ABSTRACT:
With the developing popularity of web services, an ever-increasing number of developers are making multiple services into mashups. Developers demonstrate an expanding enthusiasm for non-mainstream administrations (i.e., long-tail ones), notwithstanding, there are rare examinations attempting to address the long-tail web benefit proposal issue. The real difficulties for suggesting long-tail administrations accurately include serious sparsity of authentic use information and unsuitable nature of portrayal content.
In this project, we propose to fabricate a deep learning system to address these difficulties and perform precise long-tail suggestions. To handle the issue of the unsatisfactory nature of the portrayal content, we utilize stacked denoising autoencoders (SDAE) to perform highlight extraction. Additionally, we force the utilization records in hot administrations as a regularization of the encoding yield of SDAE, to give criticism content extraction. To address the sparsity of recorded utilization information, we take in the examples of designers’ inclination rather than modeling individual administrations. Our test results on a true dataset show that, with such joint autoencoder based feature representation and substance utilization learning system, the proposed calculation outflanks the cutting edge baselines altogether.
