Entity-level aggregation

Description

Rolling up extracted entities and their associated sentiment, context and speakers across a corpus to see which features, competitors or workflows matter to whom.

After entities are extracted, aggregation turns mentions into analysable summaries: how many participants (not mentions) referred to each feature, with what sentiment or evaluation, in which segments, and alongside which other entities. The participant-level count matters because one talkative participant can dominate mention counts. Aspect-based analysis (Hu and Liu, 2004) links opinions to the entity they target rather than to the whole document. Rigour comes from normalising entity variants (lemmatisation, alias tables), counting at the participant level, reporting co-occurrence carefully, and treating counts as pointers back to transcripts rather than as prevalence estimates for the user population.

Attributes

Loss type
Unmigrated
Data
Mixed
Effort
Days

Steps

  1. Normalise entity mentions to canonical names using an alias table (e.g. 'auto-pay', 'autopay', 'recurring payments').
  2. Attach to each mention its speaker, participant attributes, sentence-level sentiment or evaluation and context.
  3. Aggregate at participant level: number and percentage of participants mentioning each entity, by segment.
  4. Compute entity co-occurrence within participants or passages and visualise as a matrix or network.
  5. Link every aggregate cell back to its underlying quotes so analysts can read the evidence.
  6. State clearly that counts describe the sample, not population prevalence, unless sampling supports inference.

Sources

  • spaCy (en_core_web_trf transformer model)
  • spacy-transformers
  • textacy
  • displacy visualisation
  • Hu, M. & Liu, B. (2004) Mining and summarizing customer reviews. KDD
  • Liu, B. (2020) Sentiment Analysis: Mining Opinions, Sentiments, and Emotions, 2nd ed.
  • The Inquiry Series, C10 — spaCy — Building NLP Pipelines for Qualitative Research

Also known as

  • Entity-centric aggregation
  • Mention aggregation across corpus

Categories

  • Analytics & Behaviour
  • Computational & Machine Learning