Refine, combine, or compare datasets

Refine a query when matched sentences show false positives, false negatives, or ambiguous language. Combine or compare datasets when an analysis needs more than one concept.

Refine a query

Start from the existing query, results, and matched sentences. Identify terms to add, remove, quote, or exclude.

Common changes include:

• Add synonyms to improve recall.
• Add exact phrases to improve precision.
• Add plural variants manually.
• Add AND NOT exclusions for false positives.
• Use parentheses to make precedence explicit.

Combine datasets

Combine two existing datasets with AND, OR, or AND NOT from the dataset page. See My Datasets for where the actions live.

Use AND when a measure should require both concepts. Use OR when either concept should count. Use AND NOT when the first dataset should exclude a known false-positive concept represented by the second.

Combinations operate on the matched sentences of the two datasets:

• AND keeps sentences matched by both datasets.
• OR keeps sentences matched by either dataset, including sentences that only the second dataset matched.
• AND NOT keeps sentences from the first dataset that the second dataset did not match.

The combination creates a new dataset and counts as one search. It recomputes the call-level measures from the combined sentence set, so exposure, risk, and sentiment follow the same measure definitions as other datasets. The new dataset keeps zero rows for the calls covered by the combination: with OR, calls from either dataset; with AND, calls present in both; with AND NOT, calls from the first dataset.

For complex queries with large keyword groups, it can be clearer to run separate searches and combine the resulting datasets than to maintain one long query. Keep the combination logic in the reproducibility record.

Compare datasets

A comparison takes a left and right dataset and creates a new dataset. Choose how to express the difference: net difference, absolute difference, ratio, or percent difference. Creating the comparison counts as one search.

Use a comparison to check how a query refinement changed a measure, for example how many matches an added exclusion removed and for which firms. Matched sentences are labeled by source, so you can inspect which sentences belong to only one dataset.

Compare versions before finalizing

Whether you refine, combine, or compare, look at both aggregate output and matched sentences before choosing the final dataset.

Record the final definition per the reproducibility checklist: the query string, date range, filters, adjacent-sentence behavior, selected measures, combination or comparison logic, and export date.