Deep Q/A Engine

AI SUITE

    Deep Q/A Engine

    AI Suite’s Deep Q/A engine is a system in which the best answers are found and presented for user’s question from the accumulated knowledge by training it to learn various knowledge. By integrating knowledge base (KBQA), information retrieval based question and answer (IRQA), machine answering (MRCQA), and counseling and dialogue history learning based question and answer (DLQA), the Deep Q/A engine finds and presents the best response to the user’s query.

    ai12

    < Deep Q/Aprocessing summary >

    Main Features

    • Massive and accurate knowledge graph-based Q&ADeep

    Q/A engines answer questions using knowledge-level data, not just information-level data. It not only automatically extracts and learns knowledge from the data, but also infers new knowledge hidden from the learned knowledge and uses it in the Q&A./dd>

    •  Flexible Q&A using an ensemble technology

    In order to respond to various types of user queries, each question and answer module is composed in the form of ensembles in which optimal answers can be presented according to the query type. In addition, in order to respond to inquiries requiring real-time information such as weather and stock prices, real-time information can be obtained in connection with external APIs and used for Q&A.

    • Quickly applicable in various domains

    It provides a platform for managing information such as dictionaries, knowledge, and indexes to handle Q&A by generating, learning, and inferring new knowledge according to the domain to which the deep QA engine is to be applied. The platform makes it easy to build knowledge of specific domains and to implement Q&A in those domains.

    Main features and specifications

    Deep Q/A engines apply various QA methods to find the correct answer based on the natural language understanding of the engine. At this time, it understands natural language and searches for answers by referring to the knowledge graph. By providing separate management tools, knowledge building, management, and service monitoring for deep query responses can be performed according to domain.

    ai13

    < Deep Q/A engine system structure >

    • Ensemble Deep Q/A

    Deep Q/A engine provides an ensemble Deep Q/A in which answers are explored in a variety of methods, including knowledge base, semantic search, machine learning, and deep learning, based on high-precision natural language results.

    ① Knowledge Based Question Answering, KBQA

    It’s a method of Identifying the core meaning of a question and then querying the Knowledge Graph repository to search for answers. You can structure your knowledge and build a knowledge base that matches your knowledge structure, so you can query the correct answers to your questions. In addition, you can change and maintain knowledge easily and keep improving the quality of answers.

    ai14

    < Knowledge-based Q&A processing summary >

    ② Information Retrieval Question Answering, IRQA

    IRQA of the deep QA engine is divided into the processing method based on the indexing technology and the processing method based on the embedding technology using the deep artificial neural network. Index-based IRQA is the most common question-and-answer approach that builds predicted question-answer data ahead of time, searching for and responding to questions similar to user’s.

    ai15

    < Information search-based Q&A processing summary >

    ③ Machine Reading Comprehension Question Answering, MRCQA

    It’s a way, without human involvement, for machines to learn to read documents and find and present answers to questions without having to build knowledge. It provides answers to questions by combining information retrieval methods that navigate the target document with machine-reading comprehension (MRC) methods that find answers in the document.

    ai16

    < Machine reading comprehension-based Q&A processing summary >

    ④ Dialog Learning based Question Answering, DLQA

    This method automatically generates answers to questions based on deep learning models that have learned dialogue data such as actual counseling, dialogue history, and question and answer history. You can build high quality learning data in large quantities for learning and improve quality through continuous learning and evaluation. The answer generated by the deep learning model can be directly provided, or it can be used as an ensemble Q&A process that is merged with the KBQA, IRQA, etc. service.

    ai17

    < Dialogue learning-based Q&A processing summary >

    • Knowledge and language resource building management

    Deep Q/A engine provides an ensemble Deep Q/A in which answers are explored in a variety of methods, including knowledge base, semantic search, machine learning, and deep learning, based on high-precision natural language results.

    ① Knowledge building

    The knowledge curation management feature allows you to create a task for knowledge building targets, designate workers (inspectors and curators), and assign it to them. You can search the status of work by knowledge building workers, and manage the quality of the building by verifying and reflecting the work results.

    ② Language resource (dictionary) management

    It provides the ability to add dictionaries or edit dictionary entries used by the deep query engine. Dictionaries are primarily used by the Natural Language Understanding Engine and KBQA Engine to identify the meaning of questions and to match it with knowledge information.

    ③ Knowledge Base management

    You can query and manage the data of the built knowledge base. You can query the knowledge graph schema, and search class, property, and instance data separately. It provides data visualization along with detailed information on knowledge data, allowing you to explore the associated knowledge structures and provide query tools that allow you to run SPARQL queries directly.

    • Q&A quality control

    We provide quality management features that are used to monitor services through the Deep Q/A engine and to continuously improve quality.

    ① Evaluation management

    The quality evaluation process of the Deep Q/A engine involves creating an assessment set consisting of questions and answers, and periodically evaluating the QA engine’s percent correct for that assessment set. The Q&A tool provides such evaluation data management and scoring management features to enable continuous quality management.

    ② Query test

    It provides a query test feature that lets you enter specific questions to see what answers are actually provided by the Deep Q/A engine. The query test results provide log information that occurs at each stage of the Q&A processes, so that you can quickly identify why the question was not answered and correct it.

    •  Q&A service management

    It provides the configuration of Deep Q/A engine for Q&A service, system usage, and various other operation management features.

    ① Project Management

    The Deep Q/A engine can construct independent knowledge bases and question-and-answer systems based on target domain or service type. The Q&A management tool provides a feature to set up and manage these individual systems as projects. The project management function enables the configuration and operation of the Deep Q/A engine and other linked engines.

    ② Q&A history management

    Deep Q/A engine provides a history management feature to store, manage and search the service history. You can continuously improve the quality of service by checking the service status and selecting and analyzing unanswered questions for future knowledge building or debugging.

    ③ System monitoring

    It provides a hardware-based monitoring of related systems, including a Deep Q/A engine.

    ④ Other operation management (emergency answer management)

    When operating a service, there may be an answer that needs to be regarded as an exception before being processed by the Deep Q/A engine.

    If you need to convert questions during the preprocessing stage or provide a predefined answer, you can take advantage of emergency response management.

    In addition, it provides additional functions such as user management, notification management, and data management, which are required for service operation.

    Main engine screen

    Untitled web 8
    Untitled web 8