A Feature Selection Based Hybrid Recommender System Using Systolic Tree For Pattern Mining

Today, the availability of huge amounts of data over the internet and the retrieval of necessary data, e-learning content, or a movie review can result in a condition of information overload. In order to achieve some positive feedback from the users, a recommendation has to be personalized. There are hybrid recommender systems that make use of content and are based on collaborative filtering, which generate certain recommendations to users. But such personalization can result in several data overload challenges, and the vector space matrix can be quite sparse. The frequent patterns will be able to overcome data overload problems and create new challenges with exponential combinations. The work has been investigated using a Systolic tree-based pattern mining for improving the recommender system. This paper presents a novel hybrid recommender framework that integrates systolic tree-based frequent pattern mining with TF-IDF feature extraction and optimization-driven feature selection (CFS, MI, and PSO) to address data sparsity and information overload in recommendation systems. Unlike existing hybrid methods that either rely solely on collaborative filtering or static content features, the proposed approach combines feature extraction, feature selection, and frequent pattern mining within a single pipeline to improve both accuracy and efficiency. The experiments use the MovieLens-1M dataset for evaluation, and the results show the effectiveness of the proposed technique. It is observed from the results that the PSO-Systolic Tree + CF achieved a PrecisionN2 of 0.84 (vs. 0.74 for the baseline) and recallN12 of 0.81 (vs. 0.44 for the baseline), reflecting significant gains in both accuracy and coverage. Furthermore, feature selection reduced dimensionality and computation time, with CFS cutting training cost by ~28% compared to plain CF, while PSO provided the most accurate results at the expense of higher training time.