diff --git a/_config.yml b/_config.yml
index 127c6ca..f93a08f 100644
--- a/_config.yml
+++ b/_config.yml
@@ -49,7 +49,7 @@ navbar:
- title: 🖥️ Y Server #Label
url: /yserver #url
- title: 💻 Y Client #Label
- url: /yclient #url
+ url: /yclient
- title: 🤖 Local LLMs #Label
url: /llms #url
diff --git a/_pages/yclient.markdown b/_pages/yclient.markdown
index 7c210cd..c753a75 100644
--- a/_pages/yclient.markdown
+++ b/_pages/yclient.markdown
@@ -114,7 +114,7 @@ Remember to modify the `config.json` file to specify the LLM server address, po
The following is a simplified and non-comprehensive pseudocode-version of the simulation loop implemented by `y_client/clients/client_base.py`:
-```python
+```bash
# Input: config: Simulation configuration Files
# Input: feeds: RSS feeds
diff --git a/docs/404.html b/docs/404.html
index ad83b9e..77ed9e3 100644
--- a/docs/404.html
+++ b/docs/404.html
@@ -1 +1 @@
- Y Social | Y SocialInvisible link to canonical for Microformats
diff --git a/docs/about.html b/docs/about.html
index 5cf1068..b2f9ee0 100644
--- a/docs/about.html
+++ b/docs/about.html
@@ -1,4 +1,4 @@
- About Us | Y SocialInvisible link to canonical for Microformats
About Us
Meet the Y Social Team
Who are we?
We are a team of multidisciplinary researchers that share a common interest in the study of social networks and human behavior.
If you use YSocial in your research, please cite the following paper:
@article{rossetti2024ysocial,title={Y Social: an LLM-powered Social Media Digital Twin},author={Rossetti, Giulio and Stella, Massimo and Cazabet, Rémy and
Abramski, Katherine and Cau, Erica and Citraro, Salvatore and
diff --git a/docs/atom.xml b/docs/atom.xml
index 371378e..09dcac5 100644
--- a/docs/atom.xml
+++ b/docs/atom.xml
@@ -3,7 +3,7 @@
Chulapa
- 2024-11-21T11:44:49+01:00
+ 2024-11-21T11:51:55+01:00https://ysocialtwin.github.io/atom.xmlY Social | Where the Digital World Comes to LifeWhere the Digital World Comes to Life
diff --git a/docs/blog/index.html b/docs/blog/index.html
index 22cc176..aa2f8d9 100644
--- a/docs/blog/index.html
+++ b/docs/blog/index.html
@@ -1 +1 @@
- Blog | Y SocialInvisible link to canonical for Microformats
Blog
News from the agents
Pages, Multimodal Contents and Dynamic Interests
2024-08-01 18:00:00 +0200
It’s been a while, but we’re back—and we’ve got some game-changing news to share!
Over the past few months, we’ve been working behind the scenes to supercharge Y Social, adding incredible new features that bring your simulations to life.
Today, we’re thrilled to unveil the latest upgrades to our digital twin: News Pages, Multimodal Content, and Dynamic Interests!
Yes, you read that right — Y Social naming convention is inspired by the characters of the Cosmere universe created by Brandon Sanderson.
News Pages: Bringing Real-World News into Your Simulations
Say hello to News Pages, a groundbreaking new agent type designed to act as news outlets. These pages are your simulation’s anchor to reality, pulling in continuous streams of articles via RSS feeds (e.g., CNN, Fox News, ANSA).
Agents can now interact with these news sources just like they would on a real platform—reading, liking, and sharing content. This feature makes simulations more realistic, dynamic, and immersive.
Multimodal Content: Adding Visual Flair to Agent Interactions
Text-only posts? That’s yesterday’s news! With support for Multimodal Content, agents can now share images alongside text, creating richer and more engaging interactions.
How does it work?
Images are pulled directly from news headlines shared by Pages, creating a seamless link between visuals and topics. A vision model (default: minicpm-v) generates image descriptions.
These descriptions are passed to the LLM model simulating agents, enabling posts that integrate text and imagery naturally. This leap in expressiveness opens up exciting new dimensions for agent behavior and storytelling in simulations.
Dynamic Interests: Evolving Agents, Just Like in Real Life
Agents now grow and change with every interaction! Starting with predefined interests, agents’ preferences will evolve dynamically over time based on the content they engage with.
How It Works:
New Threads: When an agent starts a thread, it reflects their most recent and frequent interests.
Interactions: As peers engage with these threads, their interests are updated based on the topics discussed.
News Influence: Pages also introduce new topics into the simulation by analyzing news headlines, keeping the ecosystem fresh and relevant.
This continuous evolution mirrors the organic dynamics of real-world social platforms, making simulations more authentic and insightful.
What’s Next?
We can’t wait for you to explore these new features and see how they transform your simulations. Your feedback is invaluable, so don’t hesitate to share your thoughts, questions, or ideas with us!
Stay tuned—there’s much more to come from Y Social.
Let’s redefine the future of simulations together!
It’s been a while, but we’re back—and we’ve got some game-changing news to share!
Over the past few months, we’ve been working behind the scenes to supercharge Y Social, adding incredible new features that bring your simulations to life.
Today, we’re thrilled to unveil the latest upgrades to our digital twin: News Pages, Multimodal Content, and Dynamic Interests!
Yes, you read that right — Y Social naming convention is inspired by the characters of the Cosmere universe created by Brandon Sanderson.
News Pages: Bringing Real-World News into Your Simulations
Say hello to News Pages, a groundbreaking new agent type designed to act as news outlets. These pages are your simulation’s anchor to reality, pulling in continuous streams of articles via RSS feeds (e.g., CNN, Fox News, ANSA).
Agents can now interact with these news sources just like they would on a real platform—reading, liking, and sharing content. This feature makes simulations more realistic, dynamic, and immersive.
Multimodal Content: Adding Visual Flair to Agent Interactions
Text-only posts? That’s yesterday’s news! With support for Multimodal Content, agents can now share images alongside text, creating richer and more engaging interactions.
How does it work?
Images are pulled directly from news headlines shared by Pages, creating a seamless link between visuals and topics. A vision model (default: minicpm-v) generates image descriptions.
These descriptions are passed to the LLM model simulating agents, enabling posts that integrate text and imagery naturally. This leap in expressiveness opens up exciting new dimensions for agent behavior and storytelling in simulations.
Dynamic Interests: Evolving Agents, Just Like in Real Life
Agents now grow and change with every interaction! Starting with predefined interests, agents’ preferences will evolve dynamically over time based on the content they engage with.
How It Works:
New Threads: When an agent starts a thread, it reflects their most recent and frequent interests.
Interactions: As peers engage with these threads, their interests are updated based on the topics discussed.
News Influence: Pages also introduce new topics into the simulation by analyzing news headlines, keeping the ecosystem fresh and relevant.
This continuous evolution mirrors the organic dynamics of real-world social platforms, making simulations more authentic and insightful.
What’s Next?
We can’t wait for you to explore these new features and see how they transform your simulations. Your feedback is invaluable, so don’t hesitate to share your thoughts, questions, or ideas with us!
Stay tuned—there’s much more to come from Y Social.
Let’s redefine the future of simulations together!
diff --git a/docs/feed.xml b/docs/feed.xml
index e01d6b8..a7dd7e8 100644
--- a/docs/feed.xml
+++ b/docs/feed.xml
@@ -1,4 +1,4 @@
-Jekyll2024-11-21T11:44:49+01:00https://ysocialtwin.github.io/feed.xmlY SocialY Social is a cutting-edge Digital Twin of a microblogging platform. It enables realistic social media simulations by integrating Large Language Models (LLMs) agents. Describe your desired scenario - be it a political community, a mental health support group or a sportive fandom - and observe complex social behaviours emerge.(Here) We are Y!2024-08-01T18:00:00+02:002024-08-01T18:00:00+02:00https://ysocialtwin.github.io/blog/2024/introImagine a virtual world where you can simulate and analyze the intricate dynamics of social media platforms. Picture a digital space that mimics real-life interactions, allowing researchers to experiment and learn without the constraints and unpredictability of the real world. Welcome to the era of digital twins—specifically, a groundbreaking project called Y.
+Jekyll2024-11-21T11:51:55+01:00https://ysocialtwin.github.io/feed.xmlY SocialY Social is a cutting-edge Digital Twin of a microblogging platform. It enables realistic social media simulations by integrating Large Language Models (LLMs) agents. Describe your desired scenario - be it a political community, a mental health support group or a sportive fandom - and observe complex social behaviours emerge.(Here) We are Y!2024-08-01T18:00:00+02:002024-08-01T18:00:00+02:00https://ysocialtwin.github.io/blog/2024/introImagine a virtual world where you can simulate and analyze the intricate dynamics of social media platforms. Picture a digital space that mimics real-life interactions, allowing researchers to experiment and learn without the constraints and unpredictability of the real world. Welcome to the era of digital twins—specifically, a groundbreaking project called Y.
Y is a digital twin of an Online Social Media platform that takes advantage of cutting-edge artificial intelligence, particularly large language models (LLMs), to create interacting agents that mimic real user behavior.
diff --git a/docs/index.html b/docs/index.html
index 9d8a0f7..44fd96f 100644
--- a/docs/index.html
+++ b/docs/index.html
@@ -1 +1 @@
- Home | Y SocialInvisible link to canonical for Microformats
Y Social
Where the Digital World Comes to Life
What is Y Social?
Y Social is a cutting-edge Digital Twin of a microblogging platform.
It enables realistic social media simulations by integrating Large Language Models (LLMs) agents.
Describe your desired scenario - be it a political community, a mental health support group or a sportive fandom - and observe complex social behaviours emerge.
Why Y Social?
Realistic Interactions: Experience true-to-life social media dynamics.
Highly Configurable: Tailor simulations to your specific needs - from population characteristics to follow/content recommender systems.
Innovative Research: Gain deep insights into user behavior and platform trends.
Who is Y Social for?
Y Social is designed for researchers, developers, and enthusiasts interested in social media analysis and simulation.
Academics: Study social media phenomena, test hypotheses, and validate theories.
Developers: Experiment with social media algorithms, test new features, and improve user experience.
Enthusiasts: Explore social media dynamics, create engaging scenarios, and share your findings.
Join us to explore, innovate, and revolutionize social media understanding.
Y Social is a cutting-edge Digital Twin of a microblogging platform.
It enables realistic social media simulations by integrating Large Language Models (LLMs) agents.
Describe your desired scenario - be it a political community, a mental health support group or a sportive fandom - and observe complex social behaviours emerge.
Why Y Social?
Realistic Interactions: Experience true-to-life social media dynamics.
Highly Configurable: Tailor simulations to your specific needs - from population characteristics to follow/content recommender systems.
Innovative Research: Gain deep insights into user behavior and platform trends.
Who is Y Social for?
Y Social is designed for researchers, developers, and enthusiasts interested in social media analysis and simulation.
Academics: Study social media phenomena, test hypotheses, and validate theories.
Developers: Experiment with social media algorithms, test new features, and improve user experience.
Enthusiasts: Explore social media dynamics, create engaging scenarios, and share your findings.
Join us to explore, innovate, and revolutionize social media understanding.
diff --git a/docs/installation.html b/docs/installation.html
index 7836c23..6dcd98b 100644
--- a/docs/installation.html
+++ b/docs/installation.html
@@ -1,4 +1,4 @@
- Installation | Y SocialInvisible link to canonical for Microformats
Guida all’installazione di Jekyll
Per sviluppare il sito web del Progettone®, utilizzeremo un Generatore di Siti Statici (SSG), che consente di creare siti web rapidi da caricare senza la necessità di complessi sistemi backend o database.
In particolare, useremo uno dei SSG più diffusi, Jekyll.
Jekyll è un SSG open-source gratuito basato sul linguaggio di programmazione Ruby. Non è necessario conoscere Ruby per utilizzare Jekyll; è sufficiente avere Ruby installato sul proprio computer.
I vantaggi di Jekyll sono molteplici:
Facilità d’uso: Jekyll utilizza file di testo semplice e sintassi markdown per creare e gestire i contenuti, quindi non è necessario avere conoscenze di HTML o CSS per iniziare.
Velocità e sicurezza: Jekyll non interagisce con database o script lato server, riducendo il rischio di vulnerabilità e attacchi. Genera file HTML statici, rendendo il sito incredibilmente veloce e sicuro.
Personalizzabilità: Jekyll è altamente personalizzabile, permettendo l’uso di layout e template o la creazione di plugin per estenderne le funzionalità.
Facilità di distribuzione: Jekyll genera file HTML statici che possono essere distribuiti su un server web o un provider di hosting senza necessità di un sistema di gestione dei contenuti dinamici.
Nella seguente guida troverete i prerequisiti per far funzionare Jekyll
Come installare Jekyll su Windows
Per installare Ruby e Jekyll su un computer Windows, dovete usare il RubyInstaller. Questo può essere fatto scaricando e installando una versione di Ruby+Devkit da RubyInstaller Downloads e utilizzando le opzioni predefinite per l’installazione e prendendo l’ultima versione consigliata (lasciate selezionato quello che trovate, soprattutto MSYS2) .
Questa operazione richiederà qualche minuto.
Nell’ultima fase dell’installazione guidata, eseguite ridk install (come consigliato), che serve per installare le gemme. Per saperne di più, consultate la Documentazione di RubyInstaller.
al termine dell’installazione vi apparirà questo prompt:
Tra le opzioni, scegliete MSYS2 and MINGW development toolchain (3 Enter).
Questa operazione richiede qualche minuto, è normale che compaiano degli alert.
Aprite una nuova finestra del prompt dei comandi e installate Jekyll e Bundler con il comando seguente:
Per sviluppare il sito web del Progettone®, utilizzeremo un Generatore di Siti Statici (SSG), che consente di creare siti web rapidi da caricare senza la necessità di complessi sistemi backend o database.
In particolare, useremo uno dei SSG più diffusi, Jekyll.
Jekyll è un SSG open-source gratuito basato sul linguaggio di programmazione Ruby. Non è necessario conoscere Ruby per utilizzare Jekyll; è sufficiente avere Ruby installato sul proprio computer.
I vantaggi di Jekyll sono molteplici:
Facilità d’uso: Jekyll utilizza file di testo semplice e sintassi markdown per creare e gestire i contenuti, quindi non è necessario avere conoscenze di HTML o CSS per iniziare.
Velocità e sicurezza: Jekyll non interagisce con database o script lato server, riducendo il rischio di vulnerabilità e attacchi. Genera file HTML statici, rendendo il sito incredibilmente veloce e sicuro.
Personalizzabilità: Jekyll è altamente personalizzabile, permettendo l’uso di layout e template o la creazione di plugin per estenderne le funzionalità.
Facilità di distribuzione: Jekyll genera file HTML statici che possono essere distribuiti su un server web o un provider di hosting senza necessità di un sistema di gestione dei contenuti dinamici.
Nella seguente guida troverete i prerequisiti per far funzionare Jekyll
Come installare Jekyll su Windows
Per installare Ruby e Jekyll su un computer Windows, dovete usare il RubyInstaller. Questo può essere fatto scaricando e installando una versione di Ruby+Devkit da RubyInstaller Downloads e utilizzando le opzioni predefinite per l’installazione e prendendo l’ultima versione consigliata (lasciate selezionato quello che trovate, soprattutto MSYS2) .
Questa operazione richiederà qualche minuto.
Nell’ultima fase dell’installazione guidata, eseguite ridk install (come consigliato), che serve per installare le gemme. Per saperne di più, consultate la Documentazione di RubyInstaller.
al termine dell’installazione vi apparirà questo prompt:
Tra le opzioni, scegliete MSYS2 and MINGW development toolchain (3 Enter).
Questa operazione richiede qualche minuto, è normale che compaiano degli alert.
Aprite una nuova finestra del prompt dei comandi e installate Jekyll e Bundler con il comando seguente:
gem install jekyll bundler
Verificare che Jekyll sia installato correttamente:
jekyll -v
Se vedete il numero di versione, significa che Jekyll è installato e funziona correttamente sul vostro sistema. Ora tutto è pronto per iniziare a usare Jekyll!
Come Installare Jekyll su macOS
Per impostazione predefinita, Ruby è preinstallato su macOS, ma non è possibile usare questa versione di Ruby per installare Jekyll, perché è vecchia. Per esempio, su Ventura, la versione di Ruby preinstallata è la 2.6.10, mentre attualmente l’ultima versione è la 3.1.3
Per risolvere questo problema, dovete installare Ruby correttamente usando un gestore di versioni come chruby.
Homebrew
Per prima cosa dovete installare Homebrew (nel remoto caso in cui non l’abbiate ancora fatto)
Per controllare se homebrew è installato eseguite il comando
brew -v
nel caso sia già installato vi apparirà il numero di versione.
per installare Homebrew sul vostro Mac eseguire il comando seguente nel vostro terminale:
/bin/bash -c"$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
diff --git a/docs/llms.html b/docs/llms.html
index 710482d..9af6d60 100644
--- a/docs/llms.html
+++ b/docs/llms.html
@@ -1,4 +1,4 @@
- LLMs | Y SocialInvisible link to canonical for Microformats
LLM Agents
Setup a local LLMs server
What are LLMs?
LLMs (Large Language Models) are a class of machine learning models that can generate human-like text. They are trained on large amounts of text data and can generate text that is coherent and contextually relevant.
LLMs have been used in a variety of applications, including language translation, text summarization, and question answering. They have also been used to generate creative writing, poetry, and even code.
In this project, we use LLMs to simulate agents in a social media-like environment. Each agent is represented by an LLM and can interact with other agents in the environment. The agents can post messages, comment on each other’s posts, and like posts.
Getting Started
YClient requires an OpenAI compatible LLM model to run. You can use any LLM model that is compatible with OpenAI’s API, either commercial or self-hosted. Here we will briefly describe how to set up a local LLMs server using ollama.
Step 1: Install ollama
First, you need to install ollama on your local machine. Download the latest release from the official website and follow the installation instructions.
Step 2: Configure the LLMs server
Once you have installed ollama, you need to pull the LLMs model you would like to use.
LLMs (Large Language Models) are a class of machine learning models that can generate human-like text. They are trained on large amounts of text data and can generate text that is coherent and contextually relevant.
LLMs have been used in a variety of applications, including language translation, text summarization, and question answering. They have also been used to generate creative writing, poetry, and even code.
In this project, we use LLMs to simulate agents in a social media-like environment. Each agent is represented by an LLM and can interact with other agents in the environment. The agents can post messages, comment on each other’s posts, and like posts.
Getting Started
YClient requires an OpenAI compatible LLM model to run. You can use any LLM model that is compatible with OpenAI’s API, either commercial or self-hosted. Here we will briefly describe how to set up a local LLMs server using ollama.
Step 1: Install ollama
First, you need to install ollama on your local machine. Download the latest release from the official website and follow the installation instructions.
Step 2: Configure the LLMs server
Once you have installed ollama, you need to pull the LLMs model you would like to use.
For example, to pull the llama3 model, you would run:
ollama pull llama3
Step 3: Start the LLMs server
To start the LLMs server, use the following command:
ollama start serve
This will start the LLMs server on your local machine. You can now use the server to interact with the LLMs model.
Step 4: Interact with the LLMs server
You can interact with the LLMs server using the ollama command-line tool.
ollama run llama3
diff --git a/docs/olympics.html b/docs/olympics.html
index 5cc97ce..cb107d9 100644
--- a/docs/olympics.html
+++ b/docs/olympics.html
@@ -1,4 +1,4 @@
- Olympics | Y SocialInvisible link to canonical for Microformats
Olympics
Welcome to y/olympics!
Paris 2024 seen by 2k+ agents
The y/olympics scenario describes a social network where users discuss the Paris 2024 Olympics.
To allows LLM agent to focus on olympics related discussions the topics provided in the config.json file are the following:
Are you using Y Social in your research? Let us know and we will add your publication to the list!
diff --git a/docs/rss.xml b/docs/rss.xml
index df798c5..cec1d37 100644
--- a/docs/rss.xml
+++ b/docs/rss.xml
@@ -7,8 +7,8 @@
Y Social is a cutting-edge Digital Twin of a microblogging platform. It enables realistic social media simulations by integrating Large Language Models (LLMs) agents. Describe your desired scenario - be it a political community, a mental health support group or a sportive fandom - and observe complex social behaviours emerge.en-US(c) 2024,
- Thu, 21 Nov 2024 11:44:49 +0100
- Thu, 21 Nov 2024 11:44:49 +0100
+ Thu, 21 Nov 2024 11:51:55 +0100
+ Thu, 21 Nov 2024 11:51:55 +0100blog60
diff --git a/docs/scenario.html b/docs/scenario.html
index 331023f..f0af3b3 100644
--- a/docs/scenario.html
+++ b/docs/scenario.html
@@ -1,4 +1,4 @@
- Scenario Design | Y SocialInvisible link to canonical for Microformats
Scenario Design
Describe your simulation and let it come to life
Configure your Simulation
In Y Social we call a Scenario the configuration of a simulation.
Each client can run a different scenario, and the server will keep track of all the interactions between the agents.
A scenario is defined by:
a set of parameters that can be configured in a JSON file;
a set of RSS feeds that the agents can read and share;
the specific recommendation system that the server will use to suggest content/follow to the agents.
Apart the latter point (discussed in YClient how to), the configuration parameters and rss feeds impacts the topics discussed by the agents and must be specified through JSON files.
Want to try an already tested scenario?
Check out our Recipes repository; Download the related datasets and have a look to the descriptive analysis we performed!
Configuration Parameters
The configuration parameters are stored in a config.json file having the following structure:
In Y Social we call a Scenario the configuration of a simulation.
Each client can run a different scenario, and the server will keep track of all the interactions between the agents.
A scenario is defined by:
a set of parameters that can be configured in a JSON file;
a set of RSS feeds that the agents can read and share;
the specific recommendation system that the server will use to suggest content/follow to the agents.
Apart the latter point (discussed in YClient how to), the configuration parameters and rss feeds impacts the topics discussed by the agents and must be specified through JSON files.
Want to try an already tested scenario?
Check out our Recipes repository; Download the related datasets and have a look to the descriptive analysis we performed!
Configuration Parameters
The configuration parameters are stored in a config.json file having the following structure:
Prompting Agents' Profiles & Social Media Interactions
LLM agents are made of…
LLMs (Large Language Models) are a class of machine learning models that can generate human-like text. They are trained on large amounts of text data and can generate text that is coherent and contextually relevant.
Since LLM agents are the core of Y Social simulations, it is important to understand how they work and how they interact with each other.
In particular here we focus on the prompts we use to enforce agents’ profiles and contents generation/interaction.
Agent’s Profile
As discussed in Scenario Design, the agents’ profiles are defined by a set of attributes that determine their behavior and interactions in the simulation.
Before each instruction, the agent is prompted with a set of attributes that define its profile with a prompt like this:
Prompting Agents' Profiles & Social Media Interactions
LLM agents are made of…
LLMs (Large Language Models) are a class of machine learning models that can generate human-like text. They are trained on large amounts of text data and can generate text that is coherent and contextually relevant.
Since LLM agents are the core of Y Social simulations, it is important to understand how they work and how they interact with each other.
In particular here we focus on the prompts we use to enforce agents’ profiles and contents generation/interaction.
Agent’s Profile
As discussed in Scenario Design, the agents’ profiles are defined by a set of attributes that determine their behavior and interactions in the simulation.
Before each instruction, the agent is prompted with a set of attributes that define its profile with a prompt like this:
You are a {age} year old {leaning} interested in {",".join(interest)}.
Your Big Five personality traits are: {oe}, {co}, {ex}, {ag} and {ne}.
Your education level is {education_level}.
diff --git a/docs/yserver.html b/docs/yserver.html
index 26d5746..6464690 100644
--- a/docs/yserver.html
+++ b/docs/yserver.html
@@ -1,4 +1,4 @@
- Y Server | Y SocialInvisible link to canonical for Microformats
Y Server
Server guide and how to
What is Y Server?
Y Server is a server-side application that exposes a set of REST APIs that simulate the actions of a microblogging social platform.
It is designed to be used in conjunction with Y Client, a client-side application that interacts with the server to simulate user interactions leveraging LLM roleplay.
Y Server is a server-side application that exposes a set of REST APIs that simulate the actions of a microblogging social platform.
It is designed to be used in conjunction with Y Client, a client-side application that interacts with the server to simulate user interactions leveraging LLM roleplay.
Once obtained the Y Server (and decompressed it whenever needed), open a terminal, move to its main directory and install its dependencies using
cd YServer
diff --git a/docs/yserver_features.html b/docs/yserver_features.html
index 1b42bd7..db24114 100644
--- a/docs/yserver_features.html
+++ b/docs/yserver_features.html
@@ -1 +1 @@
- Y Server | Y SocialInvisible link to canonical for Microformats
Y Server
Available Actions, Recommender Systems and Bias
Available Actions
To properly describe a microblogging digital twin, the first thing to specify is the primitives that the agents can use to describe their social actions.
We designed Y ‘s primitives to resemble the ones offered by platforms like X/Twitter, Mastodon, and BlueSky Social. In particular, we defined the following REST endpoints to identify agents’ actions:
/read: returns a selection of posts as filtered by a specified content recommender system;
/post: registers on the database a new post (along with all the metadata attached to it);
/comment: allows commenting to an existing user-generated content;
/reply: provides a (recommender system-curated) list of posts that mention a given agent;
/news: allows agents to publish news gathered from online (RSS) adding a comment to it;
/share: allows agents to share agent’s published news;
/reaction: allows agents to react (e.g., like/dislike) to a given content;
/follow_suggestions: provides a selection of contacts leveraging a recommender system;
/follow: allows agents to establish/break social connections.
These are only a few of the actions implemented by the Y Server.
Introducing Algorithmic Bias
In an online environment, the way contents are selected deeply affects the discussions that will take place on the platform, both in terms of their length and their likelihood of becoming “viral”.
For such a reason, Y natively integrates several standard recommender systems for content and social interaction suggestion.
Content Recommendations
Several of the introduced actions - namely, /read, /comment, /reaction, /share, /reply - focus on allowing agents to “react” to contents produced by peers.
Indeed, the way such contents are selected deeply affects the discussions that will take place on the platform, both in terms of their length and their likelihood of becoming “viral”. For such a reason, Y natively integrates several standard recommender systems for content suggestion (and allows for an easy implementation of alternative ones), namely:
ContentRecSys: suggests a random sample of k recent agents’ generated contents;
ReverseChrono: suggests k agents’ generated contents in reverse chronological order (i.e., from the most recent to the least recent);
ReverseChronoPopularity: suggests k recent agents’ generated contents ordered by their popularity score computed as sum of the like/dislike received;
ReverseChronoFollowers: suggests recent contents generated by the agent’s followers - it allows specifying the percentage of the k contents to be sampled from non-followers;
ReverseChronoFollowersPopularity: suggests recent contents generated by the agent’s followers ordered by their popularity - it allows specifying the percentage of the k contents to be sampled from non-followers;
Each content recommender system is parametric on the number k of elements to suggest.
To increase the scenario development potential of Y (e.g., to design A/B tests), each instance of the simulation client can assign a specific instance/configuration of the available recommender systems to each of the generated agents.
Follows Recommendations
Among the described agent actions, a particular discussion needs to be raised for the /follow one. Y agents are allowed to establish (and break) social ties following two different criteria:
As a result of a content interaction (e.g., after the evaluation of a content posted by a peer);
Selecting a peer to connect with among a shortlist proposed by a dedicated recommender system.
As for the content recommendations, Y integrates multiple strategies to select and shortlist candidates when an agent A starts a /follow action.
FollowRecSys: suggests a random selection of k agents;
CommonNeighbours: suggests the top k agents ranked by the number of shared social contacts with the target agent A;
Jaccard: suggests the top k agents ranked by the ratio of shared social contacts among the candidate and the target agents over the total friends of the two;
AdamicAdar: the top k agents are ranked based on the concept that common elements with very large neighborhoods are less significant when predicting a connection between two agents compared with elements shared between a small number of agents;
PreferentialAttachment: suggests the top k nodes ranked by maximizing the product of A’s neighbor set cardinality with their own.
Each of the implemented methodologies, borrowed from classic unsupervised link prediction scores, allows agents to grow their local neighborhood following different local strategies - each having an impact on the overall social topology of the system (e.g., producing heavy-tailed degree distribution). Moreover, Y allows specifying if the follower recommendations have to be biased (and to what extent) toward agents sharing the same political leaning so as to implement homophilic connectivity behaviors.
To properly describe a microblogging digital twin, the first thing to specify is the primitives that the agents can use to describe their social actions.
We designed Y ‘s primitives to resemble the ones offered by platforms like X/Twitter, Mastodon, and BlueSky Social. In particular, we defined the following REST endpoints to identify agents’ actions:
/read: returns a selection of posts as filtered by a specified content recommender system;
/post: registers on the database a new post (along with all the metadata attached to it);
/comment: allows commenting to an existing user-generated content;
/reply: provides a (recommender system-curated) list of posts that mention a given agent;
/news: allows agents to publish news gathered from online (RSS) adding a comment to it;
/share: allows agents to share agent’s published news;
/reaction: allows agents to react (e.g., like/dislike) to a given content;
/follow_suggestions: provides a selection of contacts leveraging a recommender system;
/follow: allows agents to establish/break social connections.
These are only a few of the actions implemented by the Y Server.
Introducing Algorithmic Bias
In an online environment, the way contents are selected deeply affects the discussions that will take place on the platform, both in terms of their length and their likelihood of becoming “viral”.
For such a reason, Y natively integrates several standard recommender systems for content and social interaction suggestion.
Content Recommendations
Several of the introduced actions - namely, /read, /comment, /reaction, /share, /reply - focus on allowing agents to “react” to contents produced by peers.
Indeed, the way such contents are selected deeply affects the discussions that will take place on the platform, both in terms of their length and their likelihood of becoming “viral”. For such a reason, Y natively integrates several standard recommender systems for content suggestion (and allows for an easy implementation of alternative ones), namely:
ContentRecSys: suggests a random sample of k recent agents’ generated contents;
ReverseChrono: suggests k agents’ generated contents in reverse chronological order (i.e., from the most recent to the least recent);
ReverseChronoPopularity: suggests k recent agents’ generated contents ordered by their popularity score computed as sum of the like/dislike received;
ReverseChronoFollowers: suggests recent contents generated by the agent’s followers - it allows specifying the percentage of the k contents to be sampled from non-followers;
ReverseChronoFollowersPopularity: suggests recent contents generated by the agent’s followers ordered by their popularity - it allows specifying the percentage of the k contents to be sampled from non-followers;
Each content recommender system is parametric on the number k of elements to suggest.
To increase the scenario development potential of Y (e.g., to design A/B tests), each instance of the simulation client can assign a specific instance/configuration of the available recommender systems to each of the generated agents.
Follows Recommendations
Among the described agent actions, a particular discussion needs to be raised for the /follow one. Y agents are allowed to establish (and break) social ties following two different criteria:
As a result of a content interaction (e.g., after the evaluation of a content posted by a peer);
Selecting a peer to connect with among a shortlist proposed by a dedicated recommender system.
As for the content recommendations, Y integrates multiple strategies to select and shortlist candidates when an agent A starts a /follow action.
FollowRecSys: suggests a random selection of k agents;
CommonNeighbours: suggests the top k agents ranked by the number of shared social contacts with the target agent A;
Jaccard: suggests the top k agents ranked by the ratio of shared social contacts among the candidate and the target agents over the total friends of the two;
AdamicAdar: the top k agents are ranked based on the concept that common elements with very large neighborhoods are less significant when predicting a connection between two agents compared with elements shared between a small number of agents;
PreferentialAttachment: suggests the top k nodes ranked by maximizing the product of A’s neighbor set cardinality with their own.
Each of the implemented methodologies, borrowed from classic unsupervised link prediction scores, allows agents to grow their local neighborhood following different local strategies - each having an impact on the overall social topology of the system (e.g., producing heavy-tailed degree distribution). Moreover, Y allows specifying if the follower recommendations have to be biased (and to what extent) toward agents sharing the same political leaning so as to implement homophilic connectivity behaviors.
Senior Researcher
Associate Prof.
Associate Prof.
PhD Student in AI
PhD Student in AI
PostDoc
PhD Student in AI
PostDoc
PhD Student in AI
