Share on Facebook

We have witnessed the magnitude of its reach. Relationships among users of online communities, particularly those in e-commerce sites or in-depth product information sites that provide reviews and ratings, have the power to determine the success and recognition of such triple W prefixes in the global address book.    

But to translate that power down to actual numbers to increase virtual traffic and reap other commercial dividends is a challenge countless developers and corporations have struggled with. No one really knows who’s at the end of the internet line to which millions of strangers transact every day. One may own a site, establish security frameworks and resolve breaches; but beyond operator-specific initiatives, there have been no effective, quantifiable means to foster a sense of trust with users.

Lim Ee Peng, a professor of information systems at Singapore Management University’s (SMU) School of Information Systems, noted that the notion of inter-user trust is important, particularly for websites that rely on community content, like product reviews, for example. This is because users rely on information provided by other users. Can they trust the information if they do not trust its author – the larger community of users?

In the paper, “Predicting Trusts among Users of Online Communities”, Lim, together with co-authors, Jaideep Srivastava of the University of Minnesota, Young Ae Kim of the Korea Advaced Institute of Science and Technology, and Haifeng Liu, Hady W. Lauw, Minh-Tam Le and Aixin Sun of Nanyang Technological University, developed a taxonomy to quantify and predict inter-user trust.

Version 1.0

Other researchers have previously worked with an underlying assumption that users function within a web of trust – that users trust one another. However, in the real world, that’s not entirely true. Users, in general, pay little attention and do not know how to foster relationships in the online space effectively. Few users give reports on the credibility of other users, or see an active need to comment on the credibility of a review, report or rating score.

“Trust between a pair of users is an important aspect of decision making for Internet users, particularly for users of an online community where one user may rely on information provided by other users to make decisions. For example, a seller trusted by a buyer in an e-commerce website has a significant advantage against other sellers if the product quality cannot be verified in advance,” Lim wrote. It plays such a crucial role in online communities, but as the taxonomy has helped highlight, “it is often hard to assess the trustworthiness between two users without their self-reporting.”

The new taxonomy of trust

The first of its kind to predict the amount of trust between a pair of users via a classification approach, the new taxonomy gathers information from personal actions of two individuals (termed “user factors”) and their interactions (termed “interaction factors”) to ascertain the type and level of trust between them. It is able to assess the most impactful features of user behaviours that convey positive trust, and using the same factors, it can also quantify various degrees of trust within an entire online community.

While most researchers and practitioners would appreciate the ability to evaluate trust levels on their sites, many face difficulties in sourcing for the information that they need. What makes this taxonomy distinctive is its inferred trust, or a “supervised learning approach”, where it is not required for two website users to demonstrate a direct relationship - there is no need to ascertain a user’s reputation or his affinity with others, as other models have required. The authors also contended that the approach have been applied successfully in other online community studies.

Epinions, an online product review website, was the case in point. The website, open to anyone, is driven entirely by its large community of users who share reviews, comments, as well as evaluation scores. It is a popular site, because users leverage on a strong web of trust, through a varied taxonomy of interactions. For consistency, researchers examined product reviews within one category (Videos and DVDs) on the community product review site.        

Besides identifying primary drivers of that category, namely members, products, reviews, ratings for reviews, and comments, the researchers looked into various relationships these drivers could share on the site. These included competing reviews, competing ratings, as well as both competing and complementary comments. From these components, more relationships were identified and represented. User interactions that proved to be pivotal in the taxonomy include:

Connection: Two users who have established a connection by communicating about an object

Write-Rate (WR) Connection: Connection between the user who wrote a review and the other who rated it

Rate-Rate (RR) Connection: Connection between two users who put down their ratings for the same review

Write-Write (WW) Connection: Two users who are connected because they wrote reviews about the same product

Write-Comment (WC) Connection: The user who commented on a review shares a connection with the writer of the review

Comment-Comment (CC) Connection: Two users connected because they commented on the same product review, or connected because one commented on the review while the other added to that comment with his personal thoughts

User factors versus interaction factors

As user factors can determine the actions of another user and allow site developers to identify the amount of trust between two users, the researchers divided them into review-related, rating-related and comment-related categories. These three categories were divided into two groups: Distribution factors and count-based factors. Distribution factors were measured by statistical metrics while count-based factors related to specific sets of interactions.

Interaction factors, as the name suggests, comprise inter-user communication and influence. To determine if two users trusted each other, researchers looked at user interactions WR, WW, RR, WC and CC, divided these further (into localised user factors). They then put the components of the taxonomy through a five-fold cross validation. What they found was that the behaviours and contributions of users provide an reasonably accurate reflection of the amount of trust that was held. There were also stark differences between pairs of users who trusted each other and pairs who did not.  

While the taxonomy allowed the researchers to uncover large number of features of online interactions that may influence online decisions, the most significant came down to the Write-Rate (WR) connection – where one user writes a review and the other rates it. The study also showed the efficacy of interaction-based taxonomies, as they were proven to impact trust-based decisions more (as opposed to user-factors). By detailing behaviours, relationships and user actions, the taxonomy allows content administrators to accurately predict - in real numbers - how positively users view relationships on their websites.

Lim and his co-researchers expressed plans to extend the model to enhance website functionalities, in areas of identifying quality content in question-and-answer websites, as well as resolving link-prediction issues. “Although our approach is developed for online product review community, it is applicable to other online communities including e-commerce websites where sellers and buyers interact with one another,” they wrote, adding that more research is required in predicting the evolution of trust, as trust, in itself, changes over time.

Share on Facebook