Data preparation is an important step in log mining for finding significant patterns from large amounts of unstructured web log data. Data preparation is a key stage in the WUM process since it transforms raw log data into multiple user sessions, offering structured data for the pattern-finding phase. Over the past decade, the extent of server logs created by prominent internet sites has escalated to several terabytes to petabytes daily that causing issues of big data processing. This paper discussed an efficient algorithm SIARDIDC_HREF, i.e., a scalable session identification (IDC) algorithm considering robot detection and internal dummy connection, which applies referrer-oriented heuristic in the three stages of data pre-processing. Following numerous experiments on a group of nodes, the suggested SIARDIC_HREF algorithm proved to be efficient and scalable for large amounts of data. The SIARDIC_HREF algorithm has a 47.65% impact on the session count of robots’ requests and internal dummy connection requests, which is higher than the SIA_HREF algorithm. Furthermore, the speedup and size-up are examined to demonstrate the algorithm's scalability, which suggests the execution time reduces as the data nodes rises for sizable datasets. The size-up metric demonstrates that when the input data volume is increased by a factor of two, the suggested procedure's execution time does not rise proportionally for the same quantity of data nodes.