A Solution Towards to Detract Cold Start in Recommender Systems Dealing with Singular Value Decomposition

Recommender system based on collaborative filtering (CF) suffers from two basic problems known as cold start and sparse data. Appling metric similarity criteria through matrix factorization is one of the ways to reduce challenge of cold start. However, matrix factorization extract characteristics of user vectors & items, to reduce accuracy of recommendations. Therefore, SSVD two-level matrix design was designed to refine features of users and items through NHUSM similarity criteria, which used PSS and URP similarity criteria to increase accuracy to enhance the final recommendations to users. In addition to compare with common recommendation methods, SSVD is evaluated on two real data sets, IMDB &STS. Experimental results depict that proposed SSVD algorithm performs better than traditional methods of User-CF, Items-CF, and SVD recommendation in terms of precision, recall, F1-measure. Our detection emphasizes and accentuate the importance of cold start in recommender system and provide with insights on proposed solutions and limitations, which contributes to the development.