In big data era, massive heterogeneous data are generated from various data sources, the cleaning of dirty data is critical for reliable data analysis. Existing rule-based methods are generally developed in single data source environment, issues like data standardisation and duplication detection for different data type attributes, are not fully studied. In order to address these challenges, we introduce a method based on dynamic configurable rules which can integrate data detection, modification and transformation together. Secondly, we propose a type-based blocking and a varying window size selection mechanism based on classic sorted-neighbourhood algorithm. We present a reference implementation of our method in a real-life data fusion system and validate its effectiveness and efficiency using recall and precision metrics. Experimental results indicate that our method is suitable in the scenario of multiple data sources with heterogeneous attribute properties.