Less data, More precisely
Inference-based NLU engine
Natural Language Understanding Engine is a chatbot core technology that provides a response to a request by identifying the user's intent when the user inputs a natural language composed of everyday terms as text.
In the case of general rule-based chatbots, the chatbot construction is time-consuming and expensive and difficult to manage because it requires the user to input as many expressions as possible in advance in order to handle various user utterances.
Xinapse's NLU engine is equipped with an inference function designed in a general-specific way based on ambiguous Korean syntax and form. As a result, Xinapse's chatbots can accurately identify intents and provide the right answers with relatively little data compared to regular rule-based chatbots.
Algorithm rather than Intuition
Xinapse's Topic modeling combines Latent Dirichlet Allocation (LDA) and its own technology to infer the relationship between topics and topics in big data and visualize these relationships.
In the conventional method of human processing, the person who analyzes had to have expertise in the relevant field, but Topic modeling is possible even without domain-specific knowledge because the algorithm determines.
Development can be promoted based on accurate data rather than intuition, so it can focus on core services that customers really need, and use time and resources efficiently, rather than meaningless services.
More Efficient, More Accurate
Companies and institutions are interested in using unstructured data to overcome the limitations of structured data. However, data labeling is not automated and the efficiency and utilization is reduced.
Xinapse has advanced the labeling operations of manual handwriting with Auto-labeling using Semi-supervised Learning to label all of your data even if you only label part of your data. This improves the efficiency and accuracy of unstructured big data analytics.
Auto-labeling, which automates the preprocessing of big data, can dramatically reduce errors, time, and costs caused by manual handwriting.
Context between sentences
Sentence embedding is a very important technology in the field of natural language processing, and AI learns the meaning of context and classifies classes by continuously learning the semantic relation between sentences.
Xinapse uses Facenet technology to classify the nature of the document data and calculate the sentences in the document as coordinates so that artificial intelligence can recognize them. In addition, the triplet loss technology is used to calculate angles and distances between coordinates and artificial intelligence classifies data classes by continuously adjusting angles and distances between coordinates through iterative learning.
What do you want to know?
Machine Reading Comprehension
MRC is an AI-based technology in which AI reads documents like humans, understands the context of a given problem in natural language, and actively finds and answers a given question.
In the existing search system, when a user enters a question, all documents containing the keyword in the question are found. In this case, users has to read every document to find the answer. But MRC can immediately give you the answer to the question.
By applying MRC, you can improve the performance of your existing homepage or groupware to find information efficiently and increase the level of information sharing.
Guess who talk to you
By using Mel finter bank algorithm and deep learning to extract the features of speech, it is possible to synthesize natural speech in less time with less voice data.
• Technological advancement creates higher quality voice content than other technologies
• System optimization speeds up voice generation(Based on the same system, Google's open technology costs about 5 seconds and our technology costs about 1 second)
• Creates high-quality content without post-processing
• Acquire celebrity voice through partnership with the entertainment companies
• Rapid generation by minimizing pre- and post-processing in terms of voice model generation