Behavioral patterns captured in embeddings learned from interaction data are pivotal across various stages of production recommender systems. However, in the initial retrieval stage, practitioners face an inherent tradeoff between embedding expressiveness and the scalability and latency of serving components, resulting in the need for representations that are both compact and expressive. To address this challenge, we propose a training strategy for learning high-dimensional sparse embedding layers in place of conventional dense ones, balancing efficiency, representational expressiveness, and interpretability. To demonstrate our approach, we modified the production-grade collaborative filtering autoencoder ELSA, achieving up to 10× reduction in embedding size with no loss of recommendation accuracy, and up to 100× reduction with only a 2.5% loss. Moreover, the active embedding dimensions reveal an interpretable inverted-index structure that segments items in a way directly aligned with the model's latent space, thereby enabling integration of segment-level recommendation functionality (e.g., 2D homepage layouts) within the candidate retrieval model itself. Source codes, additional results, as well as a live demo are available at https://github.com/zombak79/compressed\_elsa.

Email outreach remains a cornerstone of modern marketing, enabling direct, timely communication. However, this strategy faces significant personalization challenges, since new campaigns typically lack historical interaction data and rich side information. In this work, we propose a framework that combines collaborative-filtering (CF) signals derived from a shallow autoencoder (SAE) with a Thompson Sampling-based multi-armed bandit to dynamically select small batches of recipients for each email template. We show SAEs help balance exploration and exploitation by quantifying recipient informativeness and confidence, enabling efficient personalization without retraining during active learning. To facilitate reproducibility and future research, we release a large dataset of almost 15 million recipient-message interactions, offering new insights into email outreach dynamics for CF. Our experiments show that our method outperforms multiple baselines in retrieval metrics while retaining interpretable model components.

Model Embedding CTR Size per 100M (Compression) Dimension Lift Embeddings SBERT [18] 512 (baseline) 204.8 GB Nomic [14] 768 +4.86% 307.2 GB Nomic (Matryoshka) 64 +1.89% 25.6 GB Nomic (CompresSAE) 4096* +3.44% 25.6 GB *Sparse embeddings with 32 nonzero entries.Figure 1: Comparison of embedding models used for candidate retrieval.We report online recommendation performance on a downstream task, relative to SBERT [18], with anytime-valid 99% confidence intervals.

Large language models (LLMs) are sophisticated artificial intelligence systems designed to process and understand natural language at a complex level. The recent progress of these models, culminating in chat-based LLMs, has democratized the accessibility of these sophisticated intelligent systems, showcasing how machine learning methods can help humans in daily tasks. This research addresses the growing interest in understanding the mechanisms of LLMs and in evaluating their alignment with human cognition. We introduce an innovative alignment assessment strategy in the realm of LLMs that diverges from traditional approaches, utilizing the odd-one-out triplet-based task to investigate the alignment of LLMs’ representations with human object concept mental organization. Our methodology, which incorporates image captioning and zero/few-shot learning accuracy scoring, is designed to evaluate language models’ ability to predict similarities and differences in object concepts. A comprehensive experimental evaluation was conducted, involving four captioning strategies, twenty-four LLMs across eight model families, and three scoring procedures, utilizing a significantly large dataset for enhanced understanding of LLM comprehensibility. Finally, our study explores the impact of object description comprehensiveness on model-human representation alignment and analyzes a subset of randomly selected triplets to assess how LLMs are able to represent different levels of human judgment patterns.

We propose the Burst-Induced Poisson Process (BPoP), a model designed to analyze time series data such as feeds or search queries. BPoP can distinguish between the slowly-varying regular activity of a stable audience and the bursty activity of a curious audience, often seen in viral threads. Our model consists of two hidden, interacting processes: a self-feeding process (SFP) that generates bursty behavior related to viral threads, and a non-homogeneous Poisson process (NHPP) with step function intensity that is influenced by the bursts from the SFP. The NHPP models the normal background behavior, driven solely by the overall popularity of the topic among the stable audience. Through extensive empirical work, we have demonstrated that our model fits and characterizes a large number of real datasets more effectively than state-of-the-art models. Most importantly, BPoP can quantify the stable audience of media channels over time, serving as a valuable indicator of their popularity.

We propose a robust recommender systems model which performs matrix completion and a ratings-wise uncertainty estimation jointly. Whilst the prediction module is purely based on an implicit low-rank assumption imposed via nuclear norm regularization, our loss function is augmented by an uncertainty estimation module which learns an anomaly score for each individual rating via a Graph Neural Network: data points deemed more anomalous by the GNN are downregulated in the loss function used to train the low-rank module. The whole model is trained in an end-to-end fashion, allowing the anomaly detection module to tap on the supervised information available in the form of ratings. Thus, our model’s predictors enjoy the favourable generalization properties that come with being chosen from small function space (i.e., low-rank matrices), whilst exhibiting the robustness to outliers and flexibility that comes with deep learning methods. Furthermore, the anomaly scores themselves contain valuable qualitative information. Experiments on various real-life datasets demonstrate that our model outperforms standard matrix completion and other baselines, confirming the usefulness of the anomaly detection module.

Recently, the RS research community has witnessed a surge in popularity for shallow autoencoder-based CF methods. Due to its straightforward implementation and high accuracy on item retrieval metrics, EASE is potentially the most prominent of these models. Despite its accuracy and simplicity, EASE cannot be employed in some real-world recommender system applications due to its inability to scale to huge interaction matrices. In this paper, we proposed ELSA, a scalable shallow autoencoder method for implicit feedback recommenders. ELSA is a scalable autoencoder in which the hidden layer is factorizable into a low-rank plus sparse structure, thereby drastically lowering memory consumption and computation time. We conducted a comprehensive offline experimental section that combined synthetic and several real-world datasets. We also validated our strategy in an online setting by comparing ELSA to baselines in a live recommender system using an A/B test. Experiments demonstrate that ELSA is scalable and has competitive performance. Finally, we demonstrate the explainability of ELSA by illustrating the recovered latent space.

In this paper, we introduce a non-stationary and context-free Multi-Armed Bandit (MAB) problem and a novel algorithm (which we refer to as BMAB) to solve it. The problem is context-free in the sense that no side information about users or items is needed. We work in a continuous-time setting where each timestamp corresponds to a visit by a user and a corresponding decision regarding recommendation. The main novelty is that we model the reward distribution as a consequence of variations in the intensity of the activity, and thereby we assist the exploration/exploitation dilemma by exploring the temporal dynamics of the audience. To achieve this, we assume that the recommendation procedure can be split into two different states: the loyal and the curious state. We identify the current state by modelling the events as a mixture of two Poisson processes, one for each of the possible states. We further assume that the loyal audience is associated with a single stationary reward distribution, but each bursty period comes with its own reward distribution. We test our algorithm and compare it to several baselines in two strands of experiments: synthetic data simulations and real-world datasets. The results demonstrate that BMAB achieves competitive results when compared to state-of-the-art methods.

With the advancement of information systems, means of communications are becoming cheaper, faster, and more available. Today, millions of people carrying smartphones or tablets are able to communicate practically any time and anywhere they want. They can access their e-mails, comment on weblogs, watch and post videos and photos (as well as comment on them), and make phone calls or text messages almost ubiquitously. Given this scenario, in this article, we tackle a fundamental aspect of this new era of communication: How the time intervals between communication events behave for different technologies and means of communications. Are there universal patterns for the Inter-Event Time Distribution (IED)? How do inter-event times behave differently among particular technologies? To answer these questions, we analyzed eight different datasets from real and modern communication data and found four well-defined patterns seen in all the eight datasets. Moreover, we propose the use of the Self-Feeding Process (SFP) to generate inter-event times between communications. The SFP is an extremely parsimonious point process that requires at most two parameters and is able to generate inter-event times with all the universal properties we observed in the data. We also show three potential applications of the SFP: as a framework to generate a synthetic dataset containing realistic communication events of any one of the analyzed means of communications, as a technique to detect anomalies, and as a building block for more specific models that aim to encompass the particularities seen in each of the analyzed systems.