ARSR stands for "Association Rules and Semantic Relatedness", a recommender system that combines association rule mining and semantic relatedness to create recommendations for form fields. ARSR was evaluated with focus on recommending values for fields of metadata forms. The evaluation was performed with two sets of association rules (R1 and R2). R1 was primarily used to assess the recommendation performance. R2 served as an alternative and led to similar results. This dataset contains the raw data ("collected", JSON format), generated during the evaluation. It includes i.a. input (populated fields and target field), expected output, and the top 40 generated recommendations for each test combination. In addition, the processed data ("analysed", CSV format) is provided. It is based on the raw data and is used to calculate metrics and plot the results. The source code for ARSR and the evaluation is available on GitLab (gitlab.com). Note: To perform the analysis on the raw data (collected) yourself, make sure to follow the setup instruction in the evaluation repository first. More specifically, install dependencies and unzip "data/cedar/test-instances.zip" so that URI mappings can be accessed. Then follow the instructions provided by the README file in the raw data archives (e.g. collected-R1.zip).