Searcharxiv⌕ Search

arXiv subjects

Olga Goussevskaia

Publications and source records attributed to Olga Goussevskaia.

9 recordsLinked to original sources

Learning Cross-Domain Media Similarity from Amazon Reviews

Collaborative filtering is a basic foundation of many recommendation algorithms. It works by analyzing a user's history and then suggesting items similar to those in that history. This method typically relies on user data to compute item similarity, data that is widely available to commercial systems but is usually sparse or inaccessible for independent systems and research, especially across multiple media domains. In this work, we propose a method to generalize similarity information between items using supervised learning. As input to our model, we use semantic and temporal features extracted from Amazon product reviews, which are publicly available. As the training target, we use item (product) co-occurrence in the reviewing history of Amazon customers. We evaluate the quality of the predicted similarities across two media domains: books and movies. Experimental results demonstrate the robustness of our methodology, yielding positive outcomes in zero-shot cross-domain transfer tasks. Furthermore, feature attribution analysis reveals a structural shift in how feature groups are utilized across setups: intra-domain models rely heavily on linguistic semantics, whereas cross-domain environments pivot toward behavioral popularity cues. Incorporating even limited target-domain context during training bridges this domain sparsity gap, significantly enhancing framework generalization.

cs.IR↗

Music Similarity Learning in Sparse Latin Scenarios

In this work we propose a lightweight, interpretable approach that estimates item-item music similarity from publicly available metadata. Our hybrid regression model combines content-based filtering with the classic item-based collaborative filtering. We test generalization via a zero-shot evaluation on Latin American music, a genre for which co-occurrence data is scarce. Accurately predicting similarity for these unseen, sparse pairs shows that our approach is robust to the regional gap affecting many existing systems, without the audio signal processing or the computational budget that deep multimodal approaches would require.

cs.IR↗

Opportunistic Optical Link Scheduling in the Scalable Matching Model

The Scalable Matching Model (SMM) is a useful abstraction for reconfigurable datacenter networks. In SMM, a set of $n$ (Top-of-Rack) electrical packet switches is connected via a set of optical circuit switches (OCS), each realizing a $p \times p$ matching, where $p$ is the number of ports per switch. In the case when $p \ll n$, bounded-degree networks utilize only a small portion of the OCS infrastructure at each point in time, especially when traffic patterns are skewed and bursty. In this work, we propose the Opportunistic Optical Link Scheduling (OOLS), a set of lightweight OCS link scheduling routines, designed to augment bounded-degree reconfigurable network topologies by leveraging idle OCS resources. We evaluate the performance of OOLS in the SMM through simulations, using synthetic and real-world data-center workloads. Our simulation results show that OOLS substantially improves performance, lowering communication costs and boosting throughput.

cs.NI↗

Predicting Engagement with Sponsored Content Across Account Types on Instagram

In this work we explore how users interact with sponsored content on social media platforms by collecting and analyzing a large-scale dataset of sponsored Instagram posts. To maintain transparency, we favor robust statistical analysis and explainable models over deep learning techniques. Our pipeline categorizes Instagram accounts along multiple dimensions, including audience size and entity type. We complement this with semantic features extracted from post captions and hashtags, and use these elements to train regression models that forecast engagement. We validate our approach on a dataset comprising over 15M Instagram posts authored by over 700K accounts featuring sponsored content. Our analysis shows that per-post engagement is partly predictable and highly optimized when accounts are segmented by entity type or audience tier. Our models achieve competitive predictive power, while remaining fully transparent about which features drive engagement.

cs.SI↗

A Learning-Augmented Overlay Network

This paper studies the integration of machine-learned advice in overlay networks in order to adapt their topology to the incoming demand. Such demand-aware systems have recently received much attention, for example in the context of data structures (Fu et al. in ICLR 2025, Zeynali et al. in ICML 2024). We in this paper extend this vision to overlay networks where requests are not to individual keys in a data structure but occur between communication pairs, and where algorithms have to be distributed. In this setting, we present an algorithm that adapts the topology (and the routing paths) of the overlay network to minimize the hop distance travelled by bit, that is, distance times demand. In a distributed manner, each node receives an (untrusted) prediction of the future demand to help him choose its set of neighbors and its forwarding table. This paper focuses on optimizing the well-known skip list networks (SLNs) for their simplicity. We start by introducing continuous skip list networks (C-SLNs) which are a generalization of SLNs specifically designed to tolerate predictive errors. We then present our learning-augmented algorithm, called LASLiN, and prove that its performance is (i) similar to the best possible SLN in case of good predictions ($O(1)$-consistency) and (ii) at most a logarithmic factor away from a standard overlay network in case of arbitrarily wrong predictions ($O(\log^2 n)$-robustness, where $n$ is the number of nodes in the network). Finally, we demonstrate the resilience of LASLiN against predictive errors (ie, its smoothness) using various error types on both synthetic and real demands.

cs.DC↗

MHCL: IPv6 Multihop Host Configuration for Low-Power Wireless Networks

Standard routing protocols for Low power and Lossy Networks are typically designed to optimize bottom-up data flows, by maintaining a cycle-free network topology. The advantage of such topologies is low memory footprint to store routing information (only the parent's address needs to me known by each node). The disadvantage is that other communication patterns, like top-down and bidirectional data flows, are not easily implemented. In this work we propose MHCL: IPv6 Multihop Host Configuration for Low-Power Wireless Networks. MHCL employs hierarchical address allocation that explores cycle-free network topologies and aims to enable top-down data communication with low message overhead and memory footprint. We evaluated the performance of MHCL both analytically and through simulations. We implemented MHCL as a subroutine of RPL protocol on Contiki OS and showed that it significantly improves top-down message delivery in RPL, while using a constant amount of memory (i.e., independent of network size) and being efficient in terms of setup time and number of control messages.

cs.NI↗

Mixtape Application: Music Map Methodology and Evaluation

This report discusses dimensionality reduction techniques used to create a music map - a map where the distances between songs represent their similarity and that can be used to recommend songs. We evaluate two techniques: Isomap and L-Isomap.

cs.SI↗

Mixtape Application: Last.fm Data Characterization

This report analyses data collected from Last.fm and used to create a real-time recommendation system. We collected over 2M songs and 1M tags and 372K user's listening habits. We characterize users' profiles: age, playcount, friends, gender and country. We characterized song, artist and tag popularity, genres of songs. Additionally we evaluated the co-occurrence of songs in users' histories, which can be used to compute similarity between songs.

cs.SI↗

Algorithms for Wireless Capacity

In this paper we address two basic questions in wireless communication: First, how long does it take to schedule an arbitrary set of communication requests? Second, given a set of communication requests, how many of them can be scheduled concurrently? Our results are derived in an interference model with geometric path loss and consist of efficient algorithms that find a constant approximation for the second problem and a logarithmic approximation for the first problem. In addition, we analyze some important properties of the interference model and show that it is robust to various factors that can influence the signal attenuation. More specifically, we prove that as long as such influences on the signal attenuation are constant, they affect the capacity only by a constant factor.

cs.NI↗