Small adversarial perturbations of input data can drastically change the performance of machine learning systems, thereby challenging their validity. We compare several adversarial attacks targeting ...an instrument classifier, where for the first time in Music Information Retrieval (MIR) the perturbations are computed directly on the waveform. The attacks can reduce the accuracy of the classifier significantly, while at the same time keeping perturbations almost imperceptible. Furthermore, we show the potential of adversarial attacks being a security issue in MIR by artificially boosting playcounts through an attack on a real-world music recommender system.
In recent years, several music recommendation systems have been developed with the aim of incorporating valuable information into the user’s modeling and recommendation process. The inclusion of ...emotions and contextual information in music recommendation applications is increasingly becoming a relevant aspect to improve the listening experience. Thus, the main aim of this systematic literature review (SLR) is investigating the music recommendation approaches that considers emotions and/or context (research question 1) as well as to identify the main gaps and challenges that still remain and need to be addressed by future research (research question 2). After an extensive research, 64 publications were identified to answer the research questions. The studies were analyzed and evaluated for relevance. The main approaches that consider emotions and context were identified. The results of the review indicate that most studies in the field that combine multiple approach related to emotions or context factors have improved the user’s hearing experience. The main contributions of this review are a set of aspects that we consider important to be addressed by the music recommendation systems, such as: user activity, satisfaction, feedback, cold-start problems, cognitive load, learning, personality, and user preference. In addition, we also present a broad discussion about the challenges, difficulties and limitations that exist in music recommendation systems that consider emotions and contextual factors.
This paper proposes a simulation algorithm of transition probability function based on logistic distribution. This method mainly models popularity and state transition probability functions by ...acquiring consumers’ music preferences and likes. Through this mathematical model, this paper obtains the best results that are more in line with consumer preference. This paper conducts a simulation experiment by collecting Netease cloud music data. Finally, through the comparison with the empirical data, it is further demonstrated that the algorithm model in this paper has particular practical value.
Mobile devices such as smart phones are becoming popular, and realtime access to multimedia data in different environments is getting easier. With properly equipped communication services, users can ...easily obtain the widely distributed videos, music, and documents they want. Because of its usability and capacity requirements, music is more popular than other types of multimedia data. Documents and videos are difficult to view on mobile phones' small screens, and videos' large data size results in high overhead for retrieval. But advanced compression techniques for music reduce the required storage space significantly and make the circulation of music data easier. This means that users can capture their favorite music directly from the Web without going to music stores. Accordingly, helping users find music they like in a large archive has become an attractive but challenging issue over the past few years.
With the explosive growth of music volume, music recommendation systems have become an important tool for online music platforms to alleviate the information overload problem. Through the use of deep ...learning, the multi-information fusion-based deep recommendation method has gained popularity in the field of music recommendation systems research. However, most existing studies only consider the different kinds of information of users or music and fail to capture information’s internal and external associations. In this work, we propose a hierarchical multi-information fusion method for deep music recommendation (MMusic), to fully exploit the features of each type of information and to better learn the representation of users and music. Specifically, combined with the features of music recommendation, we identify various kinds of information describing users and music, respectively. Then, we learn about the interactions within and between different kinds of information for fusion. We conduct extensive experiments on the publicly available dataset NOWPLAYINGRS. The results show that MMusic achieves the best performance compared with the baselines, which verifies the effectiveness and rationality of our model.
With the rapid increase of digital music on online music platforms, it has become difficult for users to find unknown but interesting songs. Although many collaborative filtering or content based ...recommendation methods have been proposed, they have various relatively serious some problems, including cold start, diversity of recommendations. etc. Therefore, we propose a reinforcement personal music recommendation system (RPMRS) to address these problems. RPMRS comprises two main components. First, deep representation of audio and lyrics extracted by WaveNet and Word2Vec models, respectively, and apply a proposed content based recommendation method from these. Second, we employ reinforcement learning is to learn user preferences from their song playing log. Experimental results confirm, that hybrid features are superior to audio or lyrics based features for content recommendation, largely because independent audio features significantly outperform lyrics features; and reinforcement learning improves personalized recommendations. Overall, the proposed RPMRS provides dynamic and personalized music recommendations for the user.
The performance of recommender systems highly impacts both music streaming platform users and the artists providing music. As fairness is a fundamental value of human life, there is increasing ...pressure for these algorithmic decision-making processes to be fair as well. However, many factors make recommender systems prone to biases, resulting in unfair outcomes. Furthermore, several stakeholders are involved, who may all have distinct needs requiring different fairness considerations. While there is an increasing interest in research on recommender system fairness in general, the music domain has received relatively little attention. This mini review, therefore, outlines current literature on music recommender system fairness from the perspective of each relevant stakeholder and the stakeholders combined. For instance, various works address gender fairness: one line of research compares differences in recommendation quality across user gender groups, and another line focuses on the imbalanced representation of artist gender in the recommendations. In addition to gender, popularity bias is frequently addressed; yet, primarily from the user perspective and rarely addressing how it impacts the representation of artists. Overall, this narrative literature review shows that the large majority of works analyze the current situation of fairness in music recommender systems, whereas only a few works propose approaches to improve it. This is, thus, a promising direction for future research.
Music streaming services have transformed the way people listen to music in recent years. The current streaming services majorly rely on collaborative and hybrid filtering techniques, which ...predominantly recommend popular songs. However, most present systems lack musical contents, user taste changes, and novelty parameters. In this paper, we propose a music recommendation using reinforcement learning with personalizing the individual results. The proposed method implements a Q-learning model derived from the incremental reinforcement learning algorithm based on the cumulative reward from similar songs played and liked during the session. The user profile is modeled using implicit and explicit feedback from the individual musical interactions with the system. The cumulative reward obtained from the experimental outcomes demonstrates that a combination of reinforcement learning and personalized recommendation potentially broadens the scope of recommendations by including freshness and novelty. The experimental result shows average interaction time improvement of 35% compared with existing apps.