مبحث علمي تم نشره بمجلة محكمة

Recommender Systems and User Experience: Machine Learning Algorithms In Processing The Semi-Structured Text Data Used In Recommender Systems

An Abstract

This research paper discusses the theoretical and technical problems generated when creating, applying, and implementing the algorithms used in processing semi-structured text data streamed in the recommender systems. Transforming paper-based information carriers into digital ones opens up prospects for an ever-increasing volume of information, which can extract knowledge from a weakly structured data set. Despite the relatively short life of active use of the digital media industry, the amount of data flowing into channels of communication is growing exponentially, representing a transformational consequence of data converted into digital form. In 2016, for example, the accumulated amount of data was estimated at 16 zettabytes. According to an analytical study conducted by IDC, it is expected that volume of digital data will reach 163 zettabytes by 2025. This paves the way for the emergence of recommender systems, which identify the preferences that a user tends to during visiting a particular website. Many countries have carried out a comprehensive digital transformation that covered their economic and social sphere, aspiring to implement those systems into local search engines allocated to projects related to indexing non-traditional data. This, in turn, requires addressing the challenges and flaws, which impede the access to all the systems features.

Keywords Recommender systems · User experience · Collaborative filtering · Algorithm methods · Classification problems

An Introduction

In light of the sharp increase in the amount of data, determining the relevance and permanence of information has come to the fore (Xiao and Benbasat 2007; Resnick and Varian 1997). The concept of "relevance" means "the potential to create a kind of correspondence between the received information and the information request" (Pommeranz et al. 2012). Thus, relevance is determined entirely by mathematical models used in a particular information retrieval system. The term "permanence" is referring to "the correspondence between the received information and the information needs", i.e., permanence is considered the conformity between the documents found by the information retrieval system and the information needs identified by the user (McNee et al. 2002; Cena et al. 2010; Gena et al. 2011), regardless of how fully and accurately each need is expressed.

One of the effective approaches, which can increase the pertinence of information retrieval, is the machine learning methods (Burke 2002) that represent a special case of applying intelligent data analysis. This approach is considered a distinctive path (Adomavicius and Tuzhilin 2005) that lies at the intersection of many disciplines: mathematics, informatics, statistics, probability theory, and other different controlling regulations. A singular subclass of information systems is distinguished, based on mathematical models, which help solve problems related to determining the relevance and pertinence of data in a robust process called recommendation systems. More definitely, the recommender system is a software tool (Pu and Chen 2010; Pu et al. 2012), which establishes expectations regarding the user behavior inferred from the object of information retrieval system (Venkatesh et al. 2003). The common data sources for any recommender system are the user profiles, as they collect data about the user himself (Hassenzahl 2008), through his personal traits and a list of actions that he performed during working within the system. The methods and algorithms used in recommender systems aim at increasing the accuracy and stability of machine learning approaches and techniques allocated for preprocessing and classifying large volumes of semi-structured text data so that the pertinence of information retrieval remains at the same level of consistency and sustainability (Koufaris 2003; Hsu and Lu 2004; Yu et al. 2005).