With the development of information and Internet technology, human society has stepped into an era of information overload. Owing to the overwhelming quantity of information, both information providers and information consumers are facing challenges: information providers want the information to be transferred to the target audience while information consumers need to find the information most relevant to their need. To bridge the gap, recommender systems have been designed and applied in a variety of applications to help making decisions on movies, music, news and even services and persons. In a Collaborative Working Environment, recommender systems are also needed to guide collaboration and allocate task efficiently. When people exchange information and resources, they leave traces in some way or other. For a typical Web-based Collaborative Working Environment, traces can be recorded which are mainly produced by collaborative activities or interactions. The modelled traces represent knowledge as well as experience concerning the interactive actions among users and resources. Such traces can be defined, modelled and exploited in return to offer a clue on a variety of deductions. Firstly they can indicate whether a user is active or not concerning interactions on a certain subject. Combining with users’ evaluation of the information and resources during interaction, we can further evaluate a user’s competency on each subject. This aids the decision for further collaboration because knowing the specialization of users helps to distribute tasks reasonably.This thesis focuses on implementing a recommender system by exploiting various collaborative traces in the group shared/collaborative workspace. To achieve this goal, firstly we collect traces and get them filtered by system filters. For evaluating shared resources we propose a system of vote and combine the result with collaborative traces. Furthermore, we present two mathematical approaches (TF-IDF and Bayes Classifier) with semantic meanings of traced resources and a machine learning method (Logistic Regression) with user profile to exploit traces, and then discuss comprehensive examples. As a practical experience we tested our prototype in the context of the E-MEMORAe collaborative platform. By comparing the results of experiments we assess the strengths and weaknesses of each of the three methods and in which scenario they perform better. Cases show that our exploitation framework and various methods can facilitate both personal and collaborative work and help decision-making.