RU2688758C1 - Method and system for arranging dialogue with user in user-friendly channel - Google Patents

Abstract

FIELD: computer equipment.SUBSTANCE: method for arranging dialogue with a user in a data transmission channel includes: acquiring user input data; preprocessing user input by breaking it down into sentences and words, wherein user spelling errors are corrected by using typos correction module; performing lemmatization of each word from user input; generating word dependency structure in user input by using syntax parser; generating vector model of user input words; classification by means of module for dialogue with user of part of user input to generate response; generating a user response for display in a graphical user interface; providing, through the processor, one response to the recognized user data input.EFFECT: faster user service.10 cl, 6 dwg

Description

TECHNICAL FIELD

[001] The present technical solution generally relates to automated information systems and, more specifically, to technology for intelligently and projectively generating dialog messages for building a dialogue with a user.

BACKGROUND

[002] The technical solutions described in this section may be implemented, but not necessarily represent methods and systems that have been previously considered or implemented. Therefore, unless otherwise indicated, it should not be assumed that any of the technical solutions described in this section qualify as prior art only because of their inclusion in this section.

[003] A chat bot as an automated information system is a computer-based agent that has a human-adapted graphical interface for accessing and managing information. Traditionally, a chat bot can interact with natural language users to simulate intelligent communication and provide personal assistance to users. For example, users can ask such questions in the chatbot as, for example, “Where is the nearest hotel located?” Or “What is the weather now?”, And receive relevant answers. Users can also request a chat bot to perform certain functions, including, for example, creating emails, making phone calls, searching for information, receiving data, redirecting user requests, sending a user, providing notes and reminders, and so on. Chat bots and personal digital assistant systems are widely used and are a great help for computer users and are especially useful for owners of portable electronic devices such as smartphones, tablet computers, game consoles, and so on.

[004] The term "chat bot" can also be known and used as a "conversational conversational system", "interactive system", "communication agent", "robot-interlocutor", "bot-interlocutor", "chat agent", " digital personal assistant / agent, automated online assistant and so on. All of these terms fall within the scope of this technical description.

[005] Essentially, chat bots users can ask a large number of different questions and request a wide range of information related to world and local news, weather, email content, calendar appointments, scheduled events and any other searchable content. A chat bot can be useful not only for accessing certain information, but also for generating content, planning events, writing emails, navigating, and much more. On the other hand, it is often difficult for users to understand what type of information can be requested through a chat bot at a specific point in time. For example, novice users can hardly understand or know the principles of a chat bot or its specific functionality. Users may misunderstand that several features of a chat bot application can very often be used to solve their daily tasks and needs. Therefore, there is still a need to develop chat bots and, in particular, there is a need to improve the interface of human interaction with the chat bot.

[006] In financial-credit institutions, the load on call centers is constantly growing, with the number of calls and conversations with operators amounting to millions per year. Most of the questions are predictable and do not require a unique answer, and can be closed in real time automatically.

ESSENCE OF TECHNICAL SOLUTION

[007] This technical solution is aimed at eliminating the disadvantages inherent in existing solutions known from the prior art.

[008] A technical problem (or technical problem) in this technical solution is the formation of a dialogue with the user, allowing to accurately answer the user's questions.

[009] The technical result, which manifests itself in solving the above problem, is to increase the speed of user service.

[0010] An additional technical result is to reduce the computational load on the call center information system by reducing the number of user calls and the number of conversations in the support service.

[0011] This technical result is achieved due to the implementation of the method for building a dialogue with the user in a user-friendly channel, in which the user input is received by means of a processor functionally connected to the database; pre-process user input by splitting it into sentences and words, and correct user spelling errors by using the typo correction module; perform lemmatization of each word from user input; form the structure of the dependencies of words from each other in the user input by using the syntax parser; form a vector model of user input words; classifying by means of the dialogue with the user at least part of the user input to generate a response; provide by the processor at least one response to the recognized user input.

[0012] In some embodiments, when the user input is pre-processed, the typos in the user input are further corrected by the typo correction submodule.

[0013] In some embodiments, when the user input is preprocessed, the tokenization of numerals from the user input is additionally performed.

[0014] In some embodiments, the text parser uses recurrent neural networks to analyze user text.

[0015] In some embodiments, when classifying user input, the dialog module uses a slot filling algorithm.

[0016] In some embodiments, when classifying by means of a dialogue module with a user of user input, linear dialog scripts or non-linear dialog scripts are defined.

[0017] In some embodiments, the CBOW and / or Skip-gram model is used in generating a user input vector pattern.

[0018] In some embodiments, when generating a response for a user to display in the graphical user interface, determine the class to which the user input relates.

[0019] In some embodiments, when generating a response for a user to display in the graphical user interface in the absence of a response in the dialogue, switch it to an operator.

BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The features and advantages of the present invention will become apparent from the following detailed description of the invention and the accompanying drawings, in which:

[0021] FIG. 1 shows an example of a system implementation for building a dialogue with a user in a user-friendly channel;

[0022] FIG. 2 shows an example of the implementation of the text preprocessing module;

[0023] FIG. 3 shows an example of the implementation of the transition of the phrase from one to another according to the approach of the distance of movement of words.

[0024] FIG. 4 shows an example implementation of a method for arranging a dialogue with a user in a user-friendly channel.

[0025] FIG. Figure 5 shows an example of the implementation of the word2vec approach, which allows to evaluate the semantic proximity of words.

[0026] FIG. 6 is an example of a system implementation for building up a dialogue with a user in a user-friendly channel, implemented through a set of computational components.

DETAILED DESCRIPTION OF THE INVENTION

[0027] This technical solution can be implemented on a computer, in the form of an automated information system (AIS) or computer-readable media containing instructions for performing the above method.

[0028] The technical solution may be implemented as a distributed computer system.

[0029] In this solution, a system means a computer system, a computer (electronic computer), a CNC (numerical control), a PLC (programmable logic controller), computerized control systems, and any other devices capable of performing a given, well-defined sequence of computing operations (actions, instructions).

[0030] A command processing device is an electronic unit or an integrated circuit (microprocessor) that executes machine instructions (programs).

[0031] The command processing device reads and executes machine instructions (programs) from one or more data storage devices. In the role of a storage device can act, but not limited to, hard drives (HDD), flash memory, ROM (read-only memory), solid-state drives (SSD), optical drives.

[0032] Program - a sequence of instructions intended for execution by a computer control device or a command processing device.

[0033] The following will describe the terms and concepts needed to implement this technical solution.

[0034] Virtual interlocutor, interlocutor, chat bot (eng. Chatbot) is a computer program that simulates human speech behavior when communicating with one or more interlocutors.

[0035] Lemmatization is the process of bringing a word form to a lemma — its normal (vocabulary) form. For example, “cats” after lemmatization is transformed into “cat”.

[0036] A database (DB) is a collection of data organized in accordance with a conceptual structure that describes the characteristics of this data and the relationship between them, a collection of data that supports one or more applications (ISO / IEC 2382: 2015, 2121423 " database ").

[0037] As shown in FIG. 1, the system 100 for lining up a dialogue with a user in a user-friendly data channel may contain the modules described below, between which data is exchanged.

[0038] Previously, the text preprocessing module 120 receives a request for the implementation of a service from a user (or some modification of it), which generates it through its graphical user interface 110, for example, in a mobile application on a mobile communication device.

[0039] The service requested by the user may include, for example, obtaining a loan, issuing plastic cards, deposits, leasing, servicing a current account, foreign currency transactions, etc.

[0040] At this step 401, as shown in FIG. 4, in fact, user input is received, which may be a string or a set thereof, and a character one. User input can include both a single question and a set of them. User input data is stored in database 150. The string may be, without being limited to, for example, the following: “Hello, chat bot! Transfer Denis Ivanov from my account 100 rubles to his phone. ”

[0041] The basic text preprocessing module 120 is responsible for reducing the variety of possible message texts in order to simplify the work of the following modules of the system 100. This module is intended to split the incoming user input into sentences and words, as well as their morphological analysis, syntax analysis and semantic typing tokens.

[0042] The tokenization stage involves the selection of basic text elements (tokens) bounded on two sides by dividing characters, spaces or punctuation marks. The elements here are words, numbers, dates, abbreviations, abbreviations, compound prepositions, etc. Tokenization allows you to select discrete units of text, which are the basis for further work at the stages of morphological and syntactic analysis. As a result of tokenization, each element is assigned an appropriate type: word, number, date, address, etc.

[0043] The purpose of text preprocessing is to prepare it for the qualitative classification of requests. For this, the request goes through many stages of preprocessing 402 and enrichment with various information, among which the main stages are:

[0044] typo correction;

[0045] lemmatization with the removal of morphological homonymy;

[0046] enriching the results of parsing the syntax parser and the individual

selection of a set of triples object + subject + action (SAO);

[0047] the translation of numerals from text to digital representation ("three hundred" is replaced by 1300), the allocation of name, time, etc.

[0048] The text preprocessing module 120 includes, but is not limited to, as shown in FIG. 2, text syntax parser 210, typo correction sub-module 220, Russian morphological analyzer 230 with support for removing morphological ambiguity, sub-module 240 replacing synonyms and collocations, sub-module 250 replacing numerals, monetary amounts and telephone numbers.

[0049] The typo correction module 220 is used to correct spelling errors of users, which helps to properly classify the request.

[0050] The principle of operation of this sub-module 220 is as follows:

[0051] break the sentence into words;

[0052] for each word:

a. if the word is longer than 2, return it;

b. if there is a typical correction of the error in the dictionary - a typical correction is returned;

c. if the word consists not only of letters, return it;

d. if the word is familiar to the frequency dictionary - return it

e. if the word is not familiar, the most likely correction is returned:

i. generate all options for correcting a word with one change: deleting, adding, replacing single characters or replacing two adjacent characters in a word;

ii. filter only familiar to the frequency dictionary, allocate their share (p1);

iii. if candidate 1 returns it;

iv. the probability of a specific error (p2) is distinguished (for example, the probability of an error is e / and higher than e / b);

v. highlight the probability of combining the selected correction options with the previous word (p3);

vi. calculate the cumulative probability of each candidate p = p1 ^ λ1 * p2 ^ λ2 * p3 ^ λ3 where λ1, λ2 and λ3 are adjustable parameters, and return the best.

[0053] The morphological analyzer 230 of the Russian language with support for removing morphological ambiguity is a set of algorithms that match individual words and word forms in the dictionary (lexicon, to be exact) and ascertain the grammatical characteristics of words.

The marking of the source text with grammatical information makes it very easy to make rules further when the syntax parser 210 of the text works. The work of the morphological analyzer 230 is finalized by establishing the morphological features of the words of the text. This task can be considered as a task of marking or tagging text - identifying tags (morphological features). The set of features to be installed is directly dependent on the language. In some embodiments, the morphological analyzer 230 can perform morphological analysis, which can be a dictionary (with a dictionary of foundations and endings or a dictionary of word forms) or wordless (only with a dictionary of endings; a dictionary of endings can be embedded into the algorithm of morphological analysis 230). The wordless method is used only to determine the variable of morphological information (not always unambiguously), and the vocabulary method is used in all other cases. Algorithms, a model of the formation of the morphological analyzer 230, which generates and defines word forms and examples for various natural languages, are known from the source of information [3].

[0054] Sub-module 240 replacing synonyms and collocations is configured to define synonyms and collocations in user input and the existing database 150 based on statistical measures (MI, t-score, log-likelihood), which are most often used in determining the degree of closeness between components phrases in the case. Measure Ml allows you to select stable phrases, proper names, as well as low-frequency special terms. The words for which the MI-score measure takes the greatest value are less frequent and have limited compatibility. The t-score measure also takes into account the frequency of joint occurrence of a keyword and collocate, answering the question of how random the association (cohesion) strength between collocates is. Words with the highest t-score turn out to be frequency and can be combined with many units. In some embodiments, the sub-module 240 replacing synonyms and collocations may use each of the above measures or more in combination.

[0055] The result of the text parser 210 is the structure of the dependencies of words from each other in the user input in this text (its separate sentences). For example, from the phrase “transfer a hundred rubles to mother”, system 100 makes a decision that a mother is an indirect object (addressee).

[0056] The syntax parser 210 of the text highlights the syntactic structure of the sentence, which is a dependency tree, at the nodes of which there are words of the sentence, and the branches are labeled with the names of syntactic relations.

[0057] The operation of the text parser 210 is already recognizing user input 403, as shown in FIG. four.

[0058] In some embodiments, the text parser 210 for analyzing user input may use recurrent neural networks (RNN) for its work to train the text syntax parser 210. Recurrent neural networks are a class of machine learning models based on using previous network states to calculate the current one. Each character in the source text, individual words, punctuation marks, and even whole phrases are all an atomic element of the input sequence for a neural network. In some embodiments, managed recurrent neurons (gated recurrent units, GRU) may be used. The update filter determines how much information will remain from the past state and how much will be taken from the previous layer. The reset filter works something like a forgetting filter.

[0059] For training neural networks, corps, namely tribanks, are used as a training set. In linguistics, the corpus is a set of texts selected and processed according to certain rules, used as a database for language research. They are used for statistical analysis and verification of statistical hypotheses, confirmation of linguistic rules in a given language. Tribank is a collection of parsed sentences (that is, parse graphs) prepared manually or automatically in advance. Tribanki by classification are divided into phrase-structure treebanks and dependency treebanks. The following tribanks or cases for the Russian language can be used in this technical solution without limitation: SynTagRus (1.107 thousand tokens), PUD (19 thousand tokens), GSD (99 thousand tokens), Taiga (20 thousand tokens), Dependency Treebanks, etc.

[0060] Next, apply the Transition-based dependency parsing approach, widely known from the prior art. This approach consists in an attempt to predict the sequence of actions (transitions) from some initial configuration of a phrase or user request to the final one, as a result of which the search parse tree will be obtained, which allows one to obtain sufficiently high accuracy and to achieve rather high speeds during text processing.

[0061] The Arc-standard system is one of the most popular approaches for implementing a transition-based system. The system is described by a configuration consisting of three parts: c = (s, b, A),

[0062] where: s is the data stack;

[0063] b - data buffer;

[0064] A is the set of dependencies.

[0065] The initial configuration for the sequence of symbols w ₁ , ..., w _{n is the} following before processing:

[0066] s = [ROOT] - one service character in the stack;

[0067] b = [w _l , ..., w _n ] - in the buffer, the entire sequence of characters;

- множество зависимостей пусто.[0068] A =

- many dependencies are empty.

[0069] The final configuration after processing is as follows:

[0070] s = [ROOT] - one service character in the stack;

[0071] b is empty;

[0072] A - contains the desired parse tree.

[0073] We assume that s _i where i = (1, 2, ...), the i-th top element of the stack, b _i , i = (1, 2, ...) is the i-th element of the data buffer.

[0074] The Arc-standard system approach has 4 types of operations:

[0075] SHIFT - remove b ₁ from the buffer and add it to the stack;

[0076] LEFT_ARC — adds a link from s ₁ to s _{2 in A} , with a specific label of the type of connection, and removes s ₂ from the stack;

[0077] RIGHT_ARC - similar to LEFT_ARC, only with the replacement of s ₁ and s ₂ .

[0078] SWAP: returns the second element from the stack to the buffer.

[0079] Thus, we get a total of | T | = 2N _i +1 possible actions, where N _i is the number of types of dependency labels. The purpose of the text parser 210 is to select the most appropriate action for this configuration.

[0080] For learning artificial neural network is required on the basis of available data to generate the most appropriate sequence of actions. At each step, the configuration will contain the necessary data, and the action - the answer.

[0081] The result of the work of the text parser 210 is a parse tree, where for each element of the phrase its parent and type of dependency are indicated.

The following metrics can be used to evaluate the performance of the parser:

[0082] Unlabeled Attachment Score (UAS) - the ratio of the number of elements with a correctly specified parent to the total number of elements;

[0083] Labeled Attachment Score (LAS) is the ratio of the number of elements with a correctly specified parent and type of relationship to the total number of elements.

[0084] An example of the operation of the text processing module 120 may be as follows:

[0085] Login: Request: "a person wants to take a loan for 100 million rubles a day"

[0086] Exit:

[0087] The user is not always able to accurately formulate his request for information that he needs. Moreover, even after receiving this information, its subsequent analytical processing is required in order to determine its usefulness and suitability for solving the problem. The difficulties associated with solving this problem lie in the variety of possible forms of expression of the same idea, thought, which is especially characteristic of Russian-language texts.

[0088] To solve this problem in this technical solution form a vector model of user input words. The main advantage of the vector model is the ability to search and rank documents by similarity, that is, by their proximity in the vector space by determining the distance between words.

[0089] The coordinates of the vectors are formed in such a way that the cosine or Euclidean distance between the vectors of similar words is smaller than the distance between the vectors of words that are far from the meaning. At the same time, individual components of the vector may reflect some specific category, for example, the first component may contain information about time (the past is the present - the future), the second is about physical size (small - large), the third about cost (expensive - cheap) and t .d

[0090] The third class classifies, via the dialog module 130, at least a portion of the user input for generating a response.

[0091] The system 100 includes a user dialogue module 130 that is responsible for maintaining a multi-stage user dialogue, re-using the dialogue context, or using data from available external services.

[0092] This module 130 is used to enable separate dialogue scenarios with the user on new topics to be created without a single additional line of program code / with a small and simple amount of code, so that this solution:

[0093] allowed non-programmers to start new subjects;

[0094] was easier to maintain.

[0095] At the same time, the architecture of the module 130 allows embedding with its own (compatible with the architecture of the solution) routines that correspond to specific topics and can be of arbitrary complexity.

[0096] In order to be able to classify user requests, and accordingly maintain the context of the conversation with the user, interview the client on a given set of questions, module 130 can use an approach based on the slot or form filling algorithm (from the English Slot Filling) widely known from the prior art.

[0097] When using this approach, two types of scenarios can be defined:

линейные сценарии диалога, в которых собирается информация, необходимая для ответа на запрос пользователя из базы данных 150 и/или внешних источников информации, и

linear dialogue scenarios in which information is collected necessary to respond to a user request from a database 150 and / or external sources of information, and

нелинейные сценарии, в которых собираемая информация зависит от предыдущих ответов пользователя.

non-linear scenarios in which the information collected depends on the user's previous responses.

[0098] With this approach, each linear scenario is represented in the form of a form consisting of a set of fields, a form's storage time, and a prepared response. The answer can be a chain of arbitrary actions and is performed after filling in all the required fields of the form.

[0099] Each form field is defined as a set of properties:

значение по умолчанию;

default value;

уточняющий вопрос для заполнения данного поля;

clarifying question to fill in this field;

обязательность заполнения (если поле не заполнено, но обязательно - будет задан уточняющий вопрос в диалоге пользователю);

Mandatory filling (if the field is not filled, but mandatory - a clarifying question will be asked in the dialogue to the user);

функция для заполнения (одна из заранее заготовленных, которая из результата предобработки текста или использования вспомогательных внешних сервисов/баз данных извлекает значение для заполнения поля);

the function to fill in (one of the prepared in advance, which from the result of preprocessing text or using auxiliary external services / databases extracts the value to fill in the field);

возможность использовать значение поля формы из контекста другим формам и возможность заполнить поле значением из контекста;

the ability to use the value of the form field from the context of other forms and the ability to fill the field with the value from the context;

время жизни значения поля (после которого оно будет очищено) и другие.

the lifetime of the field value (after which it will be cleared) and others.

[00100] In this approach, nonlinear scenarios are set up in the form of a unidirectional acyclic form graph with the initial form for the script. In this case, all forms are similar to forms of linear scenarios and are filled in similarly, but there are transition possibilities between them. The possibility of transition from filling one form to another is specified in the form of imposed restrictions on the field values of the first one.

[00101] Thus, the module 130 for classifying requests works according to the following algorithm:

[00102] 1. Retrieves the processed (step 402) and recognized (step 403) in the previous step text obtained from the text preprocessing module 120;

[00103] 2. Makes an ordered list of scenarios to traverse them: the last for a given user, and then any remaining. Each scenario can be represented by one form and refers to one topic / request / required service;

[00104] 3. Conditions for entering a particular scenario are defined as rules, embedded arbitrary classifiers, examples that are used to classify the platform method used in the module 140 for answering frequently asked questions, or a combination of both. The latest scripts for the user are checked for the possibility of filling out the form fields with a client request, and in case of success, control is transferred to them. Otherwise, hierarchical classification can be used to compile a list of the most likely verification scenarios, after which their specific conditions for entering the scenario are checked;

[00105] 4. Control is transferred to the most likely scenario among a set of scenarios that have passed confidence thresholds that are necessary and specific to each scenario. The user is then returned a response from this scenario by filling out the script form. In the absence of a significant difference in the probability of the two different scenarios that have passed, the user may be asked a clarifying question, which of them is implied.

[00106] In cases where the dialogue module 130 does not include any scripts matching the user's request, the module 140 for answering frequently asked questions is used. This module 140 is used when neither dialogue, client-dependent answers, but only answers to frequently asked questions are required. Minimal efforts are needed to add new topics: only a list of the questions themselves, answers to them and possible paraphrasing of questions (although the search allows for arbitrariness, but more examples still improve the quality).

[00107] Work Example:

Login: "Hello"

Exit: "Hi, hi. :)"

Login: "what is my delay"

Output: To clarify information about the presence of overdue debts, you need to call the telephone number 8-800-333-31-38 "

[00108] The necessary data for the operation of this module 140 may be as follows:

[00109] reference question name for navigation,

[00110] the answer for the user

[00111] data channels in which this issue is available,

[00112] Positive and negative examples of requests (positive are requests that should lead to this answer, negative ones are similar to positive requests that should not lead to this answer).

[00113] The classification of a query on user topics from module 130 and module 140 may be combined in some implementations. After determining the subject matter, control is transferred to the appropriate module.

[00114] In some embodiments, to search for the desired scenario, the normalized user input text is converted into a word vector (form a vector word model) using the statistical measure TF-IDF, used to evaluate the importance of the word in the context of a document that is part of a collection of documents or corpus.

[00115] TF-IDF is a statistical measure used to assess the importance of a word in the context of a document that is part of a document collection or corpus. The weight of a word is proportional to the number of words used in the document, and inversely proportional to the frequency of words in other documents in the collection.

[00116] The TF-IDF measure is often used in text analysis and information retrieval tasks, for example, as one of the criteria for the relevance of a document to a search query, when calculating the measure of document proximity for clustering.

[00117] TF (term frequency) is the ratio of the number of occurrences of a word to the total number of words in the document. The significance of a word within an individual document can be determined by the following characteristic:

[00118] where n _i is the number of occurrences of the word t _i in document d; Σ _k n _k is the total number of words in this user query and / or document.

[00119] IDF (inverse document frequency) is a value inversely proportional to the frequency with which a word occurs in collection documents.

[00120] Accounting for IDF reduces the weight of widely used words. For each unique word within a specific collection of documents, there is only one IDF value. IDF-characteristic is determined by the following relationship:

- количество документов, в которых встречается t_i.[00121] where | D | - the number of documents in the case;

- the number of documents in which t _i occurs.

[00122] Thus, the measure TF-IDF is the product of two factors:

[00123] Much of the weight in the TF-IDF measure is given to words with a high frequency within a particular document and with a low frequency of usage in other documents.

[00124] The TF-IDF measure is often used to represent collection documents as numeric vectors, reflecting the importance of using each word from a certain set of words (the number of set words determines the dimension of the vector) in each document. Such a model is called a vector model and makes it possible to compare texts by comparing the vectors representing them in some metric (Euclidean distance, cosine measure, Manhattan distance, Chebyshev distance, etc.), that is, producing cluster analysis.

[00125] In this technical solution, Word2vec, a software tool for analyzing the semantics of natural languages, which is a technology based on distributive semantics and vector representation of words, can be used to generate output data (step 404, as shown in Fig. 4).

[00126] Word2vec suggests finding links between the semantics of words according to the assumption that words that are in similar contexts tend to mean similar things, i.e. to be semantically close, as shown in FIG. 5. For example, word2vec allows you to use "mathematics" in words: "king" - "man" + "woman" = "queen". More formally, the task is as follows: maximizing cosine proximity between word vectors (dot product of vectors) that appear next to each other, and minimizing cosine proximity between word vectors that do not appear next to each other. Next to each other in this case means in close contexts.

[00127] For example, the words “analysis” and “research” are often found in similar contexts. The phrases “Scientists have analyzed algorithms” or “Scientists have conducted a study of algorithms” are quite similar. Word2vec analyzes these contexts and concludes that the words “analysis” and “research” are close in meaning.

[00128] For example, for the word "coffee", word2vec may output the following 15 nearest neighbors in the format "word - cosine distance":

[00129] Coffe 0.734483;

[00130] tea: 0.690234;

[00131] tea: 0.688656;

[00132] cappuccino 0.666638;

[00133] cofn 0.636362;

[00134] Cocoa 0.619801;

[00135] Espresso 0.599390;

[00136] coffee shop 0.595211;

[00137] Chicory 0.594247;

[00138] kofe 0.593993;

[00139] Kopuchino 0.587324;

[00140] chocolate 0.585655;

[00141] cappuccino 0.580286;

[00142] cardamom 0.566781;

[00143] Latte 0.563224.

[00144] To form a response to the user (step 404) in some embodiments of the invention, the vectors of all words that are included in the user data input are used, the average (centroid) distance is determined between them, and then the cosine distance between the user input and the centroid is determined.

[00145] In some embodiments, instead of using word2vec, vector models such as GloVe, FastText, etc. are used without being limited to.

[00146] In this technical solution, the following CBOW (Continuous Bag-of-Words) and Skip-gram models can be used for vector representation of words without being limited to.

[00147] Then, by the multi-index question vectors from the FAQ (frequently asked questions database by users), which is located in database 150, we rank a bit as close as possible to the Euclidean distance or cosine similarity. Multi-index is a generalization of the concept of an integer index to a vector index, which has found application in various areas of mathematics related to functions of many variables. The cosine similarity is a measure of the similarity between two vectors of a pre-Hilbert space, which is used to measure the cosine of the angle between them.

[00148] If two feature vectors are given (in this case, two word vectors), A and B, then the cosine similarity, cos (θ), can be represented using their scalar product and the norm:

[00149] In some embodiments, to determine the most similar questions that are known in advance, and the answers to them, the Minkowski distance (L1), Euclidean distance (L2), Jacquard Measure can be used as a metric.

[00150] For example, if we have the phrase "Jessica loves oranges and mandarins", which can be represented as a vector of words A (A: [1,1,0,0,0,1,1,1]], and the phrase “Jesse likes citrus fruits”, which can be represented as vector B (B: [0,0,1,1,1,0,0,0]). Due to the fact that there are no common words, the cosine distance will be 0. However, it must be taken into account that the meaning in these phrases is semantically similar, since “citrus” in the parse tree is the parent of the words “oranges” and “mandarins”.

[00151] Next, the result is formatted. Depending on the distance of the question to the best option and the general cut-off threshold, the user is given either a pre-prepared answer (step 405, as shown in Fig. 4), or a pair (question, answer) (if the chance of irrelevance of the answer is sufficient) or switch .

[00152] Potentially, system 100 may decide to output a list of these questions to the user's choice in the right place if the graphical user interface allows it or requires some kind of separate command. To support this mode, it is enough to pass an additional parameter to this module 140, meaning returning the list of the best options for formatting instead of the text itself.

[00153] In other embodiments of the invention, a list of frequently asked questions is preliminarily created in the database, and then when new user input is received, the Word Mover's Distance method proposed in the information source [1] is applied.

[00154] With this approach, for each word from one phrase, the nearest neighbor from the other is found. Word Mover's Distance (WMD) is the minimum distance to go from one phrase to another, as shown in FIG. 3

[00155] The minimum cumulative distance for this approach can be determined as follows.

[00156] Suppose there is a set of vectors of words and distances between words for two phrases: for example, the question sounds after preprocessing the text as “how to buy a car loan”, and the answer should be “hi give a loan machine”.

[00157] Next, module 140 takes the minimum distance between words from each column, from each line, after which they are summed up, divided by the total number of words (normalized), and get "the cost of transition from one phrase to another":

[00158] (0.64 + 0.90 + 0.00 + 0.39 + 0.64 + 0.82 + 0.88 + 0.39 + 0.00) / 9 = 0.52

[00159] In some embodiments, normalization can be performed on the total number of unique tokens or on the number of unique tokens in a phrase, without being limited to.

[00160] In some implementations, for the topic “How to block cards”, it may be an erroneous approach, “Have I locked my credit card, how can I cancel it?”. In such cases, such negative examples are stored in the database 150, their relation to each other by semantics, and take this into account in further work.

[00161] In some embodiments of the invention, not only the distance between words is used, but also additional attributes, including but not limited to: the frequency of occurrence of each word in database 150, the morphological features of each word, syntactic information.

[00162] In some embodiments of the invention, not only the distance between words is used, but also some proper weight of each word, for example, its frequency of occurrence.

[00163] If x - proportion of all those in which the word W, and at - the proportion of all the examples in this topic, in which the word W, the distance between the words is multiplied by 1-x ⁱⁿ for normalization.

[00164] Also, the problem with the traditional WMD approach may be that it does not take into account the word order, so the phrases “ATM ate a card” and “a card ate an ATM” are identical after lemmatization. To solve this problem, you can use the syntactic information from the text syntax parser 210, which displays the word parsing tree. In a specific implementation, the syntax parser 210 displays an error, since the construction of the card “ate an ATM” cannot be used. To correct this situation, a matrix of cost of transitions from one class to another is introduced and the Euclidean distances between the phrases are multiplied by the corresponding values.

[00165] To provide the ability to make antonyms further along the distance between words, and synonyms are closer in word2vec from the point of view of eg Euclidean distance, not much damaging the distances of other words in space, use the approaches described in the level of technology (source of information [2]) For example, Retrofitting.

[00166] If none of the previous steps (the operation of the module 130 and the module 140) failed to pick up the answer in the dialog of the graphical user interface, the user will receive a message about this, or the user will be switched to the operator. In some cases, there may be a proposal to switch the user to an operator for communication over another data channel.

[00167] In a specific implementation, without limitation, an operation mode can be chosen that simulates the absence of the chat bot itself and smoothly transfers to the operator at virtually any incomprehensible user request (this is done because of the small amount of chat bot content):

[00168] If the chat bot is waiting for an answer to a question asked by it and receives a message that module 130 or module 140 cannot recognize, then it gives another attempt to the user (informing him about it) and, if repeated, translates to the operator;

[00169] If the chat bot does not wait for a response, but simply waits for a request from the user and receives an unrecognizable message, then immediately transfers the user to the operator.

[00170] In general, the module behavior is set by a special parameter of the form v = [2,3,6]. In this example, the chat bot will behave as follows:

[00171] when receiving the 1st unrecognized message, the fact of its existence is forgotten after v [1] = 2 recognized messages;

[00172] when receiving the 2nd unrecognized message, the fact of its existence is forgotten after v [2] = 3 recognized messages;

[00173] when receiving the 3rd unrecognized message, the fact of its existence is forgotten after v [3] = 6 recognized messages;

[00174] Receiving a 4th unrecognized message takes the user to an operator (or, more gently, invites him to do this in the graphical user interface).

[00175] In general, if a transfer to an operator is not required, one random message is output from blanks with a single meaning: it is reported to the user that nothing has been found.

[00176] The illustrative implementations described herein may be implemented in an operating environment containing computer-executable commands (eg, software) installed on a computer, in hardware, or in a combination of software and hardware. Computer-executable commands may be written in a computer programming language or may be embodied in hardware logic. If written in a programming language in accordance with a recognized standard, such commands can be executed on a variety of hardware platforms and for interfaces of a variety of operating systems. Although not limited to this, computer programs of the software system for implementing this method can be written in any number of suitable programming languages, such as, for example, Hypertext Markup Language (HTML), Dynamic HTML, Extensible Markup Language (XML), Extensible Stylesheet Language (XSL), style semantics and document specification language (DSSSL), cascading style sheets (CSS), synchronized multimedia data integration language (SMIL), document markup language for wireless communications (WML), JavaTM, JiniTM, C, C ++, Perl, UNIX Shell, Visual Basic or Visual Basic Script, virtual reality event markup language (VRML), ColdFusionTM or other compilers, assemblers, interpreters or other computer languages or platforms.

[00177] System 100 may be implemented by an architecture, as shown in FIG. 6, and include the following components shown below, including the processor 610. In a particular implementation of this technical solution, the processor 610 may include one or more processors and / or one or more microcontrollers configured to execute instructions for performing operations associated with the operation of the above method for building a dialogue with the user in a user-friendly channel. In various embodiments of the present technical solution, the processor 610 may be implemented as single-chip, multi-chip, and / or electrical components, including one or more integrated circuits and printed circuit boards. The processor 610 may optionally contain a block of cache memory (not shown) for the temporary local storage of instructions, data, or computer addresses. For example, processor 610 may include one or more processors or one or more controllers related to specific tasks or a single multi-functional processor or controller.

[00178] The processor 610 is operatively coupled to the data input / output module 620, audio module 630.

[00179] In the present embodiment, the implementation of this technical solution, data input / output module 620 can be implemented as a touch screen that performs the functionality of both input devices (by fixing user commands in the form of touches) and user output devices (i.e. display). In other words, the touch screen is a display that determines the presence and position of user input-touch. In alternative embodiments of the present technical solution, the I / O module 620 may be implemented as a separate display and a separate input device. However, in other alternative embodiments of the present technical solution, the I / O module 620 may include a physical keyboard (comprising one or more physical buttons) in addition to the touchscreen.

[00180] The processor 610 is further coupled to a memory module 640, which contains a database 150. A memory module 640 may span one or more media and generally provide storage space for computer code to implement the above method of transmitting p2p translation information (eg, software and / or hardware). For example, memory module 640 may include various tangible computer-readable media, including read-only memory (ROM) and / or random access memory (RAM). As is known to those skilled in the art, ROM unidirectionally transmits data and instructions to processor 610, and RAM is typically used to transmit data and instructions bilaterally.

[00181] The memory module 640 may also include one or more fixed storage devices in the form of, for example, a hard disk drive (HDD), solid state drive (SSD), flash memory card (e.g., secured digital or sd card, multimedia eMMD cards), along with other types of memory, are bilaterally connected to the processor 610. Information can also be located on one or more removable media loaded or installed in the system 100 when necessary. For example, any of a number of suitable memory cards (for example, SD cards) may be loaded into system 100 on a temporary or permanent basis (using, for example, one or more sets of additional ports).

[00182] The memory module 640 may store, among other things, a series of machine-readable instructions, in which the processor 610 (as well as other components of the system 100) are configured to perform various operations described herein.

[00183] In various specific embodiments, system 100 may further comprise a wireless communication module 650 and a sensor module 660, both of which are connected to the processor 610 to simplify various functions of the system 100.

[00184] The wireless communication module 650 may be created to work through one or more wireless networks, such as a wireless personal area network (WPAN) (such as BLUETOOTH WPAN, IR personal area network), a WI-FI network (eg, 802.11a / b / g / n WI-FI network, 802.11 standards network), WI-MAX network, mobile cellular network. As a mobile cellular network, for example, a global system for mobile communications (GSM), a network of developed GSM standard with an increased data transfer rate (EDGE), a universal mobile telecommunications system (UMTS) network or a long-term development network (LTE) can be used. Additionally, the wireless communication module 650 may include hosting protocols such that system 100 can be configured as a base station for wireless devices.

[00185] The sensor module 660 may include one or more sensor devices to provide additional input and simplify various functions of the system 100. Some example embodiments of the sensor module 660 may include one or more of the devices: accelerometer, temperature measurement device device, device for measuring gravity, gyroscope, device for measuring light intensity, device for measuring acceleration force, device for measuring the surrounding geomagnetic field, a device for measuring the degree of rotation, a device for measuring atmospheric pressure, a device for measuring relative humidity, a device for measuring the orientation of the device, and so on. It should be noted that some of these devices can be implemented as hardware, software, or a combination of both.

[00186] A power supply module 670 is also proposed to provide power to one or more components of system 100. In some embodiments of the present technical solution, power supply module 670 may be implemented as a lithium-ion battery.

However, other types of rechargeable (and conventional) batteries can be used. Naturally, in other embodiments of the present technical solution, additionally or alternatively to using a battery, power supply module 670 may be implemented as a main power source capable of connecting the system 100 to a main power source, for example, a standard power cable and plug.

[00187] In some embodiments of the present technical solution, the various components of the system 100 may be connected to each other through one or more buses (including hardware and / or software), these buses are not numbered. As a non-limiting example, one or more buses may include an accelerated graphics port (AGP) or other graphics ports, an improved industry standard bus architecture (EISA), a front bus (FSB), a hypertransport bus (NT), an industry standard bus ( ISA), INFINIBAND connection, LPC-bus, memory bus, microchannel architecture bus (MCA), peripheral component connection bus (PCI), express-type peripheral component connection bus (PCI-X), storage bus serial interface information (SATA), local video electronic standards association (VLB) bus, universal asynchronous transceiver interface (UART), serial data bus for integrated circuits (I2C), serial peripheral interface bus (SPI), Secure Digital memory interface (SD), and a MultiMediaCard (MMC) memory interface, a Memory Stick (MS) memory interface, a Secure Digital Input Output (SDIO) interface, a multichannel buffered serial port (McBSP) bus, a universal serial bus (USB), a universal storage controller bus a (GPMC), random access synchronous dynamic memory (SDRC) controller bus, general purpose input / output bus (GPIO), separate video bus (S-Video), display serial interface bus (DSI), advanced bus bus architecture for microcontrollers (AMBA), or any other suitable tire or combination of two or more tires.

[00188] In some embodiments, system 100 using speech-to-text and text-to-speech modules can be used for voice service (IVR) and / or voice control of a mobile application.

[00189] It will be understood by those skilled in the art that, in the present description, the expression “receiving data” from a user means receiving electronic data from a user as an electronic (or other) signal. In addition, those skilled in the art will understand that displaying data to a user through a component of a graphical user interface (for example, an electronic device screen, etc.) may include transmitting a signal to a component of a graphical user interface, this signal contains data that can be processed, and at least some of this data can be displayed to the user through a graphical user interface component.

[00190] Some of these steps, as well as the transmission and reception of a signal, are well known in the art and, therefore, have been omitted in particular parts of this description for simplicity. Signals can be transmitted / received using optical means (for example, fiber optic connection), electronic means (for example, wired or wireless connection) and mechanical means (for example, based on pressure, temperature, or other suitable parameter).

[00191] Modifications and improvements to the above-described embodiments of the present technical solution will be clear to those skilled in the art. The foregoing description is provided by way of example only and is not limited in any way. Thus, the scope of this technical solution is limited only by the scope of the attached claims.

SOURCES OF INFORMATION

[00192] 1. Kusner M. et al. From word embeddings to document distances // International Conference on Machine Learning. - 2015. - p. 957-966.

[00193] 2. Faruqui M. et al. Retrofitting word vectors to semantic lexicons // arXiv preprint arXiv: 1411.4166. - 2014.

[00194] 3. Prutskov A.V. Generation and definitions of the forms of words of natural languages based on their successive transformations // Bulletin of the Ryazan State Radio Engineering University. - 2009. - no. 27. - p. 51.

Claims (22)

1. A method for building a dialogue with a user in a data transmission channel, comprising the following steps: get through the processor, functionally connected to the database, user input data; pre-process user input by splitting it into sentences and words, with: correct user spelling errors by using the typo correction module; perform lemmatization of each word from user input; form the structure of the dependencies of words from each other in the user input by using the syntax parser; form a vector model of user input words; classifying by means of the dialogue with the user at least part of the user input to generate a response; form the answer for the user to display in the graphical user interface; provide by the processor at least one response to the recognized user input. 2. The method according to p. 1, characterized in that the implementation of the preprocessing of user input additionally correct typos in user input through a submodule correcting typos. 3. The method according to claim 1, characterized in that in the implementation of the pre-processing of user input additionally perform the tokenization of numerals from the user input. 4. The method according to claim 1, characterized in that the text syntax parser uses recurrent neural networks to analyze user text. 5. The method according to claim 1, characterized in that when classifying user input, the dialogue module uses the slot filling algorithm. 6. A method according to claim 1, characterized in that when classifying, by means of a dialog module with a user, user input, linear dialog scripts or non-linear dialog scripts are defined. 7. The method according to claim 1, characterized in that when forming a vector model of user input words, the model CBOW and / or Skip-gram is used. 8. The method according to claim 1, characterized in that when generating a response for the user to display in the graphical user interface, determine the class to which the user input relates. 9. The method according to claim 1, characterized in that when forming a response for the user to display in the graphical user interface in the absence of a response in the dialogue switch it to the operator. 10. The system for building a dialogue with the user in the data channel, containing: a text preprocessing module made with the possibility of receiving user input and its lemmatization, correcting typos, tokenization and syntactic analysis in it to form the structure of dependencies of words from each other in user input; a user dialogue module, configured to obtain from the preprocessing module text of the processed text of user input, form a vector model of user input words and classify at least a part of the user input to form a response; a module for answering frequently asked questions, configured to receive user input from the module for maintaining a dialogue with the user and to form an answer for the user to be displayed in the graphical user interface.

Images