RimTalk Common Knowledge Enhance

Unleash a chain reaction of knowledge! Allows common sense to snowball and evolve naturally.

• Core Mechanism:
  • Example: When the common sense entry `[Hello|1.00]World` is triggered, "World" is automatically extracted as a new keyword to match `[World|1.00]Hello`.
• Fine-Grained Control:
  • 1. Chain Settings (In Mod Settings):
    • Set a loop limit for the chain reaction.
    • Toggle this entire feature on or off.
  • 2. Library Permissions (In the Editing Interface):
    • Disable Matching [❌/✓]
    • Disable Extraction [❌/✓]
    • Both are disabled by default. You must manually check the box to allow an entry to be triggered or have keywords extracted from it. This effectively protects narrative entries (like short stories) from being broken down.
  • 3. Extraction Optimization (In Mod Settings):
    • Set a limit on the max number of extractable common sense entries to prevent over-extraction.
    • Set a limit on the max number of keywords that can be extracted from a single entry.
    • Apply a custom score bonus to extracted entries to prevent them from losing relevance.
• Bug Fixes:
  • Fixed the self-reinforcement issue that caused feedback loops.
  • Fixed a bug that could corrupt the contextual keyword list.
  • Fixed an issue where pawn information was not matched when both ‘Matching’ and ‘Extraction’ were disabled.

---

This feature is experimental and disabled by default in Mod Settings.

• Introducing "Word Contextual Freedom"
  • Logic: After generating 100 initial keywords, it filters out short Chinese phrases (3 characters or less) that contain stop words.
  • Effect: Prevents the generation of meaningless Chinese fragments like "的关" (de guan) or "们的" (men de).
• Improved Contextual Keyword Extraction (New Scoring System)
  • Logic:
    • 1. Length Weight: All values are now customizable in the mod settings!
      • Chinese: 2 chars=2.0f, 3 chars=8.0f, 4 chars=10f (Highest), 5 chars=4.0f, 6 chars=6.0f, 6+ chars=0f.
      • English: Starts at 2.0f, adds 2.0f per character, up to a max of 10f.
    • 2. Word Cohesion Factor: Checks if 2-3 character substrings of longer words also appear frequently. If so, a bonus is added (e.g., +0.5f, up to a max of +6.0f). This is also customizable in the mod settings!
    • 3. Composite Score:
      Length Weight² × Cohesion Factor × √Frequency
    • 4. The top 20 keywords are selected after sorting.
• Start/End Stop Words: Add stop words in the mod settings or via "Manage Stop Words" in the injection preview.
  For example, if you want to avoid meaningless keywords like "always xxx" or "at that time xxx", you can add "always" and "at that time" to the start stop word list.

---

Say goodbye to a sea of irrelevant keywords!

• Precision Extraction:
  • Optimized the core algorithm to more accurately identify pure Chinese characters, significantly improving matching accuracy.
  • Improved pawn information extraction by using Regex to match English keywords, preventing them from being broken into single letters.
• Ultimate Keyword Control:
  • Global/Save-Specific Important Keywords: Designate keywords that must be considered, ensuring core events (like "marriage" or "raid") are never missed.
  • Negative Keyword Blacklist: Block unwanted keywords to keep your dialogue pool clean and focused.
  • Quick Access to Keyword Manager: Instantly open the keyword management interface with a single click from the preview screen.

---

• All-in-One UI:
  • The UI now integrates keyword previews, common sense scores, and content keywords, making all information clear at a glance.
• Powerful Search & Bulk Management:
  • Search Function: Instantly find any keyword, no matter how large your custom list is.
  • Bulk Import: Save time by importing entire lists of keywords in one go.
• Stability & Purity:
  • Fixed a cache update bug that occurred after editing common sense entries.
  • Implemented punctuation filtering to ensure keywords remain clean.

  code by gemini2.5pro