Personalized Implicit Learning in a Music Recommender System

A large number of internet and consumer services applications involve predicting user responses to choices. Some examples include media recommendation systems implemented by Netflix \cite{Netflix} and Spotify \cite{Spotify}. Traditionally, recommender systems have been broadly classified into two categories: Collaborative Filtering and Content-Based Recommendation. We present a novel approach to recommend songs. The goal of the system is to recommend a song that is similar to the user's previously heard songs and is also rated highly by other similar user. We achieve this by using both community ratings and the metadata of songs, thus combining the traditional collaborative filtering and content-based recommendation approaches. We train a deep neural-network to predict ratings for the song and then make recommendations.

2007

We aimed at improving the efficiency and scalability of a hybrid music recommender system based on a probabilistic generative model that integrates both collaborative data (rating scores provided by users) and content-based data (acoustic features of musical pieces). Although the hybrid system was proved to make accurate recommendations, it lacks efficiency and scalability. In other words, the entire model needs to be retrained from scratch whenever a new score, user, or piece is added. Furthermore, the system cannot deal with practical numbers of users and pieces on an enterprise scale. To improve efficiency, we propose an incremental method that partially updates the model at low computational cost. To enhance scalability, we propose a method that first constructs a small "core" model over fewer virtual representatives created from real users and pieces, and then adds the real users and pieces to the core model by using the incremental method. The experimental results revealed that the proposed system was not only efficient and scalable but also outperformed the original system in terms of accuracy.

IEEE Transactions on Audio, Speech, and Language Processing, 2008

This paper presents a hybrid music recommender system that ranks musical pieces while efficiently maintaining collaborative and content-based data, i.e., rating scores given by users and acoustic features of audio signals. This hybrid approach overcomes the conventional tradeoff between recommendation accuracy and variety of recommended artists. Collaborative filtering, which is used on e-commerce sites, cannot recommend nonbrated pieces and provides a narrow variety of artists. Content-based filtering does not have satisfactory accuracy because it is based on the heuristics that the user's favorite pieces will have similar musical content despite there being exceptions. To attain a higher recommendation accuracy along with a wider variety of artists, we use a probabilistic generative model that unifies the collaborative and content-based data in a principled way. This model can explain the generative mechanism of the observed data in the probability theory. The probability distribution over users, pieces, and features is decomposed into three conditionally independent ones by introducing latent variables. This decomposition enables us to efficiently and incrementally adapt the model for increasing numbers of users and rating scores. We evaluated our system by using audio signals of commercial CDs and their corresponding rating scores obtained from an e-commerce site. The results revealed that our system accurately recommended pieces including nonrated ones from a wide variety of artists and maintained a high degree of accuracy even when new users and rating scores were added. Index Terms-Aspect model, hybrid collaborative and contentbased recommendation, incremental training, music recommender system, probabilistic generative model. I. INTRODUCTION T HE importance of music recommender systems is increasing because many online services that manage large music collections do not provide users with fully satisfactory access to their collections [1], [2]. Standard retrieval systems Manuscript

Proceedings of the 21st international conference on World Wide Web, 2012

International Journal of Multimedia Data Engineering and Management, 2014

With the explosive growth of the World Wide Web and the rise of social media, new approaches in Music Recommendation evolve. The current study investigates how blogs and micro-blogs can improve the perceived quality of music recommendation. A literature review and expert interviews are conducted to identify important topics regarding (micro-) blogs and Music Recommendation. Subsequently, the prototype Songdice is built and tested in a user-evaluation. Songdice uses music blogs to recommend songs and rationalize those recommendations. The authors' results show that (micro-) blogs can improve the perceived quality of recommendations by creating trust, using personalization and exploiting the quality of music in the long tail. Additional research is required to determine the most effective way to use information from blogs and micro-blogs. The authors' research explores a new area in music recommendation literature and provides a starting point for further research concerning t...

The accuracy of collaborative-filtering recommender systems largely depends on three factors: the quality of the rating prediction algorithm, and the quantity and quality of available ratings. While research in the field of recommender systems often concentrates on improving prediction algorithms, even the best algorithms will fail if they are fed poor quality data during training, i.e. garbage in, garbage out. Active learning aims to remedy this problem by focusing on obtaining better quality data that more aptly reflects a user's preferences. However, traditional evaluation of active learning strategies has two major flaws, which have significant negative ramifications on accurately evaluating the system's performance (prediction error, precision, and quantity of elicited ratings). (1) Performance has been evaluated for each user independently (ignoring system-wide improvements) (2) Active learning strategies have been evaluated in isolation from unsolicited user ratings (natural acquisition).

Proceedings of the 2016 Conference on User Modeling Adaptation and Personalization, 2016

Many collaborative filtering recommender systems collect and use users' explicitly entered preferences in the form of ratings for items. However, in many real world scenarios, this form of feedback can be difficult to obtain or unavailable (e.g., news portals). In this case recommendations must be built by leveraging more abundant implicit feedback data, which only indirectly signal users' preferences or opinions. A record in such datasets is a result of an action performed by a user on an item (e.g., the item was clicked or viewed). State-of-the-art implicit feedback recommender systems predict whether the user will act on a target item and interpret this prediction as a discovered preference for the item. These models are trained by observations of user actions of one single type. For instance, they predict that a user will watch a video using a dataset of observed video watch actions. In this paper we conjecture that multiple types of user actions may be jointly exploited to predict one target type of actions. We present a general prediction model (MMF-Multiple action types Matrix Factorization) that implements this conjecture and we illustrate some practical examples. The empirical evaluation of MMF, which was conducted on a large real world dataset, shows that using multiple actions is beneficial and it can outperform a state-of-the-art implicit feedback model that uses only the target action data.

The track 2 problem in KDD Cup 2011 (music recommendation) is to discriminate between music tracks highly rated by a given user from those which are overall highly rated, but not rated by the given user. The training dataset consists of not only user rating history but also the taxonomic information of track, artist, album, and genre. This paper describes the solution of the National Taiwan University team which ranked first place in the competition. We exploited a diverse of models (neighborhood models, latent models, BPR-based models, and random-walk models) with local blending and global ensemble to achieve 97.45% in accuracy on the testing dataset.

International Journal of Information Technology, 2018

Evaluation strategies are essentials in assessing the degree of satisfaction that recommender systems can provide to users. The evaluation schemes rely heavily on user feedback, however these feedbacks may be casual, biased or spam which leads to an inappropriate evaluation. In this paper, a comprehensive approach for the evaluation of recommendation system is proposed. The implicit user feedbacks are taken for the different products on the basis of the reviews provided to them. A novel sincerity check mechanism is suggested to render the biasedness and casual among the users. Further, mathematical model is presented to classify the products preference criteria. The list of the preferred products yield different ranking. Rank aggregation algorithm is used to obtain a final ranking, which is compared with the base ranking to be evaluated. Hence, with the help of suggested methodology, an evaluation strategy is suggested that avoids the risk of fake and biased feedbacks. The comparison of the proposed approach with existing schemes shows the superiority of the aforementioned approach from various parameters. It is envisaged that the proposed evaluation scheme lays a platform for users to assess the recommender systems for their ease and reliable online shopping.

Computers in Human Behavior, 2012

ABSTRACT The goal of this research is to define and capture a series of parameters that allowed us to perform a comparative analysis and find correlations between explicit and implicit feedback on recommender systems. Most of these systems require explicit actions from the users, such as rating, and commenting. In the context of electronic books this interaction may alter the patterns of reading and understanding of the users, as they are asked to stop reading and rate the content. By simulating the behavior of an electronic book reader we have improved the feedback process, by implicitly capturing, measuring, and classifying the information needed to discover user interests. In these times of information overload, we can now develop recommender systems that are mostly based on the user’s behavior, by relying on the obtained results.