LLM-free memory classification using regex-based pattern matching.
Classifies text into categories inspired by MemPalace's general_extractor:
- :decision — "we went with X because Y", "decided to use"
- :preference — "always use X", "never do Y", "I prefer Z"
- :milestone — "it works", "breakthrough", "finally fixed"
- :problem — "bug", "error", "doesn't work", "root cause"
- :emotional — "love", "scared", "proud", "grateful"
- :note — default for unclassified content
No LLM required — uses keyword and pattern matching.
Summary
Functions
Classify a text string into a memory type.
Classify with additional context for better accuracy.
Extract entity claims from text for contradiction detection.
Extract memory type from text (returns just the type atom).
Functions
Classify a text string into a memory type.
Returns {type, confidence} where:
- type is one of [:decision, :preference, :milestone, :problem, :emotional, :note]
- confidence is a float 0.0-1.0
Examples
iex> Recollect.Classification.classify("We decided to use PostgreSQL because it supports JSON")
{:decision, 0.8}
iex> Recollect.Classification.classify("I love working with this team")
{:emotional, 0.6}Classify with additional context for better accuracy.
Takes context_hints map with :git_repo, :file_path, :os keys.
Extract entity claims from text for contradiction detection.
Returns list of claim maps: %{entity: "Kai", predicate: :works_on, object: "auth", temporal: :current}
Extract memory type from text (returns just the type atom).