Subject Matter Expert Refined Topic Models

Human-Assisted Modeling (HAM), Subject Matter Expert Refined Models (SMERM), and Subject Matter Expert Refined Topic (SMERT) Models.

The Need

Currently, unstructured data represents up to 80% of the data within an organization. This means the traditional data, such as sales figures or other statistics, are separated into different documents without a meaningful and effective way to aggregate the data. Due to this there is now an opportunity in the market for a technology that is able to incorporate "high level" critiques and is able to greatly improve on current models. This would impact companies on all levels of the data spectrum, from small companies to large corporations.

The Technology

The Ohio State University researchers, led by Dr. Theodore Allen, developed an improved way of performing statistical and computer science analysis with human-assisted modeling (HAM). Procedures often involve traditional data-like sales figures or freestyle text from reports. The traditional form of modeling leaves out the analyst's knowledge and experience. This new framework includes many types of data or "high-level" critiques that improve models. HAM can be combined with topic models and unigram models to create subject matter expert refined topics (SMERT) and subject matter expert refined models (SMERM), respectively. This would allow for large sets of unstructured data to be analyzed quickly and accurately. The market for unstructured data analysis is quickly growing as companies are beginning to search for value in their large data sets.

Commercial Applications

  • Data management
  • Manufacturing
  • Sales
  • Web companies

Benefits/ Advantages

  • More accurate and interpretable summaries
  • Lower cost in expert time/ effort
  • Could help identify the cause of quality problems, prioritize issues, and accurately predict sales

Loading icon