Reputation Systems

Digital distribution has changed the rate of growth of several industries, including music, publishing and online retail. The Internet is now a vast repository of content and a conduit for several services, otherwise considered traditional business enterprises.

eCommerce, while growing in leaps and bounds, has enabled online retail in specific, which constitutes less than 10 per cent of American retail, perhaps even less with Indian retail. The precedent set by players such as Amazon.com, Rhapsody (for music), Google's Blogger and Wordpress (as a small example set) will reflect quickly within the Indian economy with one of the biggest consumer populations in the world.

Flexible eCommerce engines introduce the parody of choice wherein a consumer's attention span is considered expendable. To enable online consumers choose wisely, be it people involved in actually buying something or simply reading or contributing to online content, recommendation systems come in handy.

While content in modern cyberspace is in abundance, as Chris Anderson argues in his book The Long Tail , content would not be easily found without accompanying context. For example, on a Friday night, if I were looking for the latest Sufi album online, how do I decide which link/collection/playlist to pick? The same applies for when I am trying to understand the nature of the political debate around infrastructure policy, who should I read?

Recommendation systems help build content or source reputation online.

The simplest example of where a person's reputation comes in handy is in the financial sector, for loans, where a credit rating often decides a member's privileges and rates. The notion of reputation has been encoded in various systems. Distributed computing, for instance, houses information wherein users need to trust both the information and its source to make usage meaningful. Just as credit-ratings help a bank decide on how to treat a customer's account, reputation scores are established in order to assist in decision making, when it comes to using the services of an entity.

Collaborative exercise

Reputation systems are often collaborative in nature, where the comments of the crowd add or take away from an entity's reputation. Reputation scores need not be constant. They vary according to time and how well information goes down with users in the system. Typically, users are allowed to rate how much they like certain content such as music or articles on blogs, etc. Examples of such systems include commercially successful services such as recommender systems and collaborative filtering. Assigning trust extends to other areas of technology including Peer-to-Peer systems, establishing trust in online social communities, online auctions and markets. These ideas have also trickled down into everyday life.

In Search technologies, for example, Google's Page Rank and the general category of blog search engines are two products that use reputation systems. Page Rank uses the number of forward links that a page has to determine its reputation. Page Rank further assigns a value of zero to sites using advertising techniques as these sites are considered to be spamming the real-estate on the Web. Simply put, greater the value, greater the trust.

Generic blog search engines use the number of times an author's name occurs, Web sites that cite the author and full-string matches to determine the trust score. Reputation systems differentiate themselves from recommender systems in relying on community feedback to determine trust value. In eCommerce, examples are found within eBay and ePinions where feedback chains, tag clouds and social text-mining as also member opinions on products are used. Feedback auctions are popular and amount to what is essentially a Sybil Attack, which essentially involves one person creating multiple identities, using those identities to create a different rating each time, thereby skewing the trust value. One way to handle this comes from the same principle as social networking. Users are connected by a social network; each person is in the network because someone else accepted him, based on trust.

In ePinions, members can choose to trust or distrust each other making this a more closely monitored social networking model of establishing reputation. In Identity Management, as used in tools such as IDOlogy and Trufina, information is matched against government public databases, for credibility. This, of course, assumes that such public data is authentic. Within Social News sites such as Slashdot or Digg, we see bottom-up filtering of contributors and community votes which determine reputation. Multiple votes are filtered out using unique IP addresses. Wikis are the most popular example of reputation systems where experts with established credentials are asked to review material before it becomes a part of the information published.

Understanding each technology in terms of the trust metrics used is important to gauge how well the reputation established on a particular forum can be relied upon. The reach of blogs and online news postings is immense. These need to be built on reputation systems so that users' confidence in the source is intact. Likewise, customer satisfaction in online retail and other similar applications too can benefit from reputation-led mushrooming of communities.

The author is founder of PatNMarks, an Intellectual Property Consulting Firm and a faculty member in Computer Engineering.

COMMENT NOW