What’s the LUS.sense
Computer program, that comprehends sense of the text and detects lies and “mistakes”.
What it can do
Program disjoints the text, reveals main sense-centers, shows their hierarchy structure and ranks them. At the same time program understands unic “likeness” of the given text/author. And accordingly it finds out lies and motivated errors.
As a result we know what the author wants to say and what – to conseal.
How it works
Program rates statistical distribution of characters of different types and groups. We proceed from the following statement: if text is the sum of various phonetic, graphical, rhythmical etc. signals, the changes in these signals dynamics will correlate with semantic changes.
1. Our program searches and compares matches between different configurations of the concentrations of selected parameters.
2. Having detected all visible matches and schemed their relationship, program builds the mode of their distortions.
3. Finally, we get a graph of the general structure of the symmetries in the text. The most significant segments of this graph are examined to find anagrams – markers of the text basic message.
The problem we solve (and how).
Our method of approach enables to solve at least two important issues:
1. Identifying a person by linguistic profile.
Analysis of linguistic profile done with LUS.sense allows – for speech – to ignore possible distortion, caused by voice modulator or software tools. Program focuses not on voice-intonation, but on combinations of phonetic patterns, characteristic for sertain person.
In case of written text LUS.sense enables to attribute the message not with individual vocabulary preferences, but with – again – combinations of phonetic patterns, which makes it possible to work with small-sized messages.
2. To reveal person main hidden thoughts.
Using the original algorithm, program finds areas of main sense concentration and defines text structure individual principles. The program also finds text segments where these principles are destroyed. And discovers anagrams present in these segments. This gives the ability a) to determine author’s true attitude to message idea, b) decode the strongest verbal markers of his internal (subconscious) speech. Actually, it enables to «hear» author’s «inner speech» main point.
Correlation of changes in the phonetic composition of the text and its rhythmic characteristics with the change of its content/meaning was discovered in the 20-ies of the last century. But no one has ever used it as a tool or developed interpretation method to analise such correlations. We have found out that these correlations – in any natural language – obey strict rules. Moreover, we formalized these rules. That’s our innovation.
Stage of the project
Barriers to overcome
1. The most important thing we need just now (pre-A stage) is to improve mathematical model, required to analyse groups with fuzzy similarities, and also to search genetic links between different distortions of these similarities.
That means: 6 labour months for mathematician, inventors and programmers.
2. Above that, to adapt LUS.sense to a certain language we’ll need a team of linguists (lexicographers, phonetisians and bilingual translators).
Financial, personnel etc security.
Sentiment Analysis (IntuView)
Kribrum (Ashmanov and Partners)
Alex Limanov – inventor,
Lucy Gurbanovsky – linguist.
Software package for the security services.
Its intended purposes:
1. To identify text author. If the program once analyses the text of a particular author from the database (not less than 1000 characters), it’ll be able to determine his authorship in short (but not less than 140 characters) messages afterwards, with a probability of 90%
2. To reveal the main idea of the text.
3. To detect hidden (unspoken/unwritten) words, including proper names.
4. To indicate text fragments with lie/error/omission.
To make English version it’ll take 8 co-workers – mathematician, 2 inventors, 2 programmers, bilingual translator, lexicographer and phonetisian – and 6 labour-months. NB: not calendar, but labour. I should also mention, that all experts will be real professionals.
— Mobile application for social media users, sort of a toy.
Its purposes are:
1. To detect hidden (unspoken/unwritten) words, proper names.
2. To indicate text fragments with lie/error/omission.
In other words, show, if your girl/boyfriend is cheating you. Or, simply tell you, what she/he is thinking about, while talking to you.
4 labour-months, 8 experts – for English version.
— Supplement to the search engines.
LUS.sense can be of great help in detecting keywords, as it purposefully looks them up in the segments of informative max. Such search not only takes less time, but also provides very accurate, efficient, structurally defined results. Our approach gives incomparably more reliable info, than present semantic and statistical methods.
4 months (in this case – calendar). 5 co-workers: mathematician, 2 inventors, 2 programmers. Cooperation with the search engine developers.
— Service for users of social networks ‘friend or foe’.
People are eager to find friends in social networks. The application, which is able to identify congenial person, seems a good idea. It’s similar to socionics, when psychologists divide people by types, due to various features.
In that case program will analyse – besides speech characteristics – some additional parameters, given in social nets. Such as: number of avatars; number of active and passive friends; musical preferences; activity per hour, etc.
3-4 calendar months
— Service for users of smart-watch / bracelet / phone.
There already exist large group of smart gadgets that combine different functions – of the watch, messenger, fitness tracker (which measures the heart rate, the number of «burnt» calories, temperature, etc.) and phone – in one device.
To this synthetic gadget we can add our own option: personal assistant (PA) with the function of situational consulting and training, thus becoming specified application.
What is situational consulting and training? What will it do?
1. Create a map – that reveals area, defining person’s state: excellent-normal-unsatisfactory. Both physical and mental. Sinse the previous states and up to the future.
2. Create a map – that reveals area «motivated errors». As and when they occurred in the past, which features most closely allow to diagnose their appearance when they are assumed to be in the future. What signs indicate that they appear in the lightweight variation (let’s call them error-light areas).
3. Prediction «motivated errors» in real time. Plus information about their genesis.
4. Create a map — that reveals coherence the communicative activity of their respective owners and psycho-physical condition. It is not only statistical, but also structural analysis, which determines the behavior and other characteristics of the various semantic correspondences. These data will be marked contacts, general — social communication of the owner, and relate them to his condition. For example: chat with a contact ‘A’, usually preceded by the appearance of a psychophysical condition of ‘N’.
The principle of data — standard for LUS.sense. The data are divided into several streams: text, biometrics, geotreker, statistics of incoming / outgoing calls, etc.
Output: Two types of recommendations from the PA — a) short-term, operational, which is updated with a predetermined frequency, and b) forecasting, with specify a lag — day, week, month, etc. PA for this must be synchronized with the calendar (planner).
Competitors in the area of security services (home-land security)
1. The software complex from IntuView: Sentiment Analysis, Docex, Name-Matcher. (And also unknown to us software used by the FBI, FSB etc.)
That’s what they do.
They find texts of potential suspects in the data, define the mood of the author in relation to the objects/subjects of text, define the case if author belongs to criminals (terrorists), they form a set of features (profile), which will help in the future to identify the author or persons like him.
Let’s try to imagine a smart terrorist who would not use the key-words, the marked words, who wouldn’t use the linguistic clichés, that are specific to the desired ethnic/religious/clan group. And, thus, this person will disappear from the field of view of the software complex.
We can identify this person by searching the same word-tokens and the same linguistic clichés, but our method, according to the anagrams and ‘hidden’ words.
2. VoiceGrid (Speech Technology Center) and others sistems voice-identify
That’s what they do.
They identify suspects by voice and photo.
The use of hardware and software for voice distortion makes it difficult to identify him. That’s why most of these biometric tools only work in combination: voice, image, fingerprinting, etc. Our method focuses not on voice / intonation, but on combinations of phonetic patterns, characteristic for sertain person. This «fingerprint» is almost impossible to distort.