We are conducting research and development of small-data fine-grained analysis technology, that extracts information required in a timely decision without being misled by fake information and without requiring a large learning data.

Many industries define their security standards these days. Criteria consisting of more than 1000 requirements and controls are not rare. On the other hand, businesses are forced to spend a great effort to examine that their security control measures (internal standards) and product security specifications meat industry standards and legal requirements.

In this research, we are developing a document analysis tool, Tact, by combining hierarchical clustering, natural language processing, topic modeling, and many other statistical analysis technologies.

The document analysis tool Tact can be applied for comparative analysis of documents, analysis of document structure, evaluation of document maturity and document deterioration, and analysis of keyword distribution in documents. Targeted documents are Japanese and English. Depending on the language, we use the different method for extracting information. Figure 1 shows the analysis results that express the degree of relevance between sentences in a document from dark to light by “profiling analysis”. If dark colored clusters or isolated dots (called noise) are formed outside the diagonal line, it seems that the document is aged and deteriorated by repeated additions and deletions, or it has not been polished sufficiently. In addition to structural analysis, documents can be associated with each other on a sentence-by-sentence basis. Figure 2 maps the frequency of occurrences of each keyword (horizontal axis) and the place of the occurrences (vertical axis) using a function called keyword distribution analysis. It can be used to identify keywords that are important in analysis.

By utilizing these document analysis functions, we have conducted to many technical consultations for guideline development and the maintenance of internal standards.

The document analysis tool Tact is not open to the public now, but we are developing hands-on exercise materials so that users who need document analysis technology can easily use Tact. Please feel free to contact us when considering technical consultation, trial of analysis tools, practice experience, etc.