Machine reading comprehension Engine MRC
Saltlux’s AI Suite machine reading comprehension (MRC) engine searches relevant documents for an answer to the inquiry and finds it. AI Suite’s machine reading comprehension engine, in particular, compensates for the drawbacks of prior technology, in which papers were fed to find answers to queries, by utilizing Saltlux’s numerous knowledge resources. As a result, it delivers an open QA type service in which answers are found without the need for document input. It can also be used to extract information from a document.
- Providing optimal machine reading comprehension results by situation
The machine reading comprehension engine is referred to by multi-learning models. You can select the optimum learning model that varies from question to question, which is far superior to other machine-reading comprehension engines that use a single learning model.
- Can be used as Open QA
Saltlux’s machine reading comprehension engine for QA makes use of a variety of knowledge resources accumulated over the last 20 years. A user can enter a query without entering a document and acquire an answer by automatically pulling a document or paragraph from a knowledge resource that it already possesses and entering it into a machine reading comprehension engine, thus improving the Q&A service.
< Machine reading comprehension through automatic document search/input >
- Service expansion in connection with AI Suite engines
The machine reading comprehension engine can enhance its usefulness by linking with other engines in AI Suite. Simple Q&A services are possible during dialogue in connection with the dialogue processing engine, and in conjunction with the Q&A technology. It provides an ensemble of deep Q/A, thus improving the quality of answers. In addition, the machine reading comprehension engine can be used when the knowledge automatic extraction engine extracts knowledge from structured/unstructured documents or table data.
Main features and specifications
- Machine reading comprehension service
The machine reading comprehension engine does not provide a separate management tool but a system installation type. The API that can access the machine reading comprehension service operates when the system is installed and run. Users can use the API to implement various services with the engine.
< Machine reading comprehension test and visualization screen >
- Learning data management
It provides a feature in which learning/evaluation data can be created and managed for machine learning. MRC’s learning data should be constructed as a pair of fingerprints, questions, and answers. Especially, it is not easy to build learning data because data should include the index information located in the corresponding fingerprint. The learning data management tool provided by the machine reading comprehension engine can easily construct data because it generates questions about learning and evaluating documents, then selects appropriate answers by a drag-and-drop method.
- Learning Management
This function allows the MRC engine to perform model learning in a web management tool. It provides UI to help intuitively understand and manage the workflow, from selecting a data set to learn, setting learning options, to evaluating learning results. You can create a learning model by choosing your own learning algorithm and setting the parameter values. Learning progress can be visualized and checked, and learning history is also managed.
- Model validation and distribution
The trained model should be reflected in the machine-reading comprehension engine of the service. It provides a feature in which target service applications that deploy a model are registered and managed while models can be distributed, backed up, or canceled with a management server.