Struct SemanticTopicModeller 

pub struct SemanticTopicModeller {
    pub topics: HashMap<u64, TopicModel>,
    pub assignments: Vec<TopicAssignment>,
    pub next_topic_id: u64,
    pub config: ModellerConfig,
}

Online topic modeller that clusters embeddings into latent topics.

New documents are assigned to the closest existing topic (by cosine similarity). If the best similarity is below ModellerConfig::new_topic_threshold and the topic limit has not been reached, a fresh topic is created.

Fields

topics: HashMap<u64, TopicModel>

All topics keyed by topic_id.

assignments: Vec<TopicAssignment>

All assignment records in insertion order.

next_topic_id: u64

Monotonically increasing counter used to allocate topic IDs.

config: ModellerConfig

Modeller configuration.

Implementations

impl SemanticTopicModeller

pub fn new(config: ModellerConfig) -> Self

Creates a new, empty topic modeller with the given configuration.

pub fn assign( &mut self, doc_id: u64, embedding: Vec<f32>, now_secs: u64, ) -> TopicAssignment

Assigns embedding (associated with doc_id) to a topic.

Algorithm

1. If there are no topics yet, always create the first topic.
2. Find the topic whose centroid has the highest cosine similarity to embedding.
3. If max_sim < new_topic_threshold and the number of topics is below max_topics, create a new topic with centroid = embedding.
4. Otherwise assign to the best topic and update its centroid via the online update rule: centroid = (1 - lr) * centroid + lr * embedding.

Returns the TopicAssignment that was recorded.

pub fn topic(&self, topic_id: u64) -> Option<&TopicModel>

Returns a reference to a topic by ID, or None if not found.

pub fn assignments_for_doc(&self, doc_id: u64) -> Vec<&TopicAssignment>

Returns all assignments for the given document, sorted by assigned_at_secs descending (most recent first).

pub fn top_topics(&self, k: usize) -> Vec<&TopicModel>

Returns up to k topics sorted by member_count descending.

pub fn relabel(&mut self, topic_id: u64, label: String) -> bool

Renames a topic. Returns false if the topic ID does not exist.

pub fn stats(&self) -> TopicModellerStats

Returns aggregate statistics for the current state of the modeller.