A Multimodal Single-Branch Embedding Network for Recommendation in Cold-Start and Missing Modality Scenarios DOI Creative Commons
Christian Ganhör, Marta Moscati, Anna Hausberger

и другие.

Опубликована: Окт. 8, 2024

Most recommender systems adopt collaborative filtering (CF) and provide recommendations based on past collective interactions. Therefore, the performance of CF algorithms degrades when few or no interactions are available, a scenario referred to as cold-start. To address this issue, previous work relies models leveraging both data side information users items. Similar multimodal learning, these aim at combining content representations in shared embedding space. In we propose novel technique for recommendation, relying Single-Branch network Recommendation (SiBraR). Leveraging weight-sharing, SiBraR encodes interaction well using same single-branch different modalities. This makes effective scenarios missing modality, including cold start. Our extensive experiments large-scale recommendation datasets from three domains (music, movie, e-commerce) providing (audio, text, image, labels, interactions) show that significantly outperforms state-of-the-art content-based RSs cold-start scenarios, is competitive warm scenarios. We SiBraR's accurate modality model able map modalities region space, hence reducing gap.

Язык: Английский

Transformations in participation, creative and political practices in the punk scene in Bogotá during the COVID-19 pandemic DOI
Javier A. Rodríguez‐Camacho, Minerva Campion, Julián Jaramillo Arango

и другие.

Punk & Post Punk, Год журнала: 2024, Номер unknown

Опубликована: Июнь 17, 2024

This article studies the effects of COVID-19 pandemic through its different phases by looking at case Bogotá’s punk scene and analysing activities collectives involved in organization music shows other types gatherings. We contribute to understanding those answering question how has transformed creative, collaborative political practices participants. consider four periods analysis: pre-pandemic, lockdown, new normal post-peak reactivation. use a mixed methodology comprising interviews with members analysis secondary data obtained from social media posts made these collectives. latter compare audience participation reactions before, during after lockdown period 2019, 2020 2021. A network confirms existence identifiable as central node scene. find that remote were never considered more than temporary imperfect alternative economic perspectives. Economic impacts severe led closure several Though discourse was not articulated such scene, thought action aligned protests Colombia unified opposition market-oriented government policies cultural field. slow return in-person took place leaving events behind, re-establishing live centre allowing for informal socialization among creators, producers audiences. By three globally, we conclude did recover completely reach level it had before pandemic.

Язык: Английский

Процитировано

0
A Multimodal Single-Branch Embedding Network for Recommendation in Cold-Start and Missing Modality Scenarios DOI Creative Commons
Christian Ganhör, Marta Moscati, Anna Hausberger

и другие.

Опубликована: Окт. 8, 2024

Most recommender systems adopt collaborative filtering (CF) and provide recommendations based on past collective interactions. Therefore, the performance of CF algorithms degrades when few or no interactions are available, a scenario referred to as cold-start. To address this issue, previous work relies models leveraging both data side information users items. Similar multimodal learning, these aim at combining content representations in shared embedding space. In we propose novel technique for recommendation, relying Single-Branch network Recommendation (SiBraR). Leveraging weight-sharing, SiBraR encodes interaction well using same single-branch different modalities. This makes effective scenarios missing modality, including cold start. Our extensive experiments large-scale recommendation datasets from three domains (music, movie, e-commerce) providing (audio, text, image, labels, interactions) show that significantly outperforms state-of-the-art content-based RSs cold-start scenarios, is competitive warm scenarios. We SiBraR's accurate modality model able map modalities region space, hence reducing gap.

Язык: Английский

Процитировано

0