WO2015065225A1

WO2015065225A1 - Personal assistant with elements of artificial intelligence and method for using same

Info

Publication number: WO2015065225A1
Application number: PCT/RU2013/000964
Authority: WO
Inventors: Андрей Юрьевич ЩЕРБАКОВ; Олег Олегович ТИХОНЕНКО
Original assignee: Андрей Юрьевич ЩЕРБАКОВ; Олег Олегович ТИХОНЕНКО
Priority date: 2013-10-29
Filing date: 2013-10-29
Publication date: 2015-05-07
Also published as: RU2014142268A

Abstract

The invention relates to the field of information technology, and more particularly to a personal assistant with elements of artificial intelligence and to a method for using same. The present invention can be used in the creation and use of computer systems, control systems and production systems having various purposes, in which a user needs to have a meaningful dialogue with a computer system and in which the accumulation of knowledge is required. By comparison with other methods of communication with computer systems, the difficulty of formulating search queries is decreased because questions are worded in natural language and search results are selected appropriately; learning and knowledge accumulation are possible, as is the taking into consideration of user experience and preferences, and an active dialogue is conducted between the system and the user, including communication initiated by the system.

Description

INDIVIDUAL ASSISTANT WITH ELEMENTS

ARTIFICIAL INTELLIGENCE AND METHOD OF ITS APPLICATION

Description

The invention relates to the field of computer science, and more specifically to an individual assistant with elements of artificial intelligence and the method of its application.

The present invention may find application in the creation and

operation of computer, control and production systems

for various purposes, in which the user needs a meaningful dialogue with a computer system and the accumulation of knowledge.

The current level of awareness of this problem is implemented in any search engine.

However, modern search engines have the following

fundamental shortcomings:

- the complexity of compiling search queries and selecting search results, the inability to learn and accumulate knowledge, taking into account the experience and preferences of the user,

lack of dialogue between the system and the user.

The objectives of the invention are solved and the fundamental disadvantages of the prototype in the face of modern search engines are eliminated in an individual assistant with elements of artificial intelligence and the method of its application.

An individual assistant with elements of artificial intelligence, designed to conduct a dialogue with the user by exchanging questions and answers, which are texts in natural language and forming tips for the user in the form of texts in natural language, consisting of components that perform operations, implemented in arbitrary

computing resources, and components that perform operations include at least the following: component recognition of texts entered by the user, the search component in the data arrays, the component for generating responses for the user, the feedback component that evaluates the responses generated for the user, the configuration component that controls

the formation of responses, a component of communication with other individual

helpers, a component of communication with search engines, and data arrays, including but not limited to the following: user question data array, user response data array, settings data array containing information about the user's preferences and individual assistant operation mode, an array of events describing information about past or upcoming events.

The way to use an individual assistant with elements

The artificial intelligence described above is as follows:

a) the user generates questions that enter the text recognition component entered by the user, which in turn passes them to the search component in the data arrays, after which the search component in the data arrays compares them with the data array of the answers to the user and if there is an identical or similar answer it is sent to the response generation component for the user,

b) if the user’s question forms an event, information about the event is an array of events that describes information about past or upcoming events,

c) if an identical or similar answer is not found in the data array, the text recognition component refers to the communication component with other individual assistants, which transfers the question to another individual assistant and expects an answer from it, another

an individual assistant functions similarly,

e) in the event that a response from another assistant is not received or contacting other assistants is impossible or prohibited by the settings component, the recognition component for texts entered by the user selects

the content of the question in the form of text, after which this text is transmitted to the communication component with search engines, the answer from which is sent to the component generating responses for the user, in which the response is compared with an array of settings data and taking into account this comparison is transmitted to the user,

f) the user, having received the response, does not necessarily evaluate it, for which the user’s response is transmitted to the feedback component, and the response is placed in the response data array with the user's rating, or placed in the event array,

h) the component for generating answers for the user independently, regardless of the user's questions, forms tips for the user in the case when a question comes up from other individual assistants, the question itself or the answer to which is indicated in the data set of settings as interesting for the user, as well as when the previously asked questions require an answer at the right time, or an event has occurred that exists in the array of events.

Thus, the complexity of compiling search queries is reduced, since the questions are formulated in a natural language and there is an adequate selection of search results, it is possible to learn and accumulate knowledge, take into account the experience and preferences of the user, an active dialogue is conducted between the system and the user, including messages initiated by the system.

The present invention is disclosed in the following technological explanations of its use and a schematic description, with reference to Figure 1, schematically depicting a block diagram of an individual assistant with elements of artificial intelligence - individual intelligence assistant (IIP).

The main function used by the IIP is to compare texts. It consists in the fact that the texts are compared among themselves in a set-theoretical sense — the common part (intersection), the words,

relating only to the first and only to the second texts. Text comparison allows you to highlight the meaning of questions asked by the user and to realize user preferences by comparing preferences with the answers found.

The measure of text similarities is calculated as follows - the similarity S

S = 2P / (tl + t2), where P is the size of the intersection of texts,

tl - the number of words in the first text,

t2 is the number of words in the second text,

For example, if the texts are equal, then

S = 2t / t + t = l

Let the user ask the IIP a question:

Where is JV »51 pharmacy located?

Then, to highlight the meaning, the IIP compares the text of the question with the texts of the question templates:

Where is

Where can I buy

Who it

The name of

When it will be

For the first question (two words coincide) S = 2x2 / 4 + 2) = 2/3, for the second S = 1x2 / 4 + 2) = 1/3, for the rest the match is 0.

Thus, the IIP will determine that a question has been asked about the location (first template) and highlight what needs to be sought (pharmacy J e51) - words that are contained only in the first text. Suppose there is no result in the data array of answers to the user and a search is performed in a search engine.

Search results:

1. Pharmacy N ° 51 network of pharmacies A5

Address: Moscow, Dmitrov highway, 45/2

Phone: 8-495-916-86-47

Opening hours: 9.00-21.00

Metro: Petrovsko-Razumovskaya

2. Address

125430, Moscow, Mitinskaya street, 40 shopping center "Mitino", floor 1

Phone 8-495-759-37-29

Opening hours 8.00-23.00 3. Address: Lipetsk, st. Belana, d.26, shopping center "Real"

Phone: +7 (4742) 45-1668, +7 (4742) 45-3607

Opening hours: Daily 9.00 - 21.30

If the user's city of location Moscow is set in preferences, the comparison will select only the first two search results and pass them to the user.

Next, we illustrate the operation of the IIP with an array of events that describe information about past or upcoming events.

Suppose the current day is March 7th.

IIP will prepare a notification to the user:

Tomorrow March 8, congratulations to women!

Let the user ask a clarifying question:

And what did I give last year?

The IIP will search the data array of events a year ago and find the following data:

Gave last year:

Women at work - sweets “Evening ringing”

Mom - money

Kate's girlfriend - bracelet

To the next user question

Where is better to buy gifts for March 8

According to a similar comparison scheme, a search query (gifts for March 8) will be highlighted and a search will be performed:

Its results, for example, will be the name of the site:

http://www.dostavka-podarkov.ru/holidays/03_08/

Gifts for women on March 8th.

This year, our online store prepared new gifts and souvenirs for women on March 8. We hope that among the rich assortment of gifts for March 8, you can find and order a suitable gift. Attention! Order gifts for March 8 in advance, and not on the last day! We also remind you that we are ready to work with wholesale and corporate customers of gifts and souvenirs on March 8. Delivery of retail and wholesale orders is carried out throughout the Russian Federation.

In our online store a wide range of gifts for women is always presented.

After which the IIP after some time will ask a question intended for its self-study:

What did you choose for gifts this time?

and remember the data entry, in order to provide it to the user if necessary.

Thus, an individual assistant with elements of artificial intelligence, designed for dialogue with the user (1) by exchanging questions and answers, which are texts in natural language and forming tips for the user in the form of texts in natural language, consists of components that perform operations, implemented on

arbitrary computing resources, and the components that perform operations include at least the following: a text recognition component entered by the user (2), a search component in data arrays (3), a component for generating responses for the user (1 1), a feedback component that evaluates responses, formed for the user (12), the settings component that controls the formation of answers (9), the communication component with other individual assistants (8), the communication component with search engines (10), and an array of data, including, but not limiting the following: user questions data array (4), user response data array (5), settings data array (6) containing information about preferences

user and individual assistant operation mode, an array of events describing information about past or upcoming events (7).

• the user generates questions that enter the text recognition component entered by the user, which in turn passes them to the search component in the data arrays, after which the search component in the data arrays compares them with the array response data to the user, and if there is an identical or similar answer, he is sent to the response generation component for the user,

• in the event that a user’s question forms an event,

event information an array of events describing information about past or upcoming events,

• in the event that an identical or similar answer in the array

no answer data found, the text recognition component accesses the communication component with other individual

assistants, who passes the question to another individual assistant and expects an answer from him, another individual assistant functions similarly,

• if no response has been received from another assistant, or

access to other helpers is impossible or forbidden by the settings component, the text recognition component entered by the user selects the content of the question in the form of text, after which this text is transmitted to the communication component with search

systems, the response from which is sent to the response generation component for the user, in which the response is compared with the settings data array and given this comparison is transmitted to the user,

• the user, having received the response, does not necessarily evaluate it, for which the user’s response is transmitted to the feedback component, and the response is placed in the response data array with the user's rating, or placed in the event array,

• the component of forming answers for the user independently, regardless of the user's questions, forms tips for the user in the case when from other individual

the helpers receive a question, the question itself or the answer to which is indicated in the settings data array as interesting for the user, as well as in the case when previously asked questions require an answer in the time has come, or an event has occurred in an array of events.

Let us explain the work of IIP with additional examples. In 1950, Alan Matheson Turing published an article entitled “Can the machine think?”. In it, Turing examined two fundamental definitions - “machine” and “think”. But if the scientist’s thoughts on the first problem were relevant mainly for his contemporaries, then the Turing criterion of the reasonableness of computer systems developed by Turing is still used.

To identify electronic intelligence, he proposed the so-called "simulation game." Initially, it involves three people — a man, a woman, and a judge. The latter is separated from the other two by the walls of the room, everyone communicates by telegraphic communication (to exclude voice recognition, in the present case it will be communication by entering texts from the keyboard). The goal of the questioner is to find out which of the other participants is a woman and which is a man. The goal of the game is for the woman to help the judge, and for the man to confuse him.

The next step was to get a car to answer questions instead of a man. Will the questioner in this case be mistaken as often as in a game where only people are participants? Alan Turing

predicted that in a five-minute test, a computer with about 120 MB of memory could trick a person in 30% of cases. Modern semantic technologies make it possible to raise the Turing bar to 70 percent or more of cases.

Artificial Intelligence Assistant (IIP) is technically implemented as an application, which is a trained pseudo-rational system connected in the Internet space with other intelligence assistants. IIP has

an extensible modular architecture that includes the core and additional modules that connect to the IIP via a standard interface and

process data in standard or documented formats.

Thus, the proposed project aims at creating a personal assistant and an interlocutor who, based on fully deterministic semantic algorithms, can solve as optimization problems intellectual work, including the systematization, training and accumulation of knowledge, and maintain a meaningful dialogue and be capable of learning.

Modular architecture has significant financial and

the strategic potential of the project, which will attract a large number of external free and commercial developers.

In addition, it is very important to have an integrating Internet resource that connects the IPS currently located in the network into a single mega-system that translates the needs of IIP users and allows you to take into account use cases and create constantly updated and

dynamic project.

Important addition modules include:

- a text generation module that allows you to generate texts from the words of the user's daily dictionaries,

- image indexing module.

Schematically, the operation of the IIP can be further illustrated as follows:

- the user (subject “I”), textually or by voice, through the input and recognition unit of semantic constructions, introduces in an arbitrary form questions or narrative sentences (hereinafter referred to as “requests” or “user requests”) intended for processing by an artificial intelligence assistant (which is the subject “You” or subject “name”, which is set at the settings stage);

- the introduced construction is compared with the dialogue available in the database and, if there are close constructions, the subject “I” is informed about the dialogue that took place earlier (“you already asked something similar ...”, in the absence of similar constructions, the request is analyzed according to the following parameters:

- are there query words in the collection of semantic constructions and if all words are in the collection, then the meaning of the query is revealed - the mentioned objects and the functions that connect them, are highlighted, for example, “Where am I going now?” - The key question is “where”, “I” - attitude to the user, “now” is real moment of time, “food” - a little-informative component, equivalent to “going,” “being.”

In the absence of query words in the collection of semantic constructions, the IIP clarifies them with the user and modifies the collection with answers

semantic constructions, i.e. The IIP learning stage is also taking place in the form of dialogue. For example, if the user enters “I am bored,” the IIP may ask with the question “What is“ bored? ”And accumulate the words that accompany the explanations in the collection of explanatory constructions.

The user can answer that “boring” - “when nothing pleases”, “nothing to do” and then the IIP offers “What to do so that you are not bored?” And the subject “I” can respond with different options “Nothing”, “Tell me a fairy tale ”,“ Tell a joke ”,“ Tell something funny. ” All four options will be processed by the IIP, in the first case the III will clarify why nothing needs to be done and “I” can answer with options, for example - “I am busy”, “I don’t know what I want.” In the second case, the IIP searches the collection of local

results for the presence of a previously told tale with a tag

“Ironic” or “funny” and offers it to the user, and in the event of his refusal, searches and compares what was found with the local results collection already in the collection and offers the user a fairy tale that has no analogues in the collection of local results and, with the consent of the user, offers him the text of the fairy tale for reading or reading it out loud and with approval

(listening to or reading to the end or most) puts it in a collection of local results. The third case is handled similarly to the second, the fourth continues the dialogue with the user clarification "What do you want to hear - poems, a fairy tale, a joke?"

In the case of requests from IIP to search engines (first of all, dropping uninformative components), it is converted to

a formal request to the search system through the interface with the search system (in the examples given, to the coordinate or location determination system for the question “Where am I going?” and to entertainment resources if the situation “I am bored” is resolved). The following situation is regulated by the IIP settings: a user query can be transformed into one search query, or into several search queries into one search system, or into several search queries into different search engines, and then the results are compared in the post-processing block of the search results.

Further, the data sampling unit, taking into account preferences, transfers the results of post-processing to the user, taking into account his preferences or given

emotional parameters, for example, the search result of a certain text can be correlated with the desire of the subject “I” to see the found text “ironic”, “funny”, “kind”, or filtered by the subject of “prose”, “poems”, “jokes”, or text similar to what the user has already read before. So in the given example, the replica “I'm bored” sets the current state of the settings to more ironic or optimistic IIP replicas, while the dictionary content is selected from the collection of case-law knowledge on a set of words in the topics “funny”, “ironic”, “sad”.

Settings and preferences can also be set in the form of queries, for example, "I want to find scientific articles on" or "I want to read romantic poems."

The subject “I” can evaluate the result of the IIP (for example, “you didn’t find what you needed” or “the first and fifth links were most useful”) and thereby change the priorities of the search and work in general, or clarify these results. User-accepted responses are placed in a collection of local results (in the form of flat text or links).

In addition, the IIP acts as a secretary, recalls regular events, for example, “You need to choose a gift for March 8th. Last time you gave ... "

To improve the quality of the IIP, a constant analysis of the text entered by the user in all applications, including the sms set, can be carried out (if the IIP or part of it is installed within the mobile

devices) and a daily dictionary is technically included in the collection of case-law knowledge - words that the user adds to his primary vocabulary. This dictionary serves as an assessment user (including his state of health), and for tracking his activity and initiating dialogue.

In addition, indexing of user-created documents, letters and dialogs on social networks is performed for their quick search, correction of preferences, updating the daily dictionary. The user may also be offered a description or annotation of the images viewed, created and selected by him. In this case, images or links to them

are placed in a collection of local results.

Given the implementation of the accumulation of knowledge and training within 2-3 months, the IIP turns into an integral alter-ego of the user and allows

minimize technical issues of search and selection of information, as well as successfully automatically represent the user in communication groups and social networks using the text generation module. Besides,

a local collection of information resources most needed by the user, independent of the availability of global networks, is automatically organized.

Compared to other ways of communicating with computer systems, the complexity of compiling search queries is reduced, since the questions are formulated in a natural language and an adequate choice is made

search results, it is possible to learn and accumulate knowledge, take into account the experience and preferences of the user, an active dialogue is conducted between the system and the user, including messages initiated by the system.

Claims

CLAIM

1. An individual assistant with elements of artificial intelligence, designed to conduct a dialogue with the user by exchanging questions and answers, which are texts in natural language and forming tips for the user in the form of texts in natural language, consisting of components that perform operations, implemented in arbitrary

computing resources, and the components that perform the operations include at least the following: a text recognition component entered by the user, a search component in the data arrays, a component for generating responses for the user, a feedback component that evaluates the responses generated for the user, a setting component that controls

forming answers, a communication component with other individual assistants, a communication component with search engines, and data arrays, including but not limited to the following: user question data array, user response data array, settings data array containing information about the user's preferences and individual operating mode assistant, an array of events that describes information about past or upcoming events.

2. The way to use an individual assistant with elements

artificial intelligence according to claim 1, in the case where it is optional

at least two individual assistants are functioning, consisting in the fact that:

a) the user generates questions that enter the text recognition component entered by the user, which in turn passes them to the search component in the data arrays, after which the search component in the data arrays compares them with the data array of the answers to the user and if there is an identical or similar answer it is sent to the response generation component for the user, b) if the user’s question forms an event, information about the event is an array of events that describes information about past or upcoming events,

an individual assistant functions similarly,

the content of the question is in the form of text, after which this text is transmitted to the communication component with search engines, the response from which is sent to the response generation component for the user, in which the response is compared with the settings data array and given this comparison is transmitted to the user,