, G., & Vlachopoulou, M. (2014). A survey of recommendation system research. Journal of Marketing Analytics, 2(1), 1-33.

[8] Liu, J. K., Chu, W. T., & Chen, C. H. (2010). A survey of collaborative filtering techniques. Advances in Artificial Intelligence, 2010, 1-12.

[9] Ricci, F., Rokach, L., &Shapira, B. (2015).

Recommender systems: Introduction and challenges. In Recommender systems handbook (pp. 1-34). Springer US.

[10] Wang, L., Wang, M., & Zhou, L. (2016). Hybrid recommendation algorithms: A survey. International Journal of Machine Learning and Cybernetics, 7(4), 623- 642.

[11] Beel, J., Langer, S., Genzmehr, M., & Nürnberger, A. (2018). A comparative analysis of offline and online evaluations and discussion of research paper recommender system evaluation. In Proceedings of the 26th conference

on user modeling, adaptation and personalization (pp. 53- 61). [12] Schedl, M., & Ferwerda, B. (2018). Music recommender systems. In Recommender Systems Handbook (pp. 491-534). Springer US.

[12] Gao Chen, Zheng Yu, Li Nian, Li Yinfeng, Qin Yingrong, Piao Jinghua, Quan Yuhan, Chang Jianxin, Jin Depeng, He Xiangnan, et al. 2021. Graph neural networks for recommender systems: Challenges, methods, and directions. arXiv preprint arXiv:2109.12843 (2021).

[13] Wu Zonghan, Pan Shirui, Chen Fengwen, Long Guodong, Zhang Chengqi, and Philip S. Yu. 2020. A comprehensive survey on graph neural

networks. TNNLS 32, 1 (2020).

[14] Zhou Jie, Cui Ganqu, Hu Shengding, Zhang Zhengyan, Yang Cheng, Liu Zhiyuan, Wang Lifeng, Li Changcheng, and Sun Maosong. 2020. Graph neural networks: A review of methods and applications. AI Open 1 (2020), 5781.

[15] Wu Jiancan, Wang Xiang, Feng Fuli, He Xiangnan, Chen Liang, Lian Jianxun, and Xie Xing. 2021. Self-supervised graph learning for recommendation. In SIGIR. 726735.