Fri. Mar 29th, 2024

Describing Technology-Assisted Review

Technology-Assisted Review: An Overview #

Technology-Assisted Review (TAR) is a widely used method for managing large volumes of electronic data in the electronic discovery process. TAR leverages machine learning algorithms to analyze electronic data and identify relevant information, reducing the time and cost of the review phase. There are several different models of TAR, including TAR 1.0, TAR 2.0, TAR 3.0 and TAR 4.0, each offering a unique approach to the TAR process.

TAR 1.0 – Simple Active Learning #

TAR 1.0, or Simple Active Learning, involves moving electronic data into a TAR (predictive coding) application, establishing a random control set of electronic data, and reviewing and coding the control set for relevance. The TAR tool is trained by continuing to establish, review, and code the control set until a sufficient number of relevant documents are identified. A seed set of electronic data is then established using random and/or judgmental sampling, and the machine learning algorithm is applied to suggest the best learning documents. This process is repeated until stabilization occurs, based on the accuracy of relevance prediction for the documents in the control set. The learning algorithm is then applied to categorize or rank all documents, and all documents categorized as relevant or ranked above a cut-off score are prepared for review. The TAR process is then validated.

TAR 1.0 – Simple Passive Learning #

TAR 1.0, or Simple Passive Learning, involves moving electronic data into a TAR (predictive coding) application, establishing a seed set of electronic data, and reviewing and coding the seed set for relevance. The machine learning algorithm is then applied to evaluate whether the documents are relevant. The training process is continued with a larger seed set until the training effectiveness is deemed sufficient. The learning algorithm is then applied to categorize or rank all documents, and all documents categorized as relevant or ranked above a cut-off score are prepared for review. The TAR process is then validated.

TAR 2.0 – Continuous Active Learning #

TAR 2.0, or Continuous Active Learning, involves moving electronic data into a TAR (predictive coding) application, establishing a seed set of relevant documents using judgemental sampling, and applying the machine learning algorithm to the collection to suggest the most likely responsive documents. The suggested documents are reviewed and feedback is provided to the machine learning algorithm, which is then repeated against the collection until few, if any, suggested documents are relevant. All documents categorized as relevant are then prepared for review. The TAR process is then validated.

TAR 3.0 – Cluster-Centric CAL #

TAR 3.0, or Cluster-Centric CAL, involves forming conceptual clusters of the collection, establishing a seed set of relevant documents using judgemental sampling, and applying the machine learning algorithm to the cluster centers and sorting by relevance score. A small number of cluster centers with the highest relevance score are reviewed and the machine learning algorithm is repeated until few relevant clusters remain. The machine learning algorithm is then applied against the collection, and it is determined whether to produce documents without review, produce documents with high relevance scores without review and perform standard CAL on the remainder of documents, or review all documents for production using standard CAL. The TAR process is then validated.

TAR 4.0 – Hybrid Multimodel IST Predictive Coding #

TAR 4.0, also known as Hybrid Multimodal IST (Intelligently Spaced Training) Predictive Coding, is a comprehensive approach to technology-assisted review in electronic discovery. The process begins with defining the scope of discovery, relevance, and related review procedures through ESI communications.

The first step is to conduct a Multimodal Early Case Assessment (ECA) to gain a better understanding of the electronic data involved in the case. A random sample is taken to determine the prevalence of relevant information.

Next, the Training Select process is initiated. This iterative process involves determining which documents to use to train the machine by using a combination of human and machine functions. The AI Document Ranking process is then used to appropriately rank the documents based on their relevance.

The Multimodal Review process is initiated to find new or irrelevant documents for the next round of training. This iterative process uses multiple techniques, including search by high-ranked documents, mid-ranked uncertain documents, random and judgmental sampling, and ad hoc searches not based on document ranking.

Once the training process is complete, Zero Error Numerics (ZEN) quality assurance tests are conducted to validate the TAR (Predictive Coding) process. Finally, the relevant documents are reviewed again and produced in a phased production process.

TAR 4.0 provides a comprehensive and flexible approach to technology-assisted review, using a combination of human and machine functions to identify relevant electronic data as efficiently and accurately as possible.

TAR – A Valuable Tool #

TAR is a valuable tool for managing large volumes of electronic data in the electronic discovery process. Each model of TAR offers a unique approach to the TAR process and provides a structured and systematic method for reducing the time and cost of the review phase while ensuring that all relevant information is identified and reviewed.

 

Powered by BetterDocs