Big data analytic untuk pembuatan rekomendasi koleksi film personal menggunakan Mlib. Apache Spark

Abstract

Introduction. The digital age is characterized by the explosion of digital information that creates problems in information retrieval. Search engines have a weakness in the keywords/queries that users can remember. Recommendations arise as solutions to provide personal information. Data Collection Method. In this paper, the researcher presented a recommendation engine model using dataset from movielends.org.Analysis Data. Alternating Least Square-Weight Regulation (ALS-WR) was used as a big data analytic algorithm in rating prediction and Cosine Similiarity as the second filter to bring items closer to the genre. Results and Discussions.The results of  Root Mean Squared Error (RMSE) from 100K datasets were 0.96 (validation) and 0.94 (test). The results RMSE from 1M dataset were 0.86 (validation) and 0.96 (test). The results  RMSE from 10M dataset were 0.81 (validation) and 0.81 (test). The result cosine similarity was 1 for 100% resemblance and it  decreased based on the similarity level. The user acceptance test was 28% user accepts the result of first recommendation, this value increased to 62% acceptance level of the user against the second recommendation. Conclusions. The final results show that 75% of respondents prefer the second recommendation  from two-stage filtering than just collaborative filtering.