Unstructured text is messy. Your process does not have to be.

One workspace takes you from raw data to reliable, reproducible results. No programming required.

Import from social feeds, inboxes, CRMs, surveys, and CSV. Deduplicate and cluster so that form letters and duplicate data does not skew your analysis.

Split the work across coders, adjudicate disagreements and replicate results. Crowdsourcing and iteration workflows are built in.

Turn your team’s coding into custom classifiers that sort by topic, sentiment or any category you define, in many languages.

Find the needles first, then sort the haystack

Build custom machine classifiers (we call them sifters) that surface the items that matter before you sort by topic or sentiment. Interactive histograms let you sample purposively at any confidence level.

  • Advanced search and sampling
  • Multilingual text mining
  • Sort by topic, sentiment or your own categories

Know whose judgement to trust

Our patented CoderRank methodology ranks every annotator and classifier against validated results, so you can weight the best coders, retrain the rest, and report reliability with confidence.

U.S. Patent 9275291-B2

Collect, clean, and analyze text data

Data scientists working on text analytics and machine-learning know cleaning data can be time consuming. Users of DiscoverText build reusable custom machine classifiers or “sifters”  to find the most (or least) relevant items before using other classifiers for sorting items into topic, sentiment, and other categories. DiscoverText combines hybrid data science methods (ex., crowdsourcing, measurement, adjudication, iteration, replication, annotator ranking) along with established e-discovery and information retrieval text analytics tools, to shorten a process that used to last weeks or months when words get sorted in spreadsheets. Our machine-learning sifters are created in hours or even just a few minutes working alone or using crowdsourcing. Academics trust DiscoverText to help them do better and more transparent scientific research resulting in scholarly publications. Legal teams use our document redaction capability to remove names, metadata, email addresses, and other sensitive information to produce Bates-stamped and spreadsheet-indexed PDF collections.

Humans and machines classify text

We have been doing this work in groups since 2005. Humans are good at some things and computers are good at others. A consistent back and forth between humans and machines increases the ability of both to learn. Our text analytics software and data science methods originate in a decade of National Science Foundation-funded research into the measurements that accelerate machine-learning. Text classification is an old, hard problem, according to no less than Plato. Our unique and proven method of adjudication creates gold standard training sets for machine-learning by ranking human annotators over time. A patented CoderRank approach is critical for ensuring accurate, reliable results when the work of humans or machines is finally evaluated. 

eDiscovery tools that work

Deduplication and automated clustering of near-duplicates gives users a high level sense of the data landscape. With Twitter data, these groupings are a roadmap to the digital footprint of viral Tweets. With public comment data, these groupings are form letters and modified forms. In large-scale surveys, duplicates and near duplicates are frequently held but independently expressed opinions among customers or employees. Our interactive machine classifier histograms allow data science teams to identify the items in a collection that add the most value when coded by humans. These text analytics tools enable purposive sampling that further accelerates the process of training machine classifiers.

Built for teams who need answers they can stand behind.

Education & Research

Code qualitative data with replicable, publishable methods. Free for students and faculty.

HR Analytics

Hear what employees actually say in surveys, exit interviews and internal feedback.

Legal

Review, redact and produce document sets faster, with defensible human-machine review.

Market Research

Turn open-ended survey answers and social chatter into clear, quantified themes.