Recommending Rating Values on Reviews for Designers

Recommending Rating Values on Reviews for Designers

Jian Jin, Ping Ji, Ying Liu
Copyright: © 2014 |Pages: 12
DOI: 10.4018/978-1-4666-5202-6.ch180
(Individual Chapters)
No Current Special Offers

Chapter Preview



Nowadays, e-commerce Web sites host a large amount of online reviews. Rich information is provided in online reviews. Designers benefit from this information to gain customer requirements. However, there are a large number of online reviews. It makes designers difficult to digest reviews efficiently. In the past few years, many researchers in computer science study how to analyze online reviews. This is known as sentiment analysis. However, state-of-art algorithms fail to point out how to use these efforts in improving product design. Moreover, before discussing about how to utilize online reviews, the quality of online reviews should be concerned since the quality of these reviews is often inversely proportional to the size of its membership (Otterbacher, 2009). From the viewpoint of different product designers, without an evaluation guideline, a review is highly possible to be rated as a useless one according to his or her design requirements, although it is recommended by many other designers. For example, one review complains about the weight and size problem of iPad and some reasons are provided. It might highly be possible to be recommended. However, this one might be also deemed as useless by battery designers because they only care about reviews about battery problems. To a particular review, designers may have different understandings from own perspectives. How to recommend rating values of online reviews from a personal perspective of a product designer is an interesting question, and this is the major focus of this chapter.

In our latest research work, the helpfulness of online reviews was defined, evaluated and predicted (Liu et al., 2013). An exploratory study was conducted for a better understanding of consumers. In this exploratory study, the helpfulness of a particular review was evaluated by whether it is helpful for product design. Two questionnaires were then distributed to understand about why some reviews were rated as so. The prediction of helpfulness of online reviews was modeled as a regression problem. Also, one conclusion in this research is that the helpfulness of online reviews, from the perspective of general product designers, can be evaluated by domain-independent features only, without a significant loss if product features are neglected. Based on this work (Liu et al., 2013), this chapter concentrates on how to recommend rating values of online reviews for a product designer. It is a different research problem, compared with the previous one. The motivation of this chapter is to find high quality online reviews which meet the design requirements of different designers. Interesting phenomena from the exploratory study contribute to model the problem.

The contributions of this chapter are at least three folds. Firstly, a recommendation method for rating values of online reviews from a personal perspective of a product designer is suggested. The recommendation method takes account of the requirements of different product designers, rather than regarding the rating values of online reviews as the same single numerical value. Secondly, the recommendation is developed based on both a generic aspect and a personal one. Three categories of domain-independent features are utilized to model the helpfulness of online reviews in the generic aspect, while product features are utilized to model the preference in the personal aspect. Thirdly, 1,000 phone reviews were selected from and the proposed method is employed to recommend the rating values of online reviews. The effectiveness of the method is evaluated by comparing the real evaluations from designers, which demonstrates the possibility to recommend rating values of online reviews from a personal perspective of a product designer automatically.

Key Terms in this Chapter

Support Vector Machine: One type of supervised learning model which is used for classification and regression analysis to analyze data and recognize patterns.

Nearest Neighborhood Method: A non-parametric method for classifying objects based on closest training examples.

Product Design: The process of creating a new product to be sold by a business to its customers.

Supervised Learning: One type of a machine learning task which intends to infer a function from labeled training data.

Online Reviews: Reviews written online by consumers who have experience to comment on the product or service delivers on its promises.

Review Recommendation: The activity that recommends reviews a user might be interested in.

Opinion Spam: The activities that try to mislead readers by giving undeserving opinions to some product or service in order to promote it or damage their reputations.

Similarity Learning: One type of a supervised machine learning task which intends to learn how similar or related two objects are.

Review Mining: The process of analyzing reviews and summarizing them into useful information that can be used to increase revenue, cuts costs, or both.

Complete Chapter List

Search this Book: