Verification pipeline

The concept of "guilty until proven otherwise" in our system begins with the premise that all content starts with a trust value of 0.0, indicating a neutral, unverified state. From this starting point, each piece of content undergoes a series of checks by different verification subsystems designed to assess its authenticity. As the content passes through these checks, it accumulates positive or negative scores based on the results from each subsystem. These individual scores are then added together to form a total cumulative trust value for the entire review. A higher cumulative trust value indicates a higher degree of authenticity, showing that the content has been validated as genuine by multiple layers of verification. This method ensures that only content that consistently demonstrates its authenticity across various checks achieves a high trust score, embodying the principle of starting from a position of skepticism and requiring proof to establish credibility.

The initial step in the verification pipeline involves breaking down user inputs into their smallest parts, starting with each marked as "not yet checked." This detailed method lets us closely examine every bit of content to make sure nothing is missed. As we check each part, it moves from being "not yet checked" to being confirmed as authentic, gradually building up a review that is true and reliable.

Technological methods

  • Authenticity and uniqueness image recognition algorithms. Analysis of similarity with known images, comparison of metadata and other technical characteristics of photographs (geo, creation time, resolution).

  • Anomaly detection and text signatures. Identifying statistical deviations and unique stylistic features.

  • Cross-check. Detecting duplicates of content (text).

User Motivation and verification

  • Identity verification and Proof of interaction. Presence of a geotag from the location, photo with a digital signature confirming authenticity, Cross-account verification, and request for photos/receipts.

  • Limit on the number of reviews. Introducing limits to prevent spam.

  • Master questionnaire. Creating personalized surveys by place category. (For example, if we decided he is a bot)

Transparency and moderation

  • Publication of rules and criteria. Open information about moderation processes. (Instructions for the user on creating trustworthy content)

  • Informing users and feedback. Notification about the status of reviews and the possibility to discuss moderation decisions.

  • Manual review and discussion of reviews. A team of moderators for quality control and interaction with businesses.

  • Public grant program. A program to attract external researchers to improve methods of detecting fakes. (Invitation to "hack" our system)

Adaptation and update

  • Regular update of moderation methods. Adapting to changing trends in review manipulations.

Identity verification

  • Cross-verification. Phone, email, social networks, crypto wallet, KYC identity verification.

  • Biometric verification. Using biometric data to confirm a user's uniqueness. (clarify regarding Android)

  • Proof of Interaction with business. Requesting receipts or selfie photos, conducting transactions.

Technological measures

  • Geolocation verification. Checking the user's location, detecting suspicious changes in device location.

  • IP Address check and proxy/VPN analysis. Identifying the use of proxy servers and VPNs.

Behavior analysis

  • ML-capcha

    • Reaction Time Analysis: Studying the time intervals between actions.

    • Behavioral Patterns: Analyzing user actions on the platform.

  • Activity history. Analyzing the user's activity history.

  • Analysis of reviews and comments. Examining the content created, text stylometry analysis.

Technological methods

  • Photo and video verification. Digital signature of photos from the location, including the name of the establishment.

  • Uniqueness checks for locations. Algorithms for analyzing images and descriptions for duplications.

  • Cross-check with other services. Data verification of the location with information from reliable sources and mapping services.

Social and community-oriented methods

  • Verification through community. Inviting users for location verification, bonuses for the role of Inspector.

  • Limit on creating locations for new users. Requirement of a certain level of activity and trust to add locations.

Verification and identification methods

  • Digital Identification and account verification. Strengthening the requirements for verifying the identity of the place creator through multi-factor authentication.

Manual and moderation methods

  • Manual moderation. Creating a team of moderators to check dubious locations and complaints.

  • Public control and feedback. Implementing a system of feedback and complaints about locations from the community.

Last updated