"While you're still gathering information manually, his AI team has delivered a 25-page in-depth report."
Last week, the GitHub Hotlist suddenly killed a dark horse - a program called BettaFish（The open source project of "Micro Opinion" (Chinese name "微舆"), which has collected 4000+ Stars within 24 hours, eventually rushed to the first place of the hot list.
What's more surprising is that the author of this AI public opinion analyzer is just a 20-year-old college student. Initially, it was just an ordinary coursework, but now it has become the meat and potatoes in the eyes of countless enterprises and research organizations.
Today, let's step into this multi-Agent opinion assistant, which has been hailed as usable by everyone, and see how it redefines the matter of information gathering with AI.
🐟 What is Micro Opinion? From Coursework to GitHub Explosion
"Micro Opinion" (BettaFish) is aInnovative Multi-Intelligence Body Public Opinion Analysis SystemThe name is taken from "Fighting Fish", a small but powerful ornamental fish, symbolizing "small but powerful, not afraid of challenges".
Its core capabilities are simple, yet subversive:
You just need to put forward the analysis requirements like chatting, and the system can fully automate the analysis of 30+ mainstream social media at home and abroad with millions of comments, and deliver a professional-grade research report.
For example, a directive like this:
"A comprehensive analysis of Wuhan University's recent online public opinion, including changes in voice volume, user profiles, risk points and opportunities"
In a few short minutes, it will give you aChapter 8, verse 25The in-depth analysis report covering:
- Brand Voice Trends and Key Events
- User Emotion Distribution and Profiling
- Public Opinion Risk Warning
- Brand Opportunity Tapping
- Multi-Platform Comparative Analysis
- Dissemination path tracking
- Public Opinion Forecasts and Suggestions
- Appendix data charts
Click here to see an example of an in-depth brand reputation analysis report of Wuhan University
What kind of tool is this? It's like having a 24-hour AI intelligence team assigned to you.
🤖 Technology Revealed: 5 Agents, One AI Expert Team
Instead of a single large model at the core of a micro-opinion, theCollaborative system consisting of multiple specialized AgentsThe workflow of a real team of experts is simulated by the fact that they each have their own role to play and debate with each other.
1️⃣ Query Agent: Breadth Search Specialist
- 7×24 hours monitoring of 30+ mainstream media at home and abroad
- Capture hot content and grab the core message
- Reach Weibo, Xiaohongshu, Jieyin, Twitter, Facebook and other platforms
2️⃣ Media Agent: a multimodal solver
- Ability to look at pictures, listen to audio, and understand short video content
- Recognize visual content from platforms such as Shake and Shuffle
- Extract structured information such as weather, stocks, etc. from search engines
3️⃣ Insight Agent: Deep Digger
- Specializing in private databases and analyzing millions of reviews
- Integration of multiple sentiment analysis models (BERT, Qwen3, multilingual)
- Recognizing Hidden Emotions, Attitudes, and Potential Risks
4️⃣ Report Agent: Master Reporter
- Dynamic selection of optimal report templates
- Automatic generation of graphical HTML reports
- Supports a variety of scenarios such as commercial branding and public events
5️⃣ Forum Facilitator: Catalyst for Thought Collision
- Most innovative design: mock expert debates
- Different Agents submit ideas and the moderator leads the discussion
- Avoiding the limitations of single-model thinking and producing more comprehensive conclusions
The mistake traditional AI tools often make is: one person thinks and a group of people nod. And micro-opinion makes AI really argue, collide and agree like human experts.
⚡ Six Core Strengths: Why it's an overnight sensation.
✅ 1. Total Surveillance, More Than Meets the Eye
The AI crawler cluster operates 24/7, not only crawling hot content, but alsoIn-depth analysis of massive user reviewsIt allows you to hear the truest and widest range of public voices and breaks up the "information cocoon".
✅ 2. Composite analysis beyond LLM
Instead of relying only on LLM, the micro-opinions incorporate theFine-tuning Model + Statistical Model + Multi-Agent CollaborationThe analysis of the results of the study was conducted in a manner that ensured both depth and accuracy.
✅ 3. True multimodal capabilities
Break through the plain text limitations and be able toParsing short videos, images, structured data cards, comprehensively grasp the dynamics of public opinion. It has watched 1,000 Jitterbug videos when others are still reading text comments.
✅ 4. Forum-style debate mechanism
The different Agents are given a unique mindset, which is chained together through a "forum" mechanism. This makes the results not one-sided, butConsensus after multiple validations and debatesThe
✅ 5. Seamless integration of public and private data
It not only analyzes public opinion, but also provides a high-security interface to support theInterfacing with internal operational databasesWe can help you to realize the in-depth analysis of "external trends + internal insights". Open up data barriers to realize in-depth analysis of "external trends + internal insights".
✅ 6. Lightweight and easily scalable
Based on a pure Python modular design, the implementation of theOne-Click Deployment. The code is clearly structured and can be easily customized by developers to turn it into a financial analysis system, market monitoring tool and more.
📊 Real-world testing: from instructions to 25-page report in a cup of coffee
Let's take a look at a real-life case study: what is the complete analysis process, taking Wuhan University's public opinion as an example?
📌 Step 1: User Questions
Enter it in the system:
"A comprehensive analysis of Wuhan University's recent online public opinion situation, including changes in voice volume, user profiles, risk points and opportunities"
📌 Step 2: The three Agents are launched in parallel
- Query Agent starts broad search: Weibo, Zhihu, Shake, Xiaohongshu...
- Media Agent analyzes multimodal content: campus photos, short videos, graphics
- Insight Agent mines historical data: comments on sentiment, topic evolution
📌 Step 3: Forum Debate
Agents submit preliminary findings:
- Query Agent: "Sound volume surges 300% during cherry blossom season at WU"
- Media Agent: "Photo Analysis Shows Alumni Nostalgia Accounts for 65%"
- Insight Agent: "Negative reviews focused on the ticket reservation system"
📌 Step 4: Report Generation
Report Agent integrates all the information, automatically selects the template of "Public Opinion Analysis of Universities", and generates an in-depth report with 8 chapters and 25 sections, including 30+ data charts.
Click to watch a video of the full run
The whole thing.No manual intervention requiredIt's like having a professional market research team on call 24 hours a day.
🚀 A must-see for techies: how to deploy your own opinion analytics team?
Deployment of a microcosm is easier than expected, even for novices.
base environment
- Operating System: Windows/Linux/MacOS
- Python 3.9+
- 2GB+ RAM
- MySQL database (optional cloud service)
Three-step rapid deployment
# 1. Create the environment
conda create -n weiyu python=3.11
conda activate weiyu
# 2. Install dependencies
pip install -r requirements.txt
playwright install chromium # install browser driver
# 3. Configure API key
cp .env.example .env # fill in your API key
# 4. Start the system
python app.pyinterviews http://localhost:5000, your AI opinion analysis team is in place.
💡 Tip: The project provides a detailedDeployment GuideIt also supports the direct use of cloud database services (containing 100,000+ real opinion data).
🌟 From homework to hot list: a college student's open source story
The author of the project, originally a general computer science student, took a big data analytics course this semester.
"My teacher at the time assigned a public opinion analysis assignment, and I thought to myself: why don't I let the AI do it for me?"
From the humble beginnings of a crawler script, to a multi-agent collaboration system, to the #1 spot on the GitHub Hotlist, it took the author less than 3 months.
Today, with this project, he has received internship offers from several tech companies and has even been contacted by investors to explore commercialization possibilities.
🔮 The Future is More Than Opinion
The current version of Micro Opinion has already realized Input→Analysis, but the team's ambition goes beyond that:
The next step is for them to implement a true predictive feature.
Based on the long-accumulated topic heat data across the web, the team plans to integrate theTemporal models, graph neural networks, multimodal fusionand other technologies to realize it:
- Topic Heat Prediction
- Public Opinion Inflection Point Warning
- Risk diffusion simulation
- Window of opportunity identification
"It starts with opinion, but goes beyond opinion." The project document reads, "Micro Opinion's goal is to become a concise and universal data analysis engine that drives all business scenarios."
👉 Project address
https://github.com/666ghj/BettaFish