In the 21st century, the development of complex talents has become a major direction in the training of human resources. Collaborative learning enriches the overall quality of learners and plays an important role in developing higher-order learning skills. Project-based learning is a student-centered instructional approach that prioritizes the development of collaboration, inquiry, and problem-solving skills among students [1]. By engaging them in small-group work on real-life projects, this approach fosters the cultivation of interpersonal skills and, more specifically, generates collaborative skills, thereby enhancing their interpersonal skills, collaboration, and communication skills ^[2][3][2,3]. In higher education, project-based learning empowers students to acquire a diverse range of knowledge and innovative skills essential for tackling future challenges and achieving success [4]. According to the consensus among scholars, the development of enhanced basic collaboration skills, such as effective communication of ideas, respect for others, and teamwork through social learning, is regarded as a crucial component of project-based learning [5]. In recent years, project-based collaborative learning is receiving attention from educational research institutes and educational administrators. The latest Educause Horizon Report (2022, 2023) highlighted the emerging significance of hybrid and collaborative learning as a crucial area of research within the realm of artificial intelligence ^[6][7][8][6,7,8]. The Chinese Government has put forward the “China’s Education Modernization 2035” initiative, aiming to foster first-class talents, with collaborative learning emerging as a vital approach to cultivate students’ multi-dimensional abilities [9].

Project-based learning (PBL) is an inquiry-based, holistic instructional approach rooted in authentic contexts. It represents a distinctive form of collaborative learning that places greater emphasis on student-centered engagement with tangible, real-world artifacts ^[10][11][10,11]. In this context, PBL has been extensively adopted in higher education, particularly in the field of engineering education for its authentic problem-solving skills training ^[11][12][13][11,12,13]. Physics, being a fundamental course within the engineering disciplines, plays a critical role in nurturing highly skilled professionals. Consequently, the development of students’ core proficiency in physics through PBL holds significant importance ^[14][15][14,15]. The effectiveness of project-based learning is commonly evaluated through a comprehensive assessment encompassing various dimensions, including cognitive, affective, behavioral, and artifact performance [16]. Instruments frequently employed to assess the effectiveness of project-based learning encompass a range of methodologies, such as self-report questionnaires, tests, interviews, observations, self-reporting, and artifact performance evaluations [17]. It is notable that the majority of project-based learning evaluations heavily depend on teacher judgment, even in the context of long-term activities. Evaluation procedures typically involve the utilization of appropriate scales, along with self-reports, reflective journals, and other relevant components [18]. There are, of course, studies that report on the use of computer-mediated construction of collaborative environments for project-based learning, but mainly as a mediating tool, such as an online peer assessment environment for conducting project-based learning [19], as a knowledge forum or as a blackboard tool to facilitate communication [20]. Rather, it is not designed to conduct real-life scenario-based project activities for intelligent assessment through computer-supported collaborative learning (CSCL). Therefore, computer-supported text mining of mutual assessment texts to assess the quality of collaboration provides a new perspective on the assessment of PBL.

Collaboration quality greatly influences the success of the collaboration, and various methods have been employed to examine collaboration quality. One such approach involves the utilization of a collaboration quality assessment tool that focuses on behavioral communication at both the individual and group levels [21]. The quality of team collaboration was assessed using the collaboration maturity model developed by Boughzala and De Vreede [22]. Furthermore, Jiang and Lou conducted a study on the quality of collaboration by incorporating participatory design into PBL activities [23]. Additionally, specific group moderation mechanisms have been identified as facilitators of student performance in PBL [24]. The assessment of collaboration quality often relies on the utilization of scales. Previous scholarly work has successfully employed such scales to measure collaboration quality in various domains, including medical teaching [25], mathematics [26], and cultural preservation participation [27]. However, it is important to note that this approach to measurement tends to be more subjective in nature.

Multimodal technology has emerged as a valuable approach for assessing the quality of collaboration. In particular, it enables the collection of various types of data, such as audio, logs, and eye movements, during face-to-face collaborative learning activities. These data, obtained through multimodal technology, facilitate effective evaluation of collaboration quality ^[28][29][28,29]. Som et al. employed a machine learning approach to evaluate the video and audio data of the collaboration process. They utilized the mixup data augmentation method as part of their analysis [30]. Chounta and Avouris conducted real-time assessments of short online collaboration activities by evaluating six dimensions of collaboration quality [31]. The primary focus of research has predominantly revolved around the utilization of technology for monitoring human collaboration. Nevertheless, it has been observed that this approach entails significant expenses, costs, and challenges in achieving widespread implementation.

Building upon this foundation, text mining techniques for collaborative learning have emerged as important tools for assessing the quality of collaboration, particularly due to the widespread adoption of CSCL and the availability of large-scale interaction data recording, which have provided the necessary conditions for their development [32]. Research on teaching and learning through text data mining conducted by Yang and An revealed that four methods, namely information extraction, text clustering, text classification, and topic modeling, are widely employed to address various educational problems [33]. Sentiment classification of texts by conditional random fields has also been used in the field of education [34]. According to Rosé et al., the integration of text classification in CSCL allows for more cost-effective analysis of the collaborative process. They further argue that monitoring tools can be utilized to some extent to assess the quality of manual coding [35].

Overall, the existing body of literature on PBL has primarily concentrated on higher-order thinking, self-regulated learning, and metacognition among students. However, comparatively less emphasis has been placed on examining the quality of collaboration within PBL contexts. To address this gap, leveraging online platforms to collect collaborative texts and employing text mining techniques for measuring collaboration quality emerges as a promising approach. This methodology offers several advantages, including enhanced objectivity, reduced operational costs, and suitability for large-scale implementation in evaluating collaborative quality within the context of PBL.

Group awareness encompasses the process through which individuals form a perception of the collective team and its overall situation. This concept entails an understanding or perception of the characteristics of a learning partner or collaborative group [36]. It has found extensive utilization in research, particularly in the domain of CSCL, with the aim of enhancing collaborative effectiveness ^[37][38][37,38]. Most scholars consider group awareness as a measure of individuals’ understanding of various aspects of the collaborative group and their perception of the information related to it in CSCL [39]. Group awareness encompasses a range of types and definitions. However, it can be broadly classified based on the generally accepted categorization into three main types: behavioral awareness, cognitive awareness, and social awareness ^[40][41][40,41]. Su et al. highlight the significance of emotional involvement as a dimension [42].

Behavioral awareness centers on the role, participation, and contribution of the group or peers in the ongoing project or activity. It involves perceiving and being aware of the tasks and work being carried out within the group. A significant body of research has been dedicated to analyzing learner data collected from online learning platforms. These data include metrics such as the number of comments, responses, likes, and other indicators that aim to capture learners’ collaborative engagement [43]. Additionally, visual graphs have been employed to explicitly label student contributions [44], fostering increased group collaboration efficiency.

Cognitive awareness pertains to the level of awareness among the group or peers regarding the knowledge acquired and constructed within the group, including knowledge relevant to the completed project. It is also regarded as a metacognitive process [45] that forms the foundation for self-regulated behavior [46]. Cognitive load theory has been employed to investigate cognitive awareness as well [47]. The measures used to assess cognitive awareness encompass various approaches. These measures include self-evaluation [48], gathering opinions and evaluations from group members ^[49][50][49,50], as well as examining the knowledge performance and information of collaborative members [51]. The dimensions of cognitive awareness involve perceptions of the current knowledge level within the collaborative group, understanding the relationships within the knowledge concept structure, and recognizing different viewpoints. Visualization tools utilized in cognitive awareness assessment include graphical representations, shared knowledge situations [52], textual visualizations such as word clouds and tags, as well as network concept maps that visualize associative relationships.

Social awareness encompasses the perception and understanding of how the group functions, including an understanding of the dynamics of interaction, the movements within the group, and the level of communication [53]. On the other hand, affective awareness involves sensing the emotional states of peers during interactions and is often derived from peer assessment data [54]. Successful collaboration relies on awareness of social and emotional awareness, as it significantly impacts the group climate and individuals’ willingness to participate [39]. The primary measures employed to assess social and emotional awareness include systematic interaction data, inter-rater data, and scale data [55]. Some scholars have delved into characterizing group perception data by comparing dimensions such as the frequency of interaction and the extent of engaged relationships [56].

Multiple scholars have confirmed the influence of group awareness information on collaborative learning, emphasizing its significance in the collaborative process ^[46][57][46,57]. Group awareness provides members with valuable insights into their cognitive, social, and other behaviors, enabling them to make positive behavioral adjustments [58]. Group awareness information has the potential to enhance learners’ self-regulation and promote increased individual contributions and peer interactions in collaborative learning [46]. Furthermore, group awareness can facilitate the establishment of connections among collaborative partners, improve knowledge sharing among group members [59], and promote more positive affective interactions [55].

Most scholars believe that the group awareness tool has a positive impact on students’ collaborative processes, group performance, and individual performance [60]. The utilization of group perception and peer assessment as a means to enhance self-perceived efficacy has been found to effectively facilitate collaborative activities [19]. On the contrary, the absence of perceived information about peers may impact the success of collaboration [61].

Recently, there has been a growing research interest in the integration of group awareness and peer assessment, with an increasing number of empirical studies focusing on collaborative learning through peer feedback and peer assessment ^[19][42][46][19,42,46]. However, there is less research on the relationship between group awareness data and collaborative quality, particularly in the context of PBL, and existing research is largely focused on online collaboration. In reality, PBL is a hands-on, face-to-face collaborative learning approach that places significant emphasis on offline practical skills development among students. Neglecting research on the group awareness quality of collaboration within PBL would be a disadvantage, and existing research has primarily concentrated on exploring group awareness through group awareness tools in the CSCL domain. At the same time, there is a scarcity of direct measurement of group awareness as it unfolds during collaboration [62]. Moreover, the conceptualization of group awareness as a theory has been studied with inconclusive evidence, lacking empirical study [63].