Increase Robustness SDAE with Imputing Missing Value To Eliminate Users Sparse Data in Case E-Commerce Recommender System

Online shopping needs a computer machine to serve product information sale for customer or buyer candidate. Relevant information served by ecommerce system famous called recommender system. The successful to applied, it will have impact to increase of marketing target achievement. The character of information served by recommender system have to be special, personalized, relevant and fit according customer profiling. There are four kind of recommender system model, however there is one model that was successful to be applied in real ecommerce industry that popular named collaborative filtering. Collaborative filtering approach need a record users or customers activity in the past to generate recommendation for example rating record, purchasing record, testimony about product. The majority collaborative filtering approaches rely on rating as fundamental computation to calculate product recommendation. However, just a little number of consumers who willing give rating for products less than a percent, according to several convince datasets such MovieLens. This problem causes of sparse product rating that will impact to product recommendation accuracy level. Sometime, in extreme condition, it is impossible to generate product recommendation. Several efforts have been conducting to handle product sparse rating, however they fail to generate product recommendation accurately when face extreme sparse data, such as matrix factorization family include SVD, NMF, SVD++. This research aims to develop a model to handle users sparse rating involving deep SDAE. One of the efforts to produce better output in handling this data sparse, our strategy is to imputing missing value by statistical method so that the input in SDAE is closer to the feasibility of data that is not too sparse. According to our experiment involve deep learning, TensorFlow, MovieLens datasets, evaluation method by root mean square error (RMSE), our approach involves reducing input missing value could address users sparse rating and increase robustness over several existing approach.