Article Preview
TopIntroduction
The colossal growth in digital information and the number of visitors on the Internet has led to a complete transformation in e-business environment. Big data has revolutionized business analytics for better modelling of customer profiles, customer-relations and satisfaction. The big data movement seeks to glean intelligence from data and translate it into business advantage. But the pertinent upheaval of data has uncovered myriad challenges, the most important of them being searching for relevant data. Due to this, recommendation systems (RS) have started playing a major role in our lives. From e-commerce to online advertising, recommender systems are inescapable in the present online environment. A recommendation system recommends products and services to users by exploiting their interest patterns and purchase decisions (Chen, Yang, Zhou et al, 2018). RS are ubiquitous with applications like recommender systems for movies (Carrer-Neto et al., 2012), tourism (Lim et al., 2016), music (Bogdanov et al., 2013), research articles (Son & Kim, 2018) etc. Product recommendations majorly rely on two factors, user ratings and product descriptions. For example, collaborative filtering methods (CF) (Schafer et al., 2007) exploit user-rating trends to decipher user-user similarity patterns. Whereas, content-based filtering methods (CBF) (Pazzani & Billsus, 2007) manipulate product descriptions to recommend products similar to the ones liked by target user. Each of these approaches have their own advantages and disadvantages and are often used in coalescence for enhanced performance results.
Social Web 2.0 has led to an upsurge in user-generated content which has essentially changed the online ecosystem. The increasing popularity of sentiment-rich resources like online feedback systems and personal blogs has made electronic word of mouth even more powerful. In the present setting, user reviews have started playing a vital role in determining the usability of product. More recently research interests have shifted to the orchestration of experiential user information (extracted from social web) for improved recommendation quality. In order to determine whether the reviewer's attitude towards a particular product is positive, negative, or neutral, sentiment analysis can be used. Sentiment analysis is a text categorization technique that interprets and classifies people’s opinions, appraisals, emotions, and attitudes towards entities like products, services, organizations, events, topics and their attributes. (Liu, 2012). Sentiment analysis can be used as an augmentation to the current recommendation systems, to automate product evaluation for ministration of purchase decisions.
CBF is a widely used recommendation technique that highly depends on availability of detailed and dependable item descriptions. Several techniques like matrix factorization (MF), probability models, nearest neighbour methods and clustering etc. have been used to augment CBF for superior results. Although traditional CBF techniques are efficient and easy to implement, they still suffer from a number of drawbacks like low prediction accuracies, lack of dependable item descriptions and inability to capture complex user-item interactions (Nassar et al., 2020). Also, classic catalogue-based item descriptions are unable to uncover factors that manipulate user purchase judgements. This leads to the multi-criteria decision problem of determining and combining estimators for rating prediction to quantify user-adherence for a product. All these drawbacks have opened avenues for exploring multi-criteria recommender systems (Nassar et al., 2020), deep learning and soft-computing based RS (Fu et al., 2018) for performance enhancement.