WeChat Chat Analyzer and Predictor

Context and Problem

WeChat chat logs are a significant part of one's digital footprint. Many users desire to analyze their communication patterns, emotional shifts, or relationship dynamics but are deeply concerned about privacy. Uploading raw records containing private conversations to third-party cloud services for analysis poses unacceptable security risks. This tool addresses this core conflict: performing deep, multi-dimensional analysis and prediction on raw chat files in a completely offline, local environment.

Core Capabilities and Workflow

The tool uses a modular architecture. The core process revolves around local files.

1. Data Ingestion and Preprocessing

Raw chat files in txt, json, docx, or pdf formats are imported via the --input parameter. The tool auto-detects encodings (e.g., utf-8, gb18030) and parses the data into a structured format. All processing occurs locally; the original file remains untouched and un-uploaded.

2. Multi-dimensional Analysis (Entry: analyze-v2)

The analysis engine consists of nine independent modules, outputting structured JSON data:
- Personality & Sentiment Insights: Employs keyword rules and jieba segmentation for heuristic MBTI and Big Five personality inference. Sentiment analysis combines lexicons, emoji mapping, and sarcasm detection, now enhanced with rolling_trend (moving average) and turning_points detection.
- Relationship & Risk Assessment: A new Relationship Health Analyzer scores interaction balance, responsiveness, emotional synchronicity, and more on a 0-100 scale. The risk detection module identifies four risk types while excluding sarcastic content.
- Predictive Scenario Simulation: Provides five types of future scenario simulations (e.g., trend, behavior, risk evolution) along with interpretations. All analysis is generated by local algorithms.

3. Conversation Prediction (Entry: predict-v2)

Fuses three-layer models: Rule-based, Retrieval-Augmented Generation (RAG), and MiroFish for next-turn prediction. The model is time-aware, adjusting predictions based on context like weekends or message intervals.

4. Auxiliary Features

• Calendar Event Extraction: Identifies and structures appointments, reminders, birthdays, etc., from chats.
• Voice Interaction: Launched via python scripts/main.py voice, supporting spoken commands with a multi-tier fallback strategy for TTS/STT to accommodate varying hardware.
• Report Generation: Using --html creates a single, offline HTML file with all charts (e.g., SVG) inlined for easy archiving and sharing.

Applicability and Important Notes

• Disclaimer: All analyses (personality, sentiment, risk, relationship health) are heuristic estimates intended for reference only. They must not be used for professional psychological assessment, clinical diagnosis, or major life decisions.
• Privacy Boundary: The tool itself makes zero network requests (except for the initial download of the RAG embedding model). However, users must ensure their acquisition of the source chat file complies with relevant laws and regulations.
• Capability Limits: Analysis depth is constrained by the text content of the chat. Non-text elements like voice messages, images, and links are currently outside the analysis scope.
• Performance Considerations: The first use of RAG functionality requires downloading a ~50-100MB local model. Complex analyses on large files may be time-consuming. A health check can be run via python scripts/main.py doctor.