Алгоритмы рейтинговых систем в онлайн-сообществах

Community rating algorithm

Reputation systems are programs or algorithms that allow users to rate each other in online communities in order to build trust through reputation. Some common uses of these systems can be found on E commerce websites such as eBay, Amazon. com, and Etsy as well as online advice communities such as Stack Exchange. These reputation systems represent a significant trend in "decision support for Internet mediated service provisions". With the popularity of online communities for shopping, advice, and exchange of other important information, reputation systems are becoming vitally important to the online experience. The idea of reputation systems is that even if the consumer can't physically try a product or service, or see the person providing information, that they can be confident in the outcome of the exchange through trust built by recommender systems. Rheingold says that these systems arose as a result of the need for Internet users to gain trust in the individuals they transact with online. The trait he notes in human groups is that social functions such as gossip "keeps us up to date on who to trust, who other people trust, who is important, and who decides who is important". Internet sites such as eBay and Amazon, he argues, seek to make use of this social trait and are "built around the contributions of millions of customers, enhanced by reputation systems that police the quality of the content and transactions exchanged through the site".

The emerging sharing economy increases the importance of trust in peer to peer marketplaces and services. Users can build up reputation and trust in individual systems but usually don't have the ability to carry those reputations to other systems. Rachel Botsman and Roo Rogers argue in their book What's Mine is Yours (2010), that "it is only a matter of time before there is some form of network that aggregates reputation capital across multiple forms of Collaborative Consumption". These systems, often referred to as reputation banks, try to give users a platform to manage their reputation capital across multiple systems.

The main function of reputation systems is to build a sense of trust among users of online communities. As with brick and mortar stores, trust and reputation can be built through customer feedback. Paul Resnick from the Association for Computing Machinery describes three properties that are necessary for reputation systems to operate effectively. It was originally aimed at email applications, but it was subsequently developed as a general architecture for a reputation based service, followed by an email specific part. However, the workhorse of email reputation remains with DNSxL's, which do not follow that protocol. Those specification don't say how to collect feedback —in fact, the granularity of email sending entities makes it impractical to collect feedback directly from recipients— but are only concerned with reputation query/response methods.

High reputation capital often confers benefits upon the holder. For example, a wide range of studies have found a positive correlation between seller rating and selling price on eBay, indicating that high reputation can help users obtain more money for their items. High product reviews on online marketplaces can also help drive higher sales volumes. Abstract reputation can be used as a kind of resource, to be traded away for short term gains or built up by investing effort. For example, a company with a good reputation may sell lower quality products for higher profit until their reputation falls, or they may sell higher quality products to increase their reputation. Some reputation systems go further, making it explicitly possible to spend reputation within the system to derive a benefit. For example, on the Stack Overflow community, reputation points can be spent on question "bounties" to incentivize other users to answer the question. Even without an explicit spending mechanism in place, reputation systems often make it easier for users to spend their reputation without harming it excessively. For example, a ridesharing company driver with a high ride acceptance score (a metric often used for driver reputation) may opt to be more selective about his or her clientele, decreasing the driver's acceptance score but improving his or her driving experience. With the explicit feedback provided by the service, drivers can carefully manage their selectivity to avoid being penalized too heavily.

Reputation systems are in general vulnerable to attacks, and many types of attacks are possible. As the reputation system tries to generate an accurate assessment based on various factors including but not limited to unpredictable user size and potential adversarial environments, the attacks and defense mechanisms play an important role in the reputation systems. Attack classification of reputation system is based on identifying which system components and design choices are the targets of attacks. While the defense mechanisms are concluded based on existing reputation systems.

The capability of the attacker is determined by several characteristics, e. g., the location of the attacker related to the system (insider attacker vs. outsider attacker). An insider is an entity who has legitimate access to the system and can participate according to the system specifications, while an outsider is any unauthorized entity in the system who may or may not be identifiable. As the outsider attack is much more similar to other attacks in a computer system environment, the insider attack gets more focus in the reputation system. Usually, there are some common assumptions: the attackers are motivated either by selfish or malicious intent and the attackers can either work alone or in coalitions.

Attacks against reputation systems are classified based on the goals and methods of the attacker. Self promoting Attack. The attacker falsely increases their own reputation. A typical example is the so called Sybil attack where an attacker subverts the reputation system by creating a large number of pseudonymous entities, and using them to gain a disproportionately large influence. A reputation system's vulnerability to a Sybil attack depends on how cheaply Sybils can be generated, the degree to which the reputation system accepts input from entities that do not have a chain of trust linking them to a trusted entity, and whether the reputation system treats all entities identically. Whitewashing Attack. The attacker uses some system vulnerability to update their reputation. This attack usually targets the reputation system's formulation that is used to calculate the reputation result. The whitewashing attack can be combined with other types of attacks to make each one more effective. Slandering Attack. The attacker reports false data to lower the reputation of the victim nodes. It can be achieved by a single attacker or a coalition of attackers. Orchestrated Attack. The attacker orchestrates their efforts and employs several of the above strategies. One famous example of an orchestrated attack is known as an oscillation attack. Denial of Service Attack. The attacker prevents the calculation and dissemination of reputation values in reputation systems by using Denial of Service method.