The main purpose of linked data is to realize the semantic web and extract knowledge through linking the data available on the web. One of the obstacles to achieving this goal is the existence of problems and errors in the published data, which causes incorrect links an
More
The main purpose of linked data is to realize the semantic web and extract knowledge through linking the data available on the web. One of the obstacles to achieving this goal is the existence of problems and errors in the published data, which causes incorrect links and as a result, invalid conclusions. Considering that the quality of the data has a direct effect on the success of the linked data project and the realization of the semantic web, it is better to evaluate the quality of each of the data sets in the early stages of publication. In this paper, a learning-based method for evaluating linked datasets is presented. For this purpose, first, the base quality model is selected and the quality features of the model are mapped to the field under study (which is the field of linked data in this article). Then, based on the mapping done, the important qualitative features in the study area are identified and described in detail by defining sub-features. In the third stage, based on past studies, the measurement metrics of each of the sub-features are extracted or defined. Then, measurement metrics should be implemented based on the type of data in the studied domain. In the next step, by selecting several data sets, the metric values are automatically calculated on the tested data sets. To use observational learning methods, it is necessary to evaluate the quality of data experimentally by experts. At this stage, the accuracy of each of the data sets is evaluated by experts, and based on the correlation study tests, the relationship between the quantitative values of the proposed metrics and the accuracy of the data is investigated. Then, by using learning methods, the effective metrics in the accuracy evaluation that have an acceptable predictability are identified. In the end, using learning methods, a quality prediction model based on the proposed criteria is presented. The results of the evaluations showed that the proposed method is scalable, efficient and applicable in addition to being automatic.
Manuscript profile
