Abstract:Next-item recommender systems are often trained using only positive feedback with randomly-sampled negative feedback. We show the benefits of using real negative feedback both as inputs into the user sequence and also as negative targets for training a next-song recommender system for internet radio. In particular, using explicit negative samples during training helps reduce training time by ~60% while also improving test accuracy by ~6%; adding user skips as additional inputs also can considerably increase user coverage alongside slightly improving accuracy. We test the impact of using a large number of random negative samples to capture a 'harder' one and find that the test accuracy increases with more randomly-sampled negatives, but only to a point. Too many random negatives leads to false negatives that limits the lift, which is still lower than if using true negative feedback. We also find that the test accuracy is fairly robust with respect to the proportion of different feedback types, and compare the learned embeddings for different feedback types.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper mainly explores how to improve the music personalized recommendation system by introducing negative feedback. Specifically, the authors attempt to solve the following key problems: 1. **Improve training efficiency and test accuracy**: - Current recommendation systems usually only use positive feedback and randomly sample negative feedback. This method may lead to overly long model training time and lower accuracy on the test set. - The paper proposes using real negative feedback as input and using it as negative samples for training. Experiments show that this can significantly reduce training time (by about 60%) while increasing test accuracy (by about 6%). 2. **Optimize the selection of negative samples**: - Selecting appropriate negative samples is crucial for improving model performance. The paper studies the impact of using different numbers of randomly - sampled negative samples on model performance. The results show that increasing the number of randomly - sampled negative samples can improve test accuracy, but exceeding a certain number will lead to false negatives, thus degrading model performance. - The authors also explore the effect of using "the hardest" negative samples as negative samples and find that it can better improve model performance. 3. **Utilize user skip behavior**: - The behavior of users skipping songs can be regarded as a kind of implicit negative feedback. The paper studies the impact of using these skip behaviors as additional input on model performance. The results show that adding skip behaviors can not only improve test accuracy but also significantly increase user coverage, especially for those users who mainly provide skip feedback. 4. **Address the cold - start problem**: - For new users or new items, due to the lack of sufficient feedback data, traditional collaborative filtering methods may not be able to provide effective recommendations. The paper explores how to alleviate this problem by introducing more types of feedback (such as skip behavior). 5. **Evaluate the embedding learning of different feedback types**: - The paper compares the embedding learning effects of different types of feedback (such as positive feedback, negative feedback, and skip behavior). The results show that skip behavior is closer to negative feedback rather than positive feedback. ### Summary This paper aims to improve the training efficiency and test accuracy of the music personalized recommendation system by introducing negative feedback and optimizing the negative sample selection strategy. In addition, the paper also explores how to use implicit feedback such as user skip behavior to enhance the performance and coverage of the recommendation system.

Negative Feedback for Music Personalization

Enhancing Sequential Music Recommendation with Negative Feedback-informed Contrastive Learning

Exploiting Negative Preference in Content-based Music Recommendation with Contrastive Learning

Leveraging Negative Signals with Self-Attention for Sequential Music Recommendation

Learning from Negative User Feedback and Measuring Responsiveness for Sequential Recommenders

Fifty Shades of Ratings: How to Benefit from a Negative Feedback in Top-N Recommendations Tasks

Personalized Negative Reservoir for Incremental Learning in Recommender Systems

Recommendations with Negative Feedback via Pairwise Deep Reinforcement Learning

Loss Aversion in Recommender Systems: Utilizing Negative User Preference to Improve Recommendation Quality

Negative Feedback Aware Hybrid Sequential Neural Recommendation Model.

Negative Sampling in Recommendation: A Survey and Future Directions

Generating Negative Samples for Sequential Recommendation

Counteracting User Attention Bias in Music Streaming Recommendation via Reward Modification

Negative Can Be Positive: Signed Graph Neural Networks for Recommendation

Context-Based Music Recommendation Algorithm Evaluation

Online Collaborative Filtering With Implicit Feedback

Evaluating Music Recommendations with Binary Feedback for Multiple Stakeholders

Music recommendation using dynamic feedback and content-based filtering

Enhancing Recommender Systems: A Strategy to Mitigate False Negative Impact

Denoising Implicit Feedback for Recommendation

Augmented Negative Sampling for Collaborative Filtering