EKG as a Knowledge Management System basis

In most KMS the top level of the knowledge structure is the familiar to all PC users hierarchy of "folders" and documents, so modeling of the knowledge structure begins with building a single hierarchy for navigation through them. In practice, the knowledge structure of almost any company is so complex that it does not fit into a single hierarchy. Mixing different bases of knowledge classification in one tree leads to the loss of its logical structure and complicates navigation.

To get around this limitation, knowledge items are beginning to be tagged with tags - randomly selected words, each of which can stand for one of the main business objects or a term from the business glossary. Tagging can be automated using machine learning algorithms. The tags themselves are usually not structured in any way and form a simple list.

To build an EKG, first an ontology model of the subject area is created, which describes the types of business objects, their possible relationships and properties. For some industries (e.g. banking) such models exist ready-made, for others they need to be developed during the project.

A business glossary (e.g., built according to the SKOS model), which presents the terms used to refer to business objects, properties, and relationships, can also be part of the model.

Then catalogs of the main business objects are built, with which knowledge elements can be associated - these can be assets, customers, employees, etc. If the company has already implemented an MDM system, it can become a source of filling such catalogs.

Multiple hierarchical facets can be created in the model to categorize knowledge items.

Filling information in a classic KMS is most often done manually. The expert needs to find knowledge sources for each issue in the organization, process them, extract the necessary information, place it in the appropriate section of the KMS and provide search metadata, including tags.

Complex project practices are often used to organize the filling process: organizing working groups and committees, conducting interviews and surveys, and coordinating the information entered into the KMS. All of this consumes expensive work time of the most competent employees of the organization.

To populate the EKG, structured and unstructured data existing in the organization are primarily used. Only information that did not previously exist in electronic form is entered manually. To extract structured data, adapters are created that extract information from corporate systems using web services or by accessing copies of their databases. To operate the adapters, rules are created in the ontology model that describe the correspondence of model elements to data structure elements in the sources (paradigm OBDA, Ontology-based data access). Extracted data can be materialized in the EKG repository or retrieved from the source on request.

To index documents, a crawler is created that traverses all file stores. Along with the usual full-text index, it identifies the business objects mentioned in each document and links them.

Thus, the EKG creates a virtual representation of each business object, enriched with all available information about it. This representation can be considered as a kind of digital shadow of the business object.

Incorrect or irrelevant data contained in a KMS can cause significant damage if used in decision-making. The vast majority of classic KMS systems do not have any tools to control the relevance and verification of data - these tasks are completely left to human experts. With rapid changes in the business environment, it is extremely difficult to control the relevance of all information in a KMS.

The "human factor" cannot be excluded as a source of errors when entering data into KMS.

Ontologies allow to record in machine-readable form the rules for validating information and the rules for drawing logical conclusions from the available facts. Such rules, constructed by the analyst, allow to automate the cleaning and enrichment of EKG content.

Any information entering the EKG retains a link to its source. When viewing any facts in the EKG, it is always possible to establish exactly when and from where they were obtained. The use of structured data, indexing of primary source documents reduces the probability of incorrect information entering the EKG. All this helps to increase the credibility of the information contained in the EKG.

To find the desired information in a classic KMS, the user uses content tree navigation, tag-based search, and full-text search. Hyperlinks between content elements and widgets such as "Related articles" are also used, the content of which can be built using machine learning algorithms.

The knowledge item found is usually a text article that contains a pre-prepared answer to a specific question or a description of a particular situation.

The user can build complex queries to EKG including several interrelated conditions, for example: "In which factoring transactions with clients of industry N for the amount of more than X million $ was risk assessment according to M methodology applied? " It is important that EKG gives precise answers to such questions based on data, and not just selecting documents in which the relevant words occur. It is possible to use the results of some queries as a set of input parameters for others.

A found knowledge item is an object that has many properties and relationships with other objects. Objects are viewed and navigated in a wiki-like interface. The relationship graph itself can also be visualized - this gives an opportunity to quickly assess the full picture of information available for each object.

Auxiliary functions of EKG are searching for relations between objects, searching for similar objects.

The process of maintaining a KMS differs little from the process of its initial content and is limited to adding new materials, correcting existing ones, and removing irrelevant ones. Significant expert labor is spent on support.

EKG automatically collects up-to-date information from source systems and documents and eliminates outdated data. Support is to extend the model as new knowledge domains are added to the EKG perimeter or as business processes change.

EKG can respond quickly to changes in requirements, processes, and source data structures because only changes to the model, not program code, need to be made to reflect them in the system.

Innovative features of KMS are most often reduced to attempts to apply neural networks to improve content search practices. In general, this segment is characterized by rather slow development of both the paradigm as a whole and the functionality of specific products.

The development of Natural Language Understanding (NLU) technologies can be used to enrich EKG content with facts extracted from text, thus increasing the accessibility of information from unstructured sources. They can be used to enrich the content of the EKG with facts extracted from the text, thus increasing the accessibility of information from unstructured sources. They will also improve the user experience of interacting with the system: the ability to ask a question in plain language, rather than constructing a query in a search form, will speed up information retrieval. A chatbot or dialog assistant using information from the EKG will allow the user to directly receive answers to questions that are important to the user, rather than having to sort through an array of query results.