Struggling with LSI keyword integration for semantic search?

hey folks, been diving deep into optimising our content for LSI keywords, specifically how google *really* parses them for semantic search relevence.

i'm running into issues where our internal scoring models for contextual density don't quite align with observed SERP movements, almost like there's a disconnect in weighting.

for example, i'm seeing this kind of output in our analysis, and i'm not sure if it's an internal miscalibration or something deeper with google's NLP:

[INFO] LSI_SCORE_CALC: 'data warehousing' - 0.78
[INFO] KEYWORD_DENSITY: 'data warehousing' - 2.1%
[WARN] SEMANTIC_RELEVANCE: 'data analytics' - LOW (expected HIGH)
[ERROR] TOPIC_CLUSTER_ALIGN: 'data science' - MISMATCH

is there a more granular approach to mapping LSI terms to specific entity relationships that i'm missing? help a brother out please...