1 of 8

Milestone 2

A Decentralized �Autonomous Marketplace �(DAM) �

Gavriel Shaw

Filtering and Matching

Explore front-end marketplace architecture suitable for filtering and matching users with suitable AI app experiences, such as: sentiment analysis, data privacy and consent management, marketplace integrity (fraud prevention)

2 of 8

Conditions for Milestone 2

Filtering and matching on the DAM (decentralized autonomous marketplace) platform depends on a self-censorship approach to curating meaningful content experiences.
Initially, the thesis revolved around using traditional tags with sliders to modulate the intensity or frequency of topics.
However, this has evolved with the introduction of NLP and intelligent filtering, which allows users more seamless, dynamic control over what they see with minimal settings friction.

This milestone aims to establish a filtering and matching architecture that allows complete personalized content curation.
Through NLP and behavioral analysis, users can specify their preferences based on themes, emotional tones, behavior patterns, and content intensity without relying on manual tagging.

3 of 8

User Requirements (part 1)

Users require intelligent control over the content they encounter

Topic Visibility Control:

Users specify which topics or emotional tones they wish to avoid. For instance:

As a User, I want to avoid content related to specific topics (e.g., political debates, violence) or emotional tones (e.g., aggressive or pessimistic content), allowing me to manage what I see.

As a User, I want filtering sliders for manual adjustments to my content preferences.

This empowers users to proactively manage their exposure to unwanted content based on context, rather than relying on tags.

Behavioral-Based Filtering:

Filtering should also evolve based on user behavior:

As a User, I want to automatically see content based on patterns of what I tend to avoid or engage with.

By implementing behavior-based filtering, the platform adapts to user preferences automatically, minimizing the need for constant manual adjustment.

Topic Intensity Control:

The ability to control not just what content is seen, but how deeply or graphically it is presented:

As a User, I want to control how intense the discussion of certain topics can be (e.g., 0-10 scale for graphic content or in-depth political coverage).

Intensity sliders offer users refined control, allowing for more nuanced curation rather than simple exclusion.

4 of 8

User Requirements (part 2)

Users require intelligent control over the content they encounter

Content Categorization:

Users should have access to automated categorization, allowing for easier management of what themes appear in their feeds:

As a User, I want the system to automatically categorize content into high-level themes (e.g., news, entertainment) and subcategories (e.g., specific political issues or genres), giving me more granular control.

This ensures that users aren’t overwhelmed with manual adjustments, allowing them to filter broadly or with precision, depending on their needs.

Visual Media Filtering:

Given the rise of video and image-based content, visual filtering is key:

As a User, I want to manage how graphic content in videos or images is displayed, using sliders to reduce or block disturbing media.

This adds another layer of control for users over how they engage with content that is more visually impactful.

Filtering Reviews:

Transparency is essential for users to refine their settings:

As a User, I want feedback on how much content was filtered or adjusted in my feed (e.g., weekly or monthly reports) so I can improve my settings.

Providing users with this kind of feedback ensures they stay informed and can adjust their preferences as their needs evolve.

5 of 8

Challenge

Beyond traditional tag-based systems, create an intelligent, adaptive filtering and matching engine that dynamically understands user preferences

The system should:

Detect meaning, themes, and tone through NLP rather than relying on manual tags.
Build a behavior-driven filtering engine that automatically adjusts user exposure to content based on patterns of engagement or avoidance.
Ensure that users remain in full control of privacy and consent, so that their data and preferences are protected and only shared with authorized parties.

This is not just about blocking content, but about shaping the experience to maximize relevance and comfort for each user.

6 of 8

Strategy

Integrating Web2 and Web3 tools to create a hybrid system that balances personalized user control with decentralized autonomy. and an intelligent content filtering system

Behavior-Driven Filtering:

A behavioral filtering engine will adjust content exposure based on user interaction patterns, minimizing the need for manual adjustments.

Web2 Example: Amplitude or Google Analytics can track user engagement and dynamically refine content recommendations.

Web3 Example: Ocean Protocol can be leveraged for decentralized data management, ensuring users retain ownership of their behavior-driven preferences.

NLP for Content Understanding:

By integrating NLP tools (e.g., GPT models, BERT, SpaCy), we can ensure that the system interprets the meaning, tone, and context of content, providing richer, context-aware filtering without reliance on tags.

Web2 Example: Use OpenAI GPT models for sophisticated content understanding, enabling content filtering based on emotional tone or subject matter.

Web3 Example: Integrate SingularityNET to decentralize the NLP services, allowing users control over the algorithms that analyze their content.

Topic Intensity Control:

Users will have intensity sliders to adjust how much depth or sensitivity they allow for specific topics, from casual mentions to in-depth discussions.

Web2 Example: TensorFlow or PyTorch can power machine learning models that dynamically control content depth and sensitivity.

Web3 Example: Self-Sovereign Identity (SSI) solutions can ensure that users’ topic preferences are stored privately and securely, outside of centralized control.

7 of 8

Strategy

an intelligent content filtering system

Visual Media Filtering:

Image recognition and video analysis tools will filter out graphic content, allowing users to adjust exposure.

Web2 Example: Google Cloud Vision for real-time filtering of disturbing visuals.

Web3 Example: Kleros can handle disputes around visual media content in a decentralized arbitration system.

Automatic Content Categorization:

The system will automatically categorize content into themes, minimizing manual tagging.

Web2 Example: Amazon Comprehend for automatic categorization based on content analysis.

Web3 Example: The Graph for indexing decentralized content, allowing a more flexible categorization of Web3 content.

User Interface with Dynamic Controls:

A user-friendly interface with slider controls for topic intensity, frequency, and emotional tone ensures seamless interaction.

Web2 Example: Bubble or Webflow can build customizable user interfaces with dynamic controls.

Web3 Example: Superfluid (https://www.superfluid.finance/) could allow for real-time adjustments to user content preferences, with changes taking immediate effect across the platform, ensuring the user experience dynamically reflects these adjustments without lag.

8 of 8

General Plan

a seamless, dynamic system for managing their digital experience while aligning with the platform’s self-censorship thesis.

Integrate Web2 and Web3 Tools:

Utilize TensorFlow for content filtering intensity, alongside decentralized tools like Ocean Protocol and SSI to store user preferences securely.

NLP-Based Filtering Architecture:

Implement NLP-powered content understanding to detect themes, emotional tones, and content intensity, integrated with a behavior-driven filtering engine.

Visual and Video Filtering:

Implement Google Cloud Vision and Amazon Rekognition for visual content moderation, ensuring users can control their exposure to disturbing images and videos.

Privacy and Consent Management:

Build a privacy control dashboard that allows users to manage how their data is used, what preferences are applied, and ensure transparency in how behavior-driven filtering works.