کشف هوش کسب و کار از بررسی های آنلاین محصولات: چارچوب قاعده استقرایی (قیاسی)

Online product reviews are a major source of business intelligence (BI) that helps managers and marketers understand customers’ concerns and interests. The large volume of review data makes it difficult to manually analyze customers’ concerns. Automated tools have emerged to facilitate this analysis, however most lack the capability of extracting the relationships between the reviews’ rich expressions and the customer ratings. Managers and marketers often resort to manually read through voluminous reviews to find the relationships. To address these challenges, we propose the development of a new class of BI systems based on rough set theory, inductive rule learning, and information retrieval methods. We developed a new framework for designing BI systems that extract the relationship between the customer ratings and their reviews. Using reviews of different products from Amazon.com, we conducted both qualitative and quantitative experiments to evaluate the performance of a BI system developed based on the framework. The results indicate that the system achieved high accuracy and coverage related to rule quality, and produced interesting and informative rules with high support and confidence values. The findings have important implications for market sentiment analysis and e-commerce reputation management.

As e-commerce supports higher interactivity among users with Web 2.0 applications, user-generated content posted on these sites is growing significantly. Users not only consume Web content, but also produce massive data of their participation, often affecting other users’ decisions. A study finds that more than three-quarters of the 2078 users reported that online product reviews had a significant influence on their purchase decisions (comScore, 2007). These online product reviews contain descriptions about user preferences, comments, and recommendations that serve as a major source of business intelligence (BI), helping managers and marketers to better understand customers. Management scholar Peter Drucker emphasizes that “what is value to the customer” may be the most important question to answer in order to realize a business’s mission and purpose (Drucker, 2003). However, the large volume of online product review data creates significant information overload problems (Bowman, Danzig, Manber, & Schwartz, 1994), making it difficult to discover BI from the reviews and to analyze customer concerns. Two major pieces of information available in each online review are its textual content and the numerical rating, which respectively indicate the aspects of customer concerns and the customer sentiment. However, neither of these two alone provides the full account of a product’s real “value” (Drucker, 2003), which is the true explanation of the customer’s satisfaction. An important task of a manager is therefore to correlate between the numerical ratings and the textual content of the reviews in order to understand what the customer values in a product. This task is typically done by manually reading and extracting key phrases or words that indicate customer concerns and by manually relating between the extracted phrases and the numerical ratings. Despite its usefulness, such analysis is time-consuming and does not scale up to the rapidly growing online reviews. Automated tools and techniques have been proposed to analyze online reviews. These works try to study the reviews’ impact on sales (Zhu & Zhang, 2010), to recommend products (Aciar, Zhang, Simoff, & Debenham, 2007), to calculate the utility of the reviews (Ding & Liu, 2007), to identify key product features (Zhang, 2008), to detect false reviews (Jindal & Liu, 2007), and to summarize review content (Zhuang, Jing, & Zhu, 2006). However, research that supports the managerial task of correlating between the numerical ratings and textual content of the reviews is not widely found. The problem of how the reviews’ textual content contributes to the numerical ratings is thus not widely addressed. Understanding this correlation in large amounts of online review data could help e-commerce managers to make effective decisions on brand management, product promotion, and reputation management. In this paper, we discuss existing works on analyzing online product reviews and critically review these existing approaches. Following a design science paradigm (Hevner, March, Park, & Ram, 2004), we develop a new framework for designing a new class of BI systems that correlate the textual content and the numerical ratings of online product reviews. In contrast to behavioral science, the design science paradigm was chosen because it emphasizes on building and evaluating innovative artifacts that address the analysis needs of e-commerce managers, marketers, and BI practitioners. In the process of building our artifacts, we drew upon the theoretical and computational foundations of data mining (Liu, 2007 and Pawlak, 1982) and information retrieval (Salton, 1989; Salton, Wong, & Yang, 1975). Based on the rough set theory and inductive rule mining methods used in the framework, we developed as an instantiation a system for extracting the relationship between hundreds of customer ratings and their corresponding textual reviews posted on Amazon.com’s Web site. To demonstrate the applicability of the system, two data mining methods were implemented to extract automatically decision rules to guide the understanding of the relationship. Using quantitative and qualitative experiments, we empirically tested the system that was configured under different methods and settings. The system’s enhanced performance was demonstrated over different types of products’ online reviews. The results have strong implications for brand management and online market sentiment analysis.

A major source of business intelligence, online product reviews impact directly on company reputation and brand perception that can be translated to substantial changes in market share and profitability. Automated tools have emerged to support the analysis of these reviews, however most lack the capability of extracting the relationships between the reviews’ rich expressions and the customer ratings. Based on the theories and methods in inductive rule learning, rough set theory (RST), and information retrieval, we developed and validated a new framework for discovering business intelligence from online product reviews. A BI system was developed as an instantiation of the framework to induce automatically decision rules that relate the keywords occurring in online reviews and the customer ratings. Using the reviews of four products sold on Amazon.com, we experimented with the association rule mining (ARM) method, RST exhaustive algorithm, and RST LEM2 algorithm to study how they contribute differently to the quality of decision rules. The experimental results suggest that ARM algorithm achieved the best scalability and efficiency, and induced rules that have the highest level of support and highest confidence for the product with the largest number of reviews, while RST algorithms produced rules that are the most informative, most interesting, and have the highest confidence values. These IT artifacts provide new tools to managers and marketers to analyze their rapidly-growing online product reviews. The results have important implications for market sentiment analysis, online reputation management, and search engine optimization. Our evaluation methodology, which includes a new metric called “word-rating score,” provides new guidelines for future research to evaluate BI analytics. Future work can consider expanding the review datasets to cover different types of product, testing other methods for BI discovery, and studying reviews with varied distributions of ratings. Considering the rapid growth of e-commerce and the widespread use of online product reviews, companies that lack the capability of efficient and effective analysis of these reviews could lose significant competitive advantage.