Matrix factorization learned by implicit alternating least squares (iALS) is a popular baseline in recommender system research publications. iALS is known to be one of the most computationally efficient and scalable collaborative filtering methods. However, recent studies suggest that its prediction quality is not competitive with the current state of the art, in particular autoencoders and other item-based collaborative filtering methods. In this work, we revisit four well-studied benchmarks where iALS was reported to perform poorly and show that with proper tuning, iALS is highly competitive and outperforms any method on at least half of the comparisons. We hope that these high quality results together with iALS’s known scalability spark new interest in applying and further improving this decade old technique.

iALS is a popular algorithm for learning matrix factorization models from implicit feedback with alternating least squares. This algorithm was invented over a decade ago but still shows competitive quality compared to recent approaches like VAE, EASE, SLIM, or NCF. Due to a computational trick that avoids negative sampling, iALS is very efficient especially for large item catalogues. However, iALS does not scale well with large embedding dimensions, d, due to its cubic runtime dependency on d. Coordinate descent variations, iCD, have been proposed to lower the complexity to quadratic in d. In this work, we show that iCD approaches are not well suited for modern processors and can be an order of magnitude slower than a careful iALS implementation for small to mid scale embedding sizes (d ~ 100) and only perform better than iALS on large embeddings d ~ 1000. We propose a new solver iALS++ that combines the advantages of iALS in terms of vector processing with a low computational complexity as in iCD. iALS++ is an order of magnitude faster than iCD both for small and large embedding dimensions. It can solve benchmark problems like Movielens 20M or Million Song Dataset even for 1000 dimensional embedding vectors in a few minutes.

Numerical evaluations with comparisons to baselines play a central role when judging research in recommender systems. In this paper, we show that running baselines properly is difficult. We demonstrate this issue on two extensively studied datasets. First, we show that results for baselines that have been used in numerous publications over the past five years for the Movielens 10M benchmark are suboptimal. With a careful setup of a vanilla matrix factorization baseline, we are not only able to improve upon the reported results for this baseline but even outperform the reported results of any newly proposed method. Secondly, we recap the tremendous effort that was required by the community to obtain high quality results for simple methods on the Netflix Prize. Our results indicate that empirical findings in research papers are questionable unless they were obtained on standardized benchmarks where baselines have been tuned extensively by the research community.

Most recommenders research aims at personalized systems, which suggest items based on user profiles. However, in reality many systems deal with item-oriented recommendations. In such setups, given a single item of interest, the system needs to provide other related items, following patterns like "people who liked this also liked...". While item-oriented systems are central in their importance, they have been approached so far using very basic tools. We identify several hurdles faced by standard approaches to the item-oriented task. First, the sparseness of observed activities prevents establishing reliable similarity relations for many item pairs. Second, we address a scalability challenge at the retrieval stage present in many real-world systems: Given an item inventory, which may encompass millions of items, it is desired to identify the most related item pairs in a sub-quadratic time. This work addresses these two challenges, thereby improving both accuracy and scalability of item-oriented recommenders. Additionally, we propose an empirical evaluation scheme for comparing the quality of different solutions with encouraging results.

We propose a collaborative filtering (CF) recommendation framework which is based on viewing user feedback on products as ordinal, rather than the more common numerical view. Such an ordinal view frequently provides a more natural reflection of the user intention when providing qualitative ratings, allowing users to have different internal scoring scales. Moreover, we can address scenarios where assigning numerical scores to different types of user feedback would not be easy. The framework can wrap most collaborative filtering algorithms, enabling algorithms previously designed for numerical values to handle ordinal values. We demonstrate our framework by wrapping a leading matrix factorization CF method. A cornerstone of our method is its ability to predict a full probability distribution of the expected item ratings, rather than only a single score for an item. One of the advantages this brings is a novel approach to estimating the confidence level in each individual prediction. Compared to previous approaches to confidence estimation, ours is more principled and empirically superior in its accuracy. We demonstrate the efficacy of the approach on two of the largest publicly available datasets: the Netflix data and the Yahoo! Music data.

The massive volume of queries submitted to major Web search engines reflects human interest at a global scale. While the popularity of many search queries is stable over time or fluctuates with periodic regularity, some queries experience a sudden and ephemeral rise in popularity that is unexplained by their past volumes. Typically the popularity surge is precipitated by some real-life event in the news cycle. Such queries form what are known as search trends. All major search engines, using query log analysis and other signals, invest in detecting such trends. The goal is to surface trends accurately, with low latency relative to the actual event that sparked the trend. This work formally defines precision, recall and latency metrics related to top-k search trend detection. Then, observing that many trend detection algorithms rely on query counts, we develop a linear auto-regression model to predict future query counts. Subsequently, we tap the predicted counts to expedite search trend detection by plugging them into an existing trend detection scheme. Experimenting with query logs from a major Web search engine, we report both the stand-alone accuracy of our query count predictions, as well as the task-oriented effects of the prediction on the emitted trends. We show an average reduction in trend detection latency of roughly twenty minutes, with a negligible impact on the precision and recall metrics.

Many websites provide commenting facilities for users to express their opinions or sentiments with regards to content items, such as, videos, news stories, blog posts, etc. Previous studies have shown that user comments contain valuable information that can provide insight on Web documents and may be utilized for various tasks. This work presents a model that predicts, for a given user, suitable news stories for commenting. The model achieves encouraging results regarding the ability to connect users with stories they are likely to comment on. This provides grounds for personalized recommendations of stories to users who may want to take part in their discussion. We combine a content-based approach with a collaborative-filtering approach (utilizing users' co-commenting patterns) in a latent factor modeling framework. We experiment with several variations of the model's loss function in order to adjust it to the problem domain. We evaluate the results on two datasets and show that employing co-commenting patterns improves upon using content features alone, even with as few as two available comments per story. Finally, we try to incorporate available social network data into the model. Interestingly, the social data does not lead to substantial performance gains, suggesting that the value of social data for this task is quite negligible.

In the Internet music scene, where recommendation technology is key for navigating huge collections, large market players enjoy a considerable advantage. Accessing a wider pool of user feedback leads to an increasingly more accurate analysis of user tastes, effectively creating a "rich get richer" effect. This work aims at significantly lowering the entry barrier for creating music recommenders, through a paradigm coupling a public data source and a new collaborative filtering (CF) model. We claim that Internet radio stations form a readily available resource of abundant fresh human signals on music through their playlists, which are essentially cohesive sets of related tracks. In a way, our models rely on the knowledge of a diverse group of experts in lieu of the commonly used wisdom of crowds. Over several weeks, we aggregated publicly available playlists of thousands of Internet radio stations, resulting in a dataset encompassing millions of plays, and hundreds of thousands of tracks and artists. This provides the large scale ground data necessary to mitigate the cold start problem of new items at both mature and emerging services. Furthermore, we developed a new probabilistic CF model, tailored to the Internet radio resource. The success of the model was empirically validated on the collected dataset. Moreover, we tested the model at a cross-source transfer learning manner -- the same model trained on the Internet radio data was used to predict behavior of Yahoo! Music users. This demonstrates the ability to tap the Internet radio signals in other music recommendation setups. Based on encouraging empirical results, our hope is that the proposed paradigm will make quality music recommendation accessible to all interested parties in the community.

Many websites provide commenting facilities for users to express their opinions or sentiments with regards to content items, such as, videos, news stories, blog posts, etc. Previous studies have shown that user comments contain valuable information that can provide insight on Web documents and may be utilized for various tasks. This work presents a model that predicts, for a given user, suitable news stories for commenting. The model achieves encouraging results regarding the ability to connect users with stories they are likely to comment on. This provides grounds for personalized recommendations of stories to users who may want to take part in their discussion. We combine a content-based approach with a collaborative-filtering approach (utilizing users' co-commenting patterns) in a latent factor modeling framework. We experiment with several variations of the model's loss function in order to adjust it to the problem domain. We evaluate the results on two datasets and show that employing co-commenting patterns improves upon using content features alone, even with as few as two available comments per story. Finally, we try to incorporate available social network data into the model. Interestingly, the social data does not lead to substantial performance gains, suggesting that the value of social data for this task is quite negligible.

Modern consumers are inundated with choices. A variety of products are offered to consumers, who have unprecedented opportunities to select products that meet their needs. The opportunity for selection also presents a time-consuming need to select. This has led to the development of recommender systems that direct consumers to products expected to satisfy them. One area in which such systems are particularly useful is that of media products, such as movies, books, television, and music. We study the details of media recommendation by focusing on a large scale music recommender system. To this end, we introduce a music rating data set that is likely to be the largest of its kind, in terms of both number of users, items, and total number raw ratings. The data were collected by Yahoo! Music over a decade. We formulate a detailed recommendation model, specifically designed to account for the data set properties, its temporal dynamics, and the provided taxonomy of items. The paper demonstrates a design process that we believe to be useful at many other recommendation setups. The process is based on gradual modeling of additive components of the model, each trying to reflect a unique characteristic of the data.

Media Websites often solicit users' comments on content items such as videos, news stories, blog posts, etc. Commenting activity increases user engagement with the sites, by both comment writers and readers, and so sites are looking for ways to increase the volume of comments. This work develops a recommender system aiming to present users with items -- news stories, in our case -- on which they are likely to comment. We combine items' content with a collaborative-filtering approach (utilizing users' co-commenting patterns) in a latent factor modeling framework. Building upon previous work, we focus on a continuous, real-time approach to address the problem above. After an initial training period during which commenting activity of users is observed, the system is tested at each subsequent comment submission event by predicting which story is being commented on by a given user at a given time. Our results show that we are able to overcome the site's inherent presentation bias and outperform a strong baseline as users' commenting history grows.