Entity attribute discovery and clustering from online reviews

The rapid increase of user-generated content (UGC) is a rich source for reputation management of entities, products, and services. Looking at online product reviews as a concrete example, in reviews, customers usually give opinions on multiple attributes of products, therefore the challenge is to automatically extract and cluster attributes that are mentioned. In this paper, we investigate efficient attribute extraction models using a semi-supervised approach. Specifically, we formulate the attribute extraction issue as a sequence labeling task and design a bootstrapped schema to train the extraction models by leveraging a small quantity of labeled reviews and a larger number of unlabeled reviews. In addition, we propose a clustering By committee (CBC) approach to cluster attributes according to their semantic similarity. Experimental results on real world datasets show that the proposed approach is effective.     

Multi-level knowledge discovery from rule bases

The discovery of multi-level knowledge is important to allow queries at and across different levels of abstraction. While there are some similarities between our research and that of others in this area, the work reported in this paper does not directly involve databases and is differently motivated. Our research is interested in taking data in the form of rule-bases and finding multi-level knowledge. This paper describes our motivation, our preferred technique for acquiring the initial knowledge known as Ripple-Down Rules, the use of Formal Concept Analysis to develop an abstraction hierarchy, and our application of these ideas to knowledge bases from the domain of chemical pathology. We also provide an example of how the approach can be applied to other prepositional knowledge bases and suggest that it can be used as an additional phase to many existing data mining approaches.     

Eliciting symptom-diagnosis knowledge from online medical Q&A

With the objective to automatically detect diseases from symptoms in free-text data, a methodology to extract symptom-diagnosis knowledge from online medical textual data in Q&A domain is proposed in this paper: (1) a term frequency-inverse document frequency and PRECISION method is adopted to retrieve symptom words from unstructured text; (2) a variable precision rough set based genetic algorithm is applied to reduce redundant symptom words, and a rough set based rule is utilized for adding discriminative symptom words assisting to discriminate diseases sharing similar symptoms; (3) by employing fuzzy linguistic variables to express the risk level of disease or severity level of symptoms, a knowledge base with fuzzy belief structure is generated. Using data extracted from a Chinese medical Q&A forum for training and testing, some classical gastrointestinal diseases serve as a case study to evaluate the efficiency of the proposed methodology. Subsequently performance comparisons are made between the proposed methodology and some other classifiers, such as the decision tree algorithms including ID3 and J45, and the Bayesian network classifier. The comparative results demonstrate that the proposed methodology outperforms the decision tree algorithms and the Bayesian network classifier.     

Streaming-media knowledge discovery

As the amount of streaming audio and video available to World Wide Web users grows, tools for analyzing and indexing this content will become increasingly important. Frequently, knowledge management applications and information portals synthesize unstructured text information from the Web, intranets and partner sites. Given this context, we crawl a statistically significant number of Web pages, detect those that contain streaming media links, crawl the media links to extract associated meta-data, then use the crawl data to build a resource list for Web media. We have used these crawl-data findings to build a media indexing application that uses content-based indexing methods     

Speech-acts based analysis for requirements discovery from online discussions

Online discussions about software applications and services that take place on web-based communication platforms represent an invaluable knowledge source for diverse software engineering tasks, including requirements elicitation. The amount of research work on developing effective tool-supported analysis methods is rapidly increasing, as part of the so called software analytics. Textual messages in App store reviews, tweets, online discussions taking place in mailing lists and user forums, are processed by combining natural language techniques to filter out irrelevant data; text mining and machine learning algorithms to classify messages into different categories, such as bug report and feature request.Our research objective is to exploit a linguistic technique based on speech-acts for the analysis of online discussions with the ultimate goal of discovering requirements-relevant information. In this paper, we present a revised and extended version of the speech-acts based analysis technique, which we previously presented at CAiSE 2017, together with a detailed experimental characterisation of its properties. Datasets used in the experimental evaluation are taken from a widely used open source software project (161120 textual comments), as well as from an industrial project in the home energy management domain. We make them available for experiment replication purposes. On these datasets, our approach is able to successfully classify messages into Feature/Enhancement and Other, with F-measure of 0.81 and 0.84 respectively. We also found evidence that there is an association between types of speech-acts and categories of issues, and that there is correlation between some of the speech-acts and issue priority, thus motivating further research on the exploitation of our speech-acts based analysis technique in semi-automated multi-criteria requirements prioritisation.     

Enhancing student knowledge acquisition from online learning conversations

This article reports a theory-driven experimental study that evaluates the effects of an annotation functionality on online social interaction and individual learning outcomes. The central hypothesis of this study is that directly addressing a part of a text by annotating it and then connecting each annotation with its related discussion can decrease coordinative interaction costs and result in a higher-quality discussion that favors greater gains in individual learning outcomes. To reach our objective, we carried out a theory-driven experimental study that compares two versions of an anchored discussion system: one with annotation functionality and one without it, both displaying the learning material side by side with its associated discussion in one window. Participants were 106 students enrolled in two sections of a blended-format course in health education. We assigned each section to a software condition. The examination of students’ online social interaction centered on a fine-grained content analysis of coordination and knowledge construction activities as well as sequential analysis of knowledge construction activities. The results indicate that annotation functionality decreased coordinative interaction costs and stimulated more elaborated discussions that favored greater gains in individual learning outcomes. Implications for research and practice are discussed.     

Finding users preferences from large-scale online reviews for personalized recommendation

Along with the growth of Internet and electronic commerce, online consumer reviews have become a prevalent and rich source of information for both consumers and merchants. Numerous reviews record massive consumers’ opinions on products or services, which offer valuable information about users’ preferences for various aspects of different entities. This paper proposes a novel approach to finding the user preferences from free-text online reviews, where a user-preference-based collaborative filtering approach, namely UPCF, is developed to discover important aspects to users, as well as to reflect users’ individual needs for different aspects for recommendation. Extensive experiments are conducted on the data from a real-world online review platform, with the results showing that the proposed approach outperforms other approaches in effectively predicting the overall ratings of entities to target users for personalized recommendations. It also demonstrates that the approach has an advantage in dealing with sparse data, and can provide the recommendation results with desirable understandability.     

一种以领域知识为中心的知识发现过程模型

针对知识发现在实际应用中的问题,提出了一种以领域知识为中心的知识发现过程模型,并将其形式化,描述了其动态语义。与已有的知识发现过程模型相比,此过程模型更能体现知识发现过程的本质特性,同时具有严格的形式化基础,为知识发现系统的设计和实际的知识发现应用提供了一个新的参考。     

Architecture for knowledge discovery and knowledge management

In this paper, we propose I-MIN model for knowledge discovery and knowledge management in evolving databases. The model splits the KDD process into three phases. The schema designed during the first phase, abstracts the generic mining requirements of the KDD process and provides a mapping between the generic KDD process and (user) specific KDD subprocesses. The generic process is executed periodically during the second phase and windows of condensed knowledge called knowledge concentrates are created. During the third phase, which corresponds to actual mining by the end users, specific KDD subprocesses are invoked to mine knowledge concentrates. The model provides a set of mining operators for the development of mining applications to discover and renew, preserve and reuse, and share knowledge for effective knowledge management. These operators can be invoked by either using a declarative query language or by writing applications.The architectural proposal emulates a DBMS like environment for the managers, administrators and end users in the organization. Knowledge management functions, like sharing and reuse of the discovered knowledge among the users and periodic updating of the discovered knowledge are supported. Complete documentation and control of all the KDD endeavors in an organization are facilitated by the I-MIN model. This helps in structuring and streamlining the KDD operations in an organization.     

Conspicuous display through photo sharing in online reviews: Evidence from an online travel platform

Compared to text, photos are more conspicuous and suited for self-presentation. Although other motivations, such as helping others, partly account for photo sharing in online reviews, we emphasize a model of social status in which a conspicuous display of distant tourism products is used as a signal of an individual's status. Drawing on self-enhancement theory, we test hypotheses concerning exhibitionistic motivation in photo posting using a novel dataset of group tourism products. An instrumental variable method is used to address the endogeneity issue. We find that the distance between departure and destination, an indicator of cost and time spent, has a positive effect on the inclination to share photos. But people of higher status refrain from the conspicuous display of their photos. Moreover, the distance effect on photo sharing is mitigated by status within a website, suggesting that people assign different marginal values to the conspicuous display of tourism products at the same distance. This study contributes to the understanding of photo-sharing behavior in online reviews and can help platform managers build strategies to enhance reach and engagement.     

Discovering business intelligence from online product reviews: A rule-induction framework

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.     

Sources and impacts of social influence from online anonymous user reviews

Consumers increasingly rely on online reviews to make purchase decisions. However, the process through which consumers are influenced by online reviews is not well understood. To fill the gap, we apply the social influence theory to theoretically explain and analyze this opinion change process. Specifically, we identify antecedents and consequences of two types of social influence from online user reviews: informational and value-expressive influence. From a survey conducted in a controlled laboratory environment, we found that perceived review quality positively impacts informational influence, while perceived review quality, consistency, and social presence jointly impact value-expressive influence. Interestingly, informational influence impacts both perceived decision quality and perceived usefulness of the website, while value-expressive influence only impacts perceived usefulness of the website. Surprisingly, consumer characteristics, including prior product expertise and self-monitoring, do not have significant effects on the formation of social influence online.     

Mining customer requirements from online reviews: A product improvement perspective

Big data commerce has become an e-commerce trend. Learning how to extract valuable and real time insights from big data to drive smarter and more profitable business decisions is a main task of big data commerce. Using online reviews as an example, manufacturers have come to value how to select helpful online reviews and what can be learned from online reviews for new product development. In this research, we first proposed an automatic filtering model to predict the helpfulness of online reviews from the perspective of the product designer. The KANO method, which is based on the classical conjoint analysis model, is then innovatively applied to analyze online reviews to develop appropriate product improvement strategies. Moreover, an empirical case study using the new method is conducted with the data we acquired from JD.com, one of the largest electronic marketplaces in China. The case study indicates the effectiveness and robustness of the proposed approach. Our research suggests that the combination of big data and classical management models can bring success for big data commerce.     

Using MineSet for knowledge discovery

MineSet aids knowledge discovery and supports decision making based on relational data. It uses visualization and data mining to arrive at interesting results. Providing diverse visualization tools lets users choose the most appropriate method for a given problem. The client-server architecture performs most of the computationally intensive tasks on a server, while the processed results return to the client for visualization. The paper discusses MineSet database visualization and data mining visualization     

Revisable knowledge discovery in databases

This article introduces the idea of using nonmonotonic inheritance networks for the storage and maintenance of knowledge discovered in data (revisable knowledge discovery in databases). While existing data mining strategies for knowledge discovery in databases typically involve initial structuring through the use of identification trees and the subsequent extraction of rules from these trees for use in rule-based expert systems, such strategies have difficulty in coping with additional information which may conflict with that already used for the automatic generation of rules. In the worst case, the entire automatic sequence may have to be repeated. If nonmonotonic inheritance networks are used instead of rules for storing knowledge discovered in databases, additional conflicting information can be inserted directly into such structures, thereby bypassing the need for recompilation. © 1996 John Wiley & Sons, Inc.     

Using online consumer reviews as a source for demographic recommendations: A case study using online travel reviews

Online consumer reviews play an important role in the decision to purchase services online, mainly due to the rich information source they provide to consumers in terms of evaluating “experience”-type products and services that can be booked using the Internet, with online travel services being a significant example. However, different types of travelers assess each quality indicator differently, depending on the type of travel they engage in, and not necessarily their cultural or age background (e.g. solo travelers, young couples with children etc.). In this study, we present architecture for a demographic recommendation system, based on a user-defined hierarchy of service quality indicator importance, and classification of traveler types. We use an algebraic approach to ascertain preferences from a large dataset that we obtained from the popular travel website Booking.com using a web crawler and compared with the customer-constructed preference matrix. Interestingly, the architecture of the evaluated recommendation system takes into account already defined demand characteristics of the hotels (such as the number of reviews of specific consumer types compared to the total number of reviews) in order to provide an example architecture for a recommendation system based on user-defined preference criteria.