Search

As of today, most movie recommendation services base their recommendations on collaborative filtering (CF) and/or content-based filtering (CBF) models that use metadata (e.g., genre or cast). In most video-on-demand and streaming services, however, new movies and TV series are continuously added. CF models are unable to make predictions in such a scenario, since the newly added videos lack interactionsa problem technically known as new item cold start (CS). Currently, the most common approach to this problem is to switch to a purely CBF method, usually by exploiting textual metadata. This approach is known to have lower accuracy than CF because it ignores useful collaborative information and relies on human-generated textual metadata, which are expensive to collect and often prone to errors. User-generated content, such as tags, can also be rare or absent in CS situations. In this paper, we introduce a new movie recommender system that addresses the new item problem in the movie domain by (i) integrating state-of-the-art audio and visual descriptors, which can be automatically extracted from video content and constitute what we call the movie genome; (ii) exploiting an effective data fusion method named canonical correlation analysis, which was successfully tested in our previous works Deldjoo et al. (in: International Conference on Electronic Commerce and Web Technologies. Springer, Berlin, pp 3445, 2016b; Proceedings of the Twelfth ACM Conference on Recommender Systems. ACM, 2018b), to better exploit complementary information between different modalities; (iii) proposing a two-step hybrid approach which trains a CF model on warm items (items with interactions) and leverages the learned model on the movie genome to recommend cold items (items without interactions). Experimental validation is carried out using a system-centric study on a large-scale, real-world movie recommendation dataset both in an absolute cold start and in a cold to warm transition; and a user-centric online experiment measuring different subjective aspects, such as satisfaction and diversity. Results show the benefits of this approach compared to existing approaches.

Popularity-based approaches are widely adopted in music recommendation systems, both in industry and research. However, as the popularity distribution of music items typically is a long-tail distribution, popularity-based approaches to music recommendation fall short in satisfying listeners that have specialized music. The contribution of this article is three-fold. We provide several quantitative measures describing the proximity of a user's music preference to the music mainstream. We define the measures at two levels: relating a listener's music preferences to the global music preferences of all users, or relating them to music preferences of the user's country. Moreover, we adopt a distribution-based and a rank-based approach as means to decrease bias towards the head of the long-tail distribution. We analyze differences between countries in terms of their level of mainstreaminess, uncover both positive and negative outliers (substantially higher and lower country-specific popularity, respectively, compared to the global mainstream), and investigate differences between countries in terms of listening preferences related to popular music artists. We use the standardized LFM-1b dataset, from which we analyze about 8 million listening events shared by about 53,000 users (from 47 countries) of the music streaming platform Last.fm. We show that there are substantial country-specific differences in listeners' music consumption behavior with respect to the most popular artists listened to. We conduct rating prediction experiments in which we tailor recommendations to a user's level of preference for the music mainstream using the proposed 6 mainstreaminess measures. Results suggest that, in terms of rating prediction accuracy, each of the presented mainstreaminess definitions has its merits., Comment: 36 pages, 4 figures, 10 tables, PLOS ONE 14(6), paper e0217389

We investigate the complex relationship between the fac- tors (i) preference for music mainstream, (ii) social ties in an online music platform, and (iii) demographics. We define (i) on a global and a country level, (ii) by several network centrality measures such as Jaccard index among users’ connections, closeness centrality, and betweenness centrality, and (iii) by country and age information. Using the LFM-1b dataset of listening events of Last.fm users, we are able to uncover country-dependent differences in consumption of mainstream music as well as in user behavior with respect to social ties and users’ centrality. We could identify that users inclined to mainstream music tend to have stronger connections than the group of less mainstreamy users. Furthermore, our analysis revealed that users typically have less connections within a country than cross-country ones, with the first being stronger social ties, though. Results will help building better user models of listeners and in turn improve personalized music retrieval and recommendation algorithms.

International audience; The traditional role of nearest-neighbor classification in musicclassification research is that of a straw man opponent for the learningapproach of the hour. Recent work in high-dimensional indexinghas shown that approximate nearest-neighbor algorithms are extremelyscalable, yielding results of reasonable quality from billions of high-dimensionalfeatures. With such efficient large-scale classifiers, the traditionalmusic classification methodology of reducing both feature dimensionalityand feature quantity is incorrect; instead the approximatenearest-neighbor classifier should be given an extensive data collectionto work with. We present a case study, using a well-known MIR classificationbenchmark with well-known music features, which shows thata simple nearest-neighbor classifier performs very competitively whengiven ample data. In this position paper, we therefore argue that nearest-neighborclassification has been treated unfairly in the literature and maybe much more competitive than previously thought.

The traditional role of nearest-neighbor classification in music classification research is that of a straw man opponent for the learning approach of the hour. Recent work in high-dimensional indexing has shown that approximate nearest-neighbor algorithms are extremely scalable, yielding results of reasonable quality from billions of high-dimensional features. With such efficient large-scale classifiers, the traditional music classification methodology of aggregating and compressing the audio features is incorrect; instead the approximate nearest-neighbor classifier should be given an extensive data collection to work with. We present a case study, using a well-known MIR classification benchmark with well-known music features, which shows that a simple nearest-neighbor classifier performs very competitively when given ample data. In this position paper, we therefore argue that nearest-neighbor classification has been treated unfairly in the literature and may be much more competitive than previously thought.

Despite their many identified shortcomings, music genres are still often used as ground truth and as a proxy for music similarity. In this work we therefore take another in-depth look at genre classification, this time with the help of music experts. In comparison to existing work, we aim at including the viewpoint of different stakeholders to investigate whether musicians and end-user music taxonomies agree on genre ground truth, through a user study among 20 professional and semi-professional music protagonists. We then compare the results of their genre judgments with different commercial taxonomies and with that of computational genre classification experiments, and discuss individual cases in detail. Our findings coincide with existing work and provide further evidence that a simple classification taxonomy is insufficient.

Recently, the LFM-1b dataset has been proposed to foster research and evaluation in music retrieval and music recommender systems, Schedl (Proceedings of the ACM International Conference on Multimedia Retrieval (ICMR). New York, 2016). It contains more than one billion music listening events created by more than 120,000 users of Last.fm. Each listening event is characterized by artist, album, and track name, and further includes a timestamp. Basic demographic information and a selection of more elaborate listener-specific descriptors are included as well, for anonymized users. In this article, we reveal information about LFM-1b’s acquisition and content and we compare it to existing datasets. We furthermore provide an extensive statistical analysis of the dataset, including basic properties of the item sets, demographic coverage, distribution of listening events (e.g., over artists and users), and aspects related to music preference and consumption behavior (e.g., temporal features and mainstreaminess of listeners). Exploiting country information of users and genre tags of artists, we also create taste profiles for populations and determine similar and dissimilar countries in terms of their populations’ music preferences. Finally, we illustrate the dataset’s usage in a simple artist recommendation task, whose results are intended to serve as baseline against which more elaborate techniques can be assessed. Version of record