Unverified50% confidenceFactExact time
The standard practice is to generate one embedding vector per chunk; when a chunk mixes multiple topics, that vector is the average of the semantics of each topic, making it hard for a query targeting a single detail to match precisely.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the hypergraph setting, the greedy algorithm at each step preferentially selects the sentence node participating in the greatest number of hyperedges, which is equivalent to preferentially selecting the sentence covering the most key topics.63% similarUnverifiedThe output space of a classification problem is a set of finite, unordered, mutually exclusive labels.63% similarUnverifiedMulti-query retrieval uses an LLM to automatically generate 3–5 semantically equivalent but differently phrased variant queries for the same question, retrieves results for each separately, and takes the union to improve recall.62% similarUnverifiedMeasures to deal with duplicate chunks include exact deduplication via content hashing, near-duplicate clustering based on embedding similarity, establishing normalization policies, and weighting by source metadata at query time, none of which require changing the model.61% similarUnverifiedEmbedding-based retrieval converts files or code blocks into vectors and stores them in an index, taking the most similar blocks at query time, which can turn stuffing the whole repo in into taking a small block by relevance, compressing the cost of a single call to a constant level.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1056903API
curl https://kongchang.com/api/v1/knowledge/claims/1056903MCP
get_claim(id=1056903)