Sentiment aggregation of targeted features by capturing their dependencies: Making sense from customer reviews
Introduction
New product development (NPD) is a multi-step process that enables product designers to bring novel and innovative products to the market. Ideation is the first step of NPD in which product designers come up with ideas that introduce newness into a product (Cooper, 2008; Fojt, 1996; Koen et al., 2002; Koen, Bertels, & Kleinschmidt, 2014). One way for product designers to find ideas that introduce newness to a product is to search for the uncommon and unique features from products of other domains (woo Kang & Tucker, 2015; Moreno, Hernandez, Yang, & Otto, 2014; Tucker & Kang, 2012; Mirtalaie, Hussain, Chang, & Hussain, 2017a). In this paper, we term such novel ideas as targeted features. To diminish the risk of product failure, product designers, after determining the targeted features ascertain customers’ opinions toward them. This assists them to decide whether or not to consider a targeted feature any further (Geise, 2017; Grönlund, Sjödin, & Frishammar, 2010).
Due to the accelerated evolution in Web 2.0, customers’ opinions are easily accessible from Social Media (SM) portals. Feature-based sentiment analysis as a field of study is utilized to automate the extraction of people’s opinions and emotions from SM towards different features of a product (Hussein, 2018; Ireland & Liu, 2018; Jin, Ji, & Gu, 2016; Liu, 2012; Liu, Jiang, & Zhao, 2018; Rathan, Hulipalled, Venugopal, & Patnaik, 2018; Tuarob & Tucker, 2015; Tucker & Kim, 2011; Wang & Wang, 2014). After the features from customer reviews are extracted, they need to be aggregated for product designers to have a meaningful summary of what customers are saying towards a feature. Sentiment aggregation as a sub-field is used to address this task. This task also handles the mismatch cases when customers in their reviews refer to a feature with a different terminology as compared to how it is mentioned in a product’s standard terminology. Techniques such as topic-modelling (Lau, Li, & Liao, 2014; Suleman & Vechtomova, 2015; Ye, Li, Guo, Qian, & B, 2015) or ontology-based (Agarwal, Mittal, Bansal, & Garg, 2015; Mukherjee & Joshi, 2013, 2014; Tamilselvam, Nagar, Mishra, & Dey, 2017; Umamaheswari & Priya, 2016; woo Kang & Tucker, 2016) are used to reduce the sparseness between the features extracted from customer reviews and how they are mentioned according to the product’s standard terminology. This is done by mapping the extracted features from customer reviews to the high-level components of a product, available from a generic ontology or lexicon like ConceptNet and WordNet. While these approaches have no issues in matching the commonly known features of a product, they fail when customers in their reviews speak about a product’s technical features. This is because such technical features are not commonly available in the generic lexicon, and thus they remain unmatched (Ganesan, Zhai, & Viegas, 2012). Thus, an approach is needed to map the product’s technical features mentioned in SM reviews to a detailed product lexicon, so that they are utilized by the product designers (woo Kang & Tucker, 2016).
A product’s feature does not work in isolation. Instead, it not only collaborates with other features to achieve its purpose but also may influence different other features to deliver their output (Apel, Kolesnikov, Siegmund, Kästner, & Garvin, 2013). For example, a camera’s ‘advanced scene recognition system (ASRS)’ works collaboratively with features such as ‘flash’ and ‘exposure’ to accomplish its job. Furthermore, ASRS also influences other features, such as the camera’s ‘image’ in achieving its functionality. Such features are the dependent features of ASRS. Since such dependencies play a vital role in fulfilling a feature’s tasks (Ferber, Haag, & Savolainen, 2002), while computing the sentiment score of a targeted feature, we argue that the polarity of its dependent features too should also be considered. Although (Donaldson & Calder, 2012; Lienhard, Greevy, & Nierstrasz, 2007; Siegmund & Kolesnikov, 2012) detect features interactions/dependencies in various domains, to the best of our knowledge, they have not been considered while analysing the polarity of a targeted feature. Thus, the sentiment value which they determine for that targeted feature may not be accurate.
As (Geise, 2017; Grönlund et al., 2010) mention, product designers use customers’ opinions toward the targeted features to decide whether or not to consider them further. Thus, addressing the gaps mentioned above are important for product designers when they reach the stage of making informed decisions about which targeted features to consider beyond the ideation step of NPD. In this paper, we address these gaps, and the research question we solve is:
How can a product designer capture technical details about a feature mentioned in customer reviews and ascertain the polarity of a targeted feature by considering its dependencies?
We fill these gaps in this paper by proposing the Sentiment Aggregation framework for Targeted Features (SA-TF). SA-TF builds on the previous work of extracting product features from reviews along with their associated degree of sentiment (Mirtalaie et al., 2017) and contributes to the theory and practice of sentiment analysis from SM in three ways as follows:
- 1
It contributes to the theory by developing an approach that maps the extracted product features (both high-level and technical) to the most appropriate component of the product. This is achieved by generating a product tree from the product’s specification document and developing a heuristic approach to map each extracted (review) feature to its best representative node in a tree.
- 2
It contributes to the theory by developing an approach that utilises hierarchical relations of a feature (both high-level and technical ones) when determining its overall aggregated polarity. It also develops an approach that determines the sentiment of a targeted feature by capturing its dependencies with other features of the product.
- 3
It contributes to practice by assisting product designers in taking the full potential of customers’ opinions from SM. Accurately determining customer opinion will lead product designers to innovate in their products better and incorporate ideas which are supported by the customers. This will diminish the risk of product failure and maintain a competitive advantage against their competitors.
The paper is organized as follows. Section 2 presents a review of the literature and shows how the gap addressed in the paper has theoretical groundings regarding the use of SM in the literature. Section 3 presents the adopted research methodology and the data analysis done by SA-TF. Section 4 presents the technical details on the process of feature matching in SA-TF. Sections 5 and 6 details on the process of sentiment aggregation and dependency consideration, respectively. Section 7 explains and compares the results of SA-TF with other approaches. Section 8 summarises the main findings along with discussing the theoretical and practical implications of the study. Section 9 concludes the paper with a discussion on future work.
Section snippets
Literature review
As our contribution in this paper spans across both the management and technical side of using SM, in this section, we present a literature review of each. The management side of literature review demonstrates how our considered use of SM concurs with the philosophical foundations and theories of SM helping in decision-making behaviour. The technical side of literature review focusses on sentiment aggregation and highlights the existing gaps in making sense of technical knowledge mentioned by
Research approach and data selection
As noted in the literature review, no such technical approach exists to address the research question. So, in this paper, we develop new techniques that incorporate new concepts to achieve our aim. We adopt a scientific research approach to solve our research question as it concurs with its spirit of making something work Galliers (1992). Furthermore, as our developed model will be able to replicate the results on a different dataset, it satisfies the scenarios in which this methodology can be
Generating the source product tree
The PSD of a source product includes almost all the features of a product and is freely available on the Internet2 . PSD is in the form of a table where the first column presents the basic features of the product. Each basic feature has its own sub-features or attributes which are depicted in the second column of the table. The third column represents the value (if any) of each attribute. We
Sentiment aggregation phase
In this phase, the overall sentiment value of each tree’s node will be computed by aggregating a node’s polarity score with that of its parent and ancestor. Although the idea of considering hierarchical relationships in computing the polarity of a feature is not new, SA-TF looks at it from a different viewpoint. To explain this difference with an example, Fig. 8 presents the polarity tuple as well as the overall sentiment score for some of the nodes in the product tree. The existing approaches
Dependency consideration phase
This phase computes the polarity of a targeted feature by considering its dependencies. This process is illustrated in the next sub-sections.
Results
In Section 7.1, we demonstrate the effectiveness of SA-TF in mapping the review features to the product tree nodes generated from the PSD and compare the results with the product tree generated by ConceptNet (Agarwal et al., 2015; Mukherjee & Joshi, 2013). Section 7.2 demonstrates the benefits and applicability of our polarity aggregation approach when the product designer’s objective is on a product’s detailed features rather than its functionalities. In Section 7.3, we evaluate the accuracy
Main findings
Scholars in different domains have examined how social media (SM) has impacted humans both in a positive and negative way. For example, scholars such as Neiger et al. (2012); Alalwan, Rana, Dwivedi, and Algharabat (2017)); Elbanna, Bunker, Levine, and Sleigh (2019)) discuss how social media has improved for the better workings domains such as healthcare, marketing and energy management respectively. On the other hand, scholars such as Kizgin et al. (2019); Dhir, Yossatorn, Kaur, and Chen (2018))
Conclusion and future work
SA-TF proposed in this paper assists product designers in a stepwise manner to make sense of review features and link them to the targeted features of their interest. Furthermore, it also considers the dependencies of a targeted feature when determining its aggregated sentiment value. What sets SA-TF apart from existing approaches of sentiment aggregation is that it focusses on the detailed features rather than the high-level ones. Thus, it is more applicable in cases when product designers
Declaration of Competing Interest
Both authors declare they have no competing interests.
Acknowledgements
The first author acknowledges UNSW Canberra for their financial support in the form of scholarship, which has led to this work. Both authors acknowledge the support of Professor Farookh Hussain of UTS in providing students who have worked on developing the prototype of SA-TF discussed in this paper.
References (99)
- et al.
Social media in marketing: A review and analysis of the existing literature
Telematics and Informatics
(2017) - et al.
Feature-interaction detection based on feature-based specifications
Computer Networks
(2013) - et al.
The influence of information technology in new product development: Observations of an empirical study of the access of engineering design information
International Journal of Information Management
(1997) - et al.
Online social media fatigue and psychological wellbeing—A study of compulsive use, fear of missing out, fatigue, anxiety and depression
International Journal of Information Management
(2018) - et al.
Modular modelling of signalling pathways and their cross-talk
Theoretical Computer Science
(2012) - et al.
Emergency management in the changing world of social media: Framing the research agenda with the stakeholders through engaged scholarship
International Journal of Information Management
(2019) - et al.
A role for startups in unleashing the disruptive power of social media
International Journal of Information Management
(2016) - et al.
Polarization and acculturation in US Election 2016 outcomes – Can twitter analytics predict changes in voting preferences
Technological Forecasting and Social Change
(2019) - et al.
Consumers acceptance of artificially intelligent (AI) device use in service delivery
International Journal of Information Management
(2019) - et al.
We’re all connected: The power of the social media ecosystem
Business Horizons
(2011)
Social media competitive analysis and text mining: A case study in the pizza industry
International Journal of Information Management
A survey on sentiment analysis challenges
Journal of King Saud University - Engineering Sciences
Application of data analytics for product design: Sentiment analysis of online product reviews
CIRP Journal of Manufacturing Science and Technology
Social media mining for product planning: A product opportunity mining approach based on topic modeling and sentiment analysis
International Journal of Information Management
Identifying comparative customer requirements from product online reviews for competitor analysis
Engineering Applications of Artificial Intelligence
Exploring reviews and review sequences on e-commerce platform: A study of helpful reviews on Amazon.iN
Journal of Retailing and Consumer Services
Social analytics: Learning fuzzy product ontologies for aspect-oriented sentiment analysis
Decision Support Systems
Using contextual features and multi-view ensemble learning in product defect identification from online discussion forums
Decision Support Systems
Fundamental studies in design-by-analogy: A focus on domain- knowledge experts and applications to transactional design
Design Studies
Reputations at risk: Engagement during social media crises
Public Relations Review
Theories of social media: Philosophical foundations
Engineering
Pre- and post-launch emotions in new product development: Insights from twitter analytics of three products
International Journal of Information Management
Co-citation and cluster analyses of extant literature on social networks
International Journal of Information Management
Predicting the “helpfulness” of online consumer reviews
Journal of Business Research
The role of digital and social media marketing in consumer behavior
Current Opinion in Psychology
Social media analytics – Challenges in topic discovery, data collection, and data preparation
International Journal of Information Management
Does privacy assurance on social commerce sites matter to millennials?
International Journal of Information Management
Business intelligence in online customer textual reviews: Understanding consumer perceptions and influential factors
International Journal of Information Management
Exploiting user experience from online customer reviews for product design
International Journal of Information Management
Social media for intelligent public information and warning in disasters: An interdisciplinary review
International Journal of Information Management
The role of online product recommendations on customer decision making and loyalty in social shopping communities
International Journal of Information Management
Sentiment analysis using common-sense and context information
Computational Intelligence and Neuroscience
Exploring feature interactions in the wild
Proceedings of the 5th International Workshop on Feature-Oriented Software Development - FOSD’ 13
Social media as tool for facilitating knowledge creation and innovation in small and medium enterprises
Baltic Journal of Management
Sentiment prediction based on dempster-shafer theory of evidence
Mathematical Problems in Engineering
Sentiment analysis using cosine similarity measure
IEEE 2nd International Conference on Recent Trends in Information Systems (ReTIS)
Making sense of social media streams through semantics: A survey
Semantic Web
When friends recommend: Online purchasing behavior of Russian and Dutch people when prompted by recommendations from Facebook friends
Russian Journal of Communication
Increasing attributional certainty via social media: Learning about others one bit at a time
Journal of Computer-Mediated Communication
Inferring activities from social media data
Transportation Research Record
Perspective: The stage‐gate® idea‐to‐Launch process ‐ update, what’s new and NexGen systems
The Journal of Product Innovation Management
The science of influence: How social media affects decision making in the healthcare, travel, retail and financial Industries
Role of WhatsApp messenger in the laboratory management system: A boon to communication
Journal of Medical Systems
Using social media to engage students
Artificial Intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy
International Journal of Information Management
The influence of social media on the consumers’ hotel decision journey
Journal of Hospitality and Tourism Technology
Feature interaction and dependencies: Modeling features for reengineering a legacy product line
International Software Product Lines Conference (SPLC)
Learning theory and its application to the use of social media in medical education
Postgraduate Medical Journal
Social media as an information source for rapid flood inundation mapping
Natural Hazards and Earth System Sciences Discussions
Cited by (5)
Predicting Mobile Applications Adoption Based on Online Review Data and Integrated Independent Subspaces Method
2023, Journal of Modern InformationSupporting Product Development by a Trend Analysis Tool Applying Aspect-Based Sentiment Detection
2022, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)Intelligent method of a competitive product choosing based on the emotional feedbacks coloring
2021, CEUR Workshop Proceedings