Journal of Information Organization Renuka Mahajan, J S Sodhi 5 3 2015 http://www.dline.info/jio/fulltext/v5n3/v5n3_1.pdf An important application of web usage mining is mining web log data, where the sequences of web pages accessed by various web users, over a period of time, are recorded on the web server. We propose a new optimized technique in realm of an e-learning site that pre-processes the web log data to recommend the best links for a learner to visit next. We propose a novel methodology, by partitioning the database, on the basis of the learner’s knowledge level, to create a specialized suffix tree(s) from the existing sequences of previous ‘n’ learners’ path. Further to reduce the overhead of re-mining the web patterns from the whole web data, we propose that a web traversal pattern should be regarded significant, only if it qualifies the minimum threshold of length and frequency in the database. These significant patterns are added to generalize suffixes. These are then mined, using the most efficient mining algorithm after a comparative analysis of various algorithms, to find the most frequent navigation paths for recommendation to new learner. We conducted experiments in web log mining on a real case study of an Indian e-learning site. The proposed methodology is verified by experiments with promising results on computational time. This speed up obtained, in Web Pattern Mining, is a meaningful approach for building recommender based e-learning system, to predict the future learning paths.