As we already wrote in our first article , we in Just AI specialize in artificial intelligence technologies for understanding the natural language (NLU). We have our own platform for developing colloquial assistants, both text and voice. The platform has advanced functionality for solving NLP-tasks (log clustering, search for semantically close phrases, etc.) and its own development language (Just AI DSL). Using DSL, the team of linguistic developers writes code for chatbot scripts, integration modules for communication with external systems, and autotests.
 
 
In this article we will talk ...

By implementing modern information systems, large companies expect to make decisions faster, discover hidden reserves and opportunities for the business, analyze the accumulated experience and build forecasts based on the identified regularities. However, the real return on information systems is often much lower, and implementation time and costs are higher than expected. The reasons can be a great many, including those related to ineffective management, the human factor, the outdated infrastructure.
 
 
Essential shortcomings are also in the information systems themselves. In this article, I ...

We will talk about the use of fashionable "Word embedding" not exactly for the purpose - namely to correct typos (strictly speaking, and mistakes too, but we assume that people are literate and sealed). The habr was pretty close article , but there will be a little more about the other.
 
 
 
Visualization of Word2Vec model, received by the student. She studied on the "Lord of the Rings". Clearly something in a black dialect.
 
article , which is well told about this. In order not to repeat, but not to chase the reader through the links ...

In the first part of Article we started talking about a new class of high-level domain models called conceptual. Unlike other similar models in conceptual models, the links between concepts are concepts themselves, and the model is built on the basis of identifying and describing the abstractions that served as the formation (definition) of the concepts of the subject domain. This allows end users to build and update domain models through simple and natural operations of creating, modifying and removing concepts and their entities.
 
 
Here, in the second part, let's talk about how a fully ...

The notion of space and time (space)
 
In the construction of the concept of space and time, our notion of ourselves as an observer participates. We represent ourselves as an observer who realizes his position in the Absolute Space and is synchronized with the Absolute Time, determines the size of the studied objects and the duration of the time intervals studied. In this case, the spatial dimensions and time intervals that a person is able to recognize lie in rather narrow intervals beyond which human perception is powerless to perceive anything. To step into another time scale or space, imagination ...

This work is a continuation of what was said earlier in Article " Neural networks, genetic algorithms and so on Myths and reality. Version II » . In most of the articles devoted to the analysis of texts that the author has managed to study, the analysis of a text is mainly understood to be two completely practical tasks, connected either with the extraction of a context or the translation of a text from one language to another. In the first case, as a rule, it is either about "cleaning" the analyzed content and comparing a section of the text to the standard in accordance with a pre-defined ...

At school, I needed tremendous efforts to understand the concept of Cantor about the multitude of power continuum. But then I realized that I did not understand anything, but only learned the rules of working with such objects. In the field of understanding, there was a white spot, so it remained white. Since then, I have repeatedly returned to this issue until I became acquainted with statistical physics and sopromat.
 
 
In these sciences, the concepts of matter and the concept of state were very well defined. It was said that the definition of a substance requires a volume of a minimum size...

Any model has limited accuracy. The more accurate the model needs to be built, the more information it will have to store. If there is an opportunity to collapse an array of data by some of the criteria, then this convolution allows us to drastically reduce the amount of stored information. However, such a convolution is not modeled by the standard modeling methods, because it requires the modeling of statements simultaneously and relative to the sets of objects, rather than objects of these sets. In fact, we need a tool for modeling both first-order predicates and second-order predicates.
 
I ...

It's been a long time since the publication of my first article on the topic of natural language processing. I continued to actively explore this topic, every day discovering something new.
 
Today I would like to talk about one of the ways to classify search queries, into separate categories using a neural network on Keras. The sphere of cars was chosen as the subject area of ​​inquiries.
 
The basis was a data set of ~ 32000 search queries, classified into 14 classes: Autoinsurance, Auto Insurance, Driving License, Complaints, Entry to the State Traffic Safety Inspectorate, Record at the MADI...