KGB mapping knowledge to play smart technology featured in Q direction

Intelligent natural language question and answer that is given issue, by semantic understanding of the issues and resolve, then use to query the knowledge base, reasoning the answer. And dialogue systems, different interactive dialogue dialogue robot, intelligent questions and answers with the following features:
Answer: The answer is the answer to the entity or entities in the relationship between knowledge of, or no-answer (that is, the problem can not find the answer in the KB), of course, this is not necessarily the only answer, such as what Chinese cities. The system will reply dialogue is a natural language sentence, sometimes even to consider Context. Evaluation criteria: recall (Recall), precision ratio (Precision), F1-Score. The evaluation criteria of the dialogue system to manual evaluation based, as well as BLEU and Perplexity.
Q mainstream approach knowledge of
semantic parser (Semantic Parsing): This method is a partial linguistic approach, the main idea is to form natural language into a series of formal logic (logic form), by the logical form of bottom up analysis, to obtain a logical form may be expressed throughout the semantic problems, query in the knowledge base by respective query (similar to lambda-Caculus), so the answer.
Information extraction (Information Extraction): This class method by extracting problems entity, the entity can be obtained in the center of the node by querying the knowledge base entity subgraph in the knowledge base, each of the sub-graph node or edge may be as a candidate answers questions by observing certain rules or templates based on information extraction, feature vectors to get the problem, build a classifier for screening candidate answers questions by entering the feature vectors to get the final answer.
Vector model (Vector Modeling): The ideas and methods of information extraction and rather close to obtaining candidates to answer the questions, the questions and the candidate answers are mapped to a distributed expression (Distributed Embedding), distributed by the expression of the training data training, making the score expression vector problem and the correct answer is (usually in the form of dot) can be screened according to the scores expression vector and the expression of candidate answers questions as high as possible after the completion of training model, get the final answer.
KGB mapping knowledge now perform the following functions: 1 document parsing: KGB knowledge map engine, and can be easily parsed version of the document in multiple formats: TXT, DOC, EXCEL, PPT , PDF, XML and so on. In particular PDF file, output can be directly resolved as word file format, file important information in the table and text format retention. For image information, OCR text can automatically identify and extract the information in the picture. 2. knowledge extraction: KGB knowledge map engine, adaptive recognition from structured forms and unstructured text and extract key knowledge (subject, object, time, place, amount, terms, etc.), the accuracy rate of up to 90 percent, rapid generation of knowledge. 3, knowledge association: KGB knowledge map knowledge associated with dig engine, a link to a knowledge entity with full knowledge of the facts of significance. And has a strong knowledge reasoning ability, reasoning and conclusions of the implicit knowledge, knowledgeable map. 4, knowledge, more experience: KGB knowledge map knowledge processing plant capable of checking the quality of intelligence, including knowledge of a variety of errors and conflicts intelligent automatic verification and correction, the more accurate knowledge engineer knowledge check to ensure the accuracy of the knowledge map .
In the application industry, KGB knowledge map has the following characteristics: 1, cross-cutting scalable: map processing plants have a common knowledge map construction engine. Knowledge extraction, knowledge associated with the quality of the verification process does not rely on specific business knowledge, combined with the knowledge map construction needs of the user, the user can quickly build the knowledge map of the field. 2, intelligent verification of the quality of knowledge: knowledge map processing plant intelligent verification and validation of knowledge on a variety of errors and conflict, and the knowledge base for real-time automatic updates to ensure the accuracy of the knowledge map. 3, human-computer services: human knowledge map processing plant composition: 90% + 10% artificial machine, only need to provide corpus, you can quickly get corresponding knowledge map construction results.