Article Preview
Top1. Introduction
With the increase in resources available on the Web, resource searching has become more important and challenging. In a traditional search system, a user first issues a query containing some keywords. Then, the search engine retrieves resources according to the match between the keywords in the query and the descriptions of the available resources. Although different users may have different personal interests and expect different resources to be identified, a traditional search system always returns the same results for users issuing the same query. Therefore, to retrieve resources that not only satisfy a user’s basic information needs but also satisfy the user’s interests, it is necessary to conduct a personalized search based on user profiles.
Many studies (Ghorab, Zhou, O’Connor, & Wade, 2013; Haiduc et al., 2013) have been conducted on using data that reflect a user’s interests to construct user profiles for personalized searching. These data include a user’s stated interests (Ma, Pant, & Sheng, 2007), browsing history (Matthijs & Radlinski, 2011), social annotations (Xu, Bao, Fei, Su, & Yu, 2008), and microblogging behavior (Younus, O’Riordan, & Pasi, 2014). However, few studies have investigated leveraging users’ reviews and ratings of reviews on e-commerce websites for personalized searching.
In recent years, websites such as Amazon1, Epinions2, and Ciao UK3 have become extremely popular. These websites allow users to post reviews of products and rate the helpfulness of other users’ reviews. For example, a user can choose whether a review is helpful or not on Amazon, and can assign a helpfulness rating (ranging from 1 to 5) to a review on Epinions. These data (reviews and ratings of reviews) provide rich information for identifying personal interests or concerns. Each review includes several features describing a product. For example, there may be features such as battery life, flash, and shutter speeds in a review of a camera. Intuitively, a user may be interested in (or be concerned about) a product feature mentioned in a review they rate as useful. It is also intuitive for a user to be disinterested in (or unconcerned about) a product feature that appears in a review they rate as useless. For instance, if a user writes a review on the Galaxy S5 stating that “The processor of the phone is very fast,” then a phone’s processor may be the feature with which the user is most concerned. If another user considers this to be a useful review (i.e., rates it with a high score), then a phone’s processor may also be the feature with which that user is most concerned. By contrast, if a user considers it to be a useless review (i.e., rates it with a low score), a phone’s processor may be a feature with which this user is unconcerned.
Based on the above discussion, we believe that users’ reviews and ratings of reviews on e-commerce websites reflect a user’s concerns. Therefore, they are very useful for user profiling, and can improve the precision of user profiles. Moghaddam, Jamali, & Ester (2012) used users’ reviews to achieve personalized review recommendations based on tensor factorization. However, existing studies do not make use of them for personalized searching.