Generative Data Transformation: From Mixed to Unified Data
#Generative Data Transformation #TASE #Recommendation Systems #Multi-domain Data #Contrastive Decoding #Data-centric Framework #Cold Start Problem #arXiv Research
📌 Key Takeaways
- Researchers developed TASE, a data-centric framework for recommendation systems
- The framework addresses data sparsity and cold start problems in multi-domain scenarios
- TASE uses contrastive decoding to encode cross-domain context into target sequences
- The approach outperforms existing model-centric solutions with better generalization
- Code for the framework is publicly available for further research
📖 Full Retelling
Researchers Jiaqing Zhang and eight collaborators from various institutions introduced TASE (Target-Aligned Sequential Regeneration), a novel data-centric framework for recommendation systems, in a paper submitted to arXiv on February 26, 2026, aiming to solve the persistent challenges of data sparsity and cold start problems that degrade model performance when integrating multiple data domains. The paper addresses a fundamental issue in artificial intelligence where recommendation model performance is intrinsically tied to the quality, volume, and relevance of training data. While recent research has attempted to enrich target domains with auxiliary data, inherent domain gaps can degrade mixed-domain data quality, leading to negative transfer and diminished performance. The researchers identified that existing 'model-centric' paradigms struggle to capture subtle, non-structural sequence dependencies across domains, resulting in poor generalization and high computational demands. To overcome these limitations, the team developed TASE, which employs a contrastive decoding mechanism to adaptively encode cross-domain context into target-domain sequences. This innovative approach enables standard models to learn intricate dependencies without requiring complex fusion architectures. According to their experimental results, TASE outperforms traditional model-centric solutions and demonstrates strong generalization capabilities across various sequential models. By generating enriched datasets, TASE effectively combines the strengths of both data-centric and model-centric approaches, representing a significant advancement in recommendation system design.
🏷️ Themes
Artificial Intelligence, Data Science, Recommendation Systems
📚 Related People & Topics
Tel Aviv Stock Exchange
Stock exchange located in Tel Aviv, Israel
The Tel Aviv Stock Exchange (TASE; Hebrew: הַבּוּרְסָה לִנְיָירוֹת עֵרֶךְ בְּתֵל אָבִיב), colloquially known as The Bursa, is the only public stock exchange in Israel and a public company itself, listed on its own exchange since August 1, 2019. It is regulated by the Securities Law (1968) and is und...
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
--> Computer Science > Artificial Intelligence arXiv:2602.22743 [Submitted on 26 Feb 2026] Title: Generative Data Transformation: From Mixed to Unified Data Authors: Jiaqing Zhang , Mingjia Yin , Hao Wang , Yuxin Tian , Yuyang Ye , Yawen Li , Wei Guo , Yong Liu , Enhong Chen View a PDF of the paper titled Generative Data Transformation: From Mixed to Unified Data, by Jiaqing Zhang and 8 other authors View PDF HTML Abstract: Recommendation model performance is intrinsically tied to the quality, volume, and relevance of their training data. To address common challenges like data sparsity and cold start, recent researchs have leveraged data from multiple auxiliary domains to enrich information within the target domain. However, inherent domain gaps can degrade the quality of mixed-domain data, leading to negative transfer and diminished model performance. Existing prevailing \emph{model-centric} paradigm -- which relies on complex, customized architectures -- struggles to capture the subtle, non-structural sequence dependencies across domains, leading to poor generalization and high demands on computational resources. To address these shortcomings, we propose \textsc , a \emph{data-centric} framework for \textbf arget-\textbf lign\textbf d \textbf equenti\textbf l \textbf egeneration, which employs a contrastive decoding mechanism to adaptively encode cross-domain context into target-domain sequences. It employs contrastive decoding to encode cross-domain context into target sequences, enabling standard models to learn intricate dependencies without complex fusion architectures. Experiments show \textsc outperforms model-centric solutions and generalizes to various sequential models. By generating enriched datasets, \textsc effectively combines the strengths of data- and model-centric paradigms. The code accompanying this paper is available at~ \textcolor this https URL }. Comments: Accepted by The Web Conference 2026 (WWW '26) Subjects: Artificial Intelligence (cs.AI)...
Read full article at source