Memory System - Conversation Segment Enhancer

File: cogs/memory/conversation_segment_enhancer.py

The ConversationSegmentEnhancer is a post-processing module that refines the results returned by the SearchEngine. Its primary goal is to add a layer of contextual relevance to the search results, making them more useful for the StructuredContextBuilder.

ConversationSegmentEnhancer Class

__init__(self, memory_manager: MemoryManager)

Initializes the enhancer with an instance of the MemoryManager, which it uses to perform the initial search.

async search_enhanced_segments(self, search_context: SearchContext) -> List[EnhancedSegment]

This is the main method of the class. It takes a SearchContext object, performs a search, and then enhances the results.

• Parameters:
  • search_context (SearchContext): A data class containing the search query, channel ID, and information about the current participants in the conversation.
• Process:
  1. It performs a standard search using the memory_manager.search_memory() method.
  2. It passes the raw results to _enhance_search_results().
  3. The enhanced results are then passed to _rank_segments_by_relevance() for final scoring and sorting.
  4. The top N results are returned.
• Returns: A list of EnhancedSegment objects.

Enhancement Logic

The "enhancement" process involves adding new metadata and calculating a more nuanced relevance score.

_enhance_search_results(...)

This method converts the raw message dictionaries from the search result into EnhancedSegment data classes. The key step here is determining the is_participant_related flag. It checks if the author of a retrieved message is one of the active participants in the current conversation (as defined in the search_context).

_rank_segments_by_relevance(...)

This method calculates a new, enhanced relevance score for each segment to provide a more context-aware ranking. The final score is a combination of several factors:

• Base Score: The original relevance score from the SearchEngine.
• Participant Boost: A significant score bonus (participant_score_boost) is added if is_participant_related is True. This prioritizes memories involving the people who are currently talking.
• Recency Boost: A smaller bonus (recency_score_boost) is added for recent messages (within the last 7 days), with the bonus decreasing as the message gets older.
• Content Quality Boost: A minor bonus is added for messages that are likely to be more meaningful (e.g., they are of a certain length, contain questions, or include specific keywords like "plan" or "issue").

The segments are then re-sorted based on this new, enhanced score before being returned.