With the increasing prevalence of collaborative learning in digital environments, the accurate assessment of student interactions has become a fundamental challenge. Learning Analytics (LA), as a data-driven approach, offers unique potential for illuminating group processes by tracking learners' digital traces. However, the research literature in this domain is characterized by the fragmentation of models and heterogeneity of indicators. Aiming to create coherence in this field, this paper presents a qualitative meta-synthesis of online teamwork assessment models based on learning analytics. This study was conducted through a systematic search in reputable scientific databases; after screening 2790 records and assessing quality, 45 eligible studies were ultimately analyzed using thematic synthesis. Findings revealed that existing models fall into three main categories: instructor-centered (utilizing dashboards for pedagogical intervention), learner-centered (utilizing data for self-regulation), and conceptual frameworks (linking pedagogy with data analysis). Furthermore, a distinct evolutionary trend was observed, transitioning from product-oriented and superficial indicators (e.g., grades and post counts) toward deeper, process-oriented indicators (e.g., Social Network Analysis, Natural Language Processing, and multimodal data). The results highlight key challenges such as data interpretability, the need for user data literacy, and the misalignment of tools with authentic educational contexts. This meta-synthesis concludes that future research must focus on designing human-centered dashboards and integrating quantitative and qualitative analyses to bridge the gap between technical capabilities and pedagogical integration.