CN111858840A - Method and device for intent recognition based on concept map - Google Patents

Abstract

The present invention belongs to the field of intention recognition, and discloses an intention recognition method and device based on a concept map, the method comprising: collecting user voice information; performing grammatical analysis on the voice information, extracting question words and keywords in the voice information; matching the question words with nodes in a question concept map to obtain matching nodes, wherein the question concept map is pre-constructed; obtaining the user's intention according to the intention concepts corresponding to the matching nodes and the keywords. The present invention generates a question concept map from various random questions, so that when obtaining the user's intention in the generated question concept map according to the question words in the voice information, it is easier and more accurate to obtain the user's intention, and in the question concept map, the various question levels are clear, making the search and recognition speed faster.

Description

Method and device for intent recognition based on concept map

technical field

The invention belongs to the technical field of semantic recognition, and in particular relates to a method and device for intention recognition based on a concept map.

Background technique

With the rapid development of smart terminals and network technologies, people are more and more accustomed to using smart terminals to fulfill various needs. As the most convenient and natural way for humans to express their thoughts, natural language has gradually become the most mainstream human-computer interaction method in the field of intelligent services.

In the human-computer interaction scenario, intent recognition is an indispensable link. It mainly analyzes the voice input by the user, understands the user's intent, converts it into a structured data format that the machine can understand, and then gives corresponding feedback. Therefore, in human-computer interaction scenarios, accurate identification of user intent is the basis for making correct responses. For students in lower grades, because they are in the stage of language learning, there may be situations such as incomplete language expression and randomness in the process of language expression. Therefore, in the voice electronic products used by students, if the conventional intention recognition method is used, the intention recognition is likely to be inaccurate, so that the intelligent terminal cannot make a correct response, which affects the user experience.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method and device for recognizing intent based on a concept map, which generates a concept map from various random questions, so as to obtain user intent more easily and more accurately.

The technical scheme provided by the present invention is as follows:

In one aspect, a method for recognizing intent based on a concept map is provided, including:

Collect user's voice information;

Syntax analysis is carried out to the voice information, and the question words and keywords in the voice information are extracted;

Matching the question word with the node in the question concept graph to obtain a matching node, wherein the question concept graph is pre-built;

According to the intent concept corresponding to the matched node and the keyword, the user's intent is acquired.

Further preferably, the method for constructing the concept map of the question method is:

Collect a large number of user corpora;

extract the question words in the corpus;

Obtain the intent concept of the question word with the same intent in the corpus;

establishing a mapping relationship between the question word and the intent concept;

According to the mapping relationship between the question word and the intent concept, a question concept map is constructed.

Further preferably, acquiring the user's intent according to the intent concept corresponding to the matched node and the keyword specifically includes:

Search the target word matching the keyword from the preset thesaurus;

According to the semantics of the target word, determine the semantics of the keyword;

The user's intent is acquired according to the intent concept corresponding to the matched node and the semantics of the keyword.

Further preferably, the voice information is grammatically analyzed, and after extracting the question words and keywords in the voice information, it also includes:

When the keyword contains a specific keyword, receiving a touch signal from the user;

obtaining image information of the user-selected area according to the touch signal;

The acquiring the user's intent according to the intent concept corresponding to the matched node and the keyword specifically includes:

The user's intent is acquired according to the intent concept corresponding to the matched node, the keyword, and the image information.

Further preferably, acquiring the user's intent according to the intent concept corresponding to the matched node, the keyword, and the image information specifically includes:

Identify text information in the image information;

replacing specific keywords in the keywords with the text information;

The user's intent is acquired according to the intent concept corresponding to the matched node, the keyword, and the text information.

On the other hand, a device for recognizing intent based on a concept map is also provided, comprising:

The acquisition module is used to collect the user's voice information;

Extraction module, for carrying out grammatical analysis to described speech information, extracts the question word and key word in described speech information;

a matching module, configured to match the question word with the node in the question concept graph to obtain the matched node, wherein the question concept graph is pre-built;

An intent acquisition module, configured to acquire the user's intent according to the intent concept corresponding to the matched node and the keyword.

Further preferably, it also includes building blocks;

The building blocks include:

The corpus collection unit is used to collect a large number of user corpora;

a word extraction unit, used to extract the question words in the corpus;

a concept acquisition unit, used to acquire the intent concept of the question word with the same intent in the corpus;

a relationship establishing unit, configured to establish a mapping relationship between the question word and the intent concept;

The construction unit is used for constructing a question concept map according to the mapping relationship between the question word and the intention concept.

Further preferably, the intent acquisition module includes:

A word search unit, used to search for target words matching the keyword from a preset thesaurus;

a semantic determination unit, configured to determine the semantics of the keyword according to the semantics of the target word;

An intent acquisition unit, configured to acquire the user's intent according to the intent concept corresponding to the matched node and the semantics of the keyword.

Further preferably, it also includes:

a signal receiving module for receiving a touch signal of the user when the keyword contains a specific keyword;

an image acquisition module, configured to acquire image information of the user-selected area according to the touch signal;

The intent acquisition module is further configured to acquire the user's intent according to the intent concept corresponding to the matched node, the keyword and the image information.

Further preferably, the intent acquisition module includes:

an identification unit for identifying text information in the image information;

A replacement unit, used to replace the specific keywords in the keywords with the text information;

An intent acquisition unit, configured to acquire the user's intent according to the intent concept corresponding to the matched node, the keyword, and the text information.

Compared with the prior art, the present invention provides a method and device for recognizing intent based on a concept map, which has the following beneficial effects: It is easier and more accurate for French words to obtain user intentions in the generated question concept map, and in the question concept map, the levels of various questions are clear, which makes the search and recognition faster.

Description of drawings

The preferred embodiments will be described below in a clear and easy-to-understand manner with reference to the accompanying drawings, and further description will be given of the above-mentioned characteristics, technical features, advantages and implementations of a method and device for identifying intent based on a concept map.

FIG. 1 is a schematic flowchart of a first embodiment of a method for recognizing intent based on a concept map of the present invention;

2 is a schematic flowchart of a second embodiment of a conceptual map-based intent identification method of the present invention;

3 is a schematic flowchart of a third embodiment of a conceptual map-based intent identification method of the present invention;

4 is a schematic flowchart of a fourth embodiment of a conceptual map-based intent recognition method of the present invention;

5 is a schematic flowchart of a fifth embodiment of a conceptual map-based intent recognition method of the present invention;

FIG. 6 is a schematic structural block diagram of an embodiment of an intention recognition device based on a conceptual diagram of the present invention;

FIG. 7 is a schematic structural block diagram of another embodiment of an intention recognition device based on a conceptual diagram of the present invention.

Explanation of reference numerals

100, acquisition module; 200, extraction module; 300, matching module; 400, intent acquisition module; 410, word search unit; 420, semantic determination unit; 430, intent acquisition unit; 440, identification unit; 450, replacement unit; 500 510, corpus collection unit; 520, word extraction unit; 530, concept acquisition unit; 540, relationship establishment unit; 550, construction unit; 600, signal receiving module; 700, image acquisition module.

Detailed ways

In order to more clearly describe the embodiments of the present invention or the technical solutions in the prior art, the specific embodiments of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative efforts, and obtain other implementations.

In order to keep the drawings concise, the drawings only schematically show the parts related to the present invention, and they do not represent its actual structure as a product. In addition, in order to make the drawings concise and easy to understand, in some drawings, only one of the components having the same structure or function is schematically shown, or only one of them is marked. As used herein, "one" not only means "only one", but also "more than one".

In the present invention, the intent recognition is carried out through the concept map, which can improve the accuracy of the intent recognition; the intent recognition method based on the concept map of the present invention can be applied to intelligent terminal equipment; For the convenience of understanding, all explanations take the tutoring machine as the subject, but those skilled in the art should understand that the intent recognition method based on the concept map can also be applied to other intelligent terminal devices, as long as the corresponding functions can be realized.

According to the first embodiment provided by the present invention, as shown in FIG. 1 , a method for recognizing intent based on a concept map includes:

S100 collects user's voice information;

Specifically, the user's voice may be collected through a microphone or other voice collection devices. Microphones and other voice collection devices may be built-in devices in the tutoring machine, or may be external devices. The voice information may be voice input by the user in real time.

S200 performs grammatical analysis on the voice information, and extracts question words and keywords in the voice information;

Specifically, after acquiring the voice information, the voice information may be converted into text information first, and then word segmentation and grammatical analysis are performed on the converted text information to extract question words and keywords from the text information. Converting the voice information into text can be achieved by using various existing technical means. The present invention is not limited to using a certain conversion method, and will not be described in detail here.

The question word here refers to the word expressing various questions in the process of voice interaction, such as "how to do", "how to solve", "I don't understand" and so on. The keywords here refer to the nouns with actual meaning in the voice information, that is, the nouns with the corresponding entities; for example, "Chinese video" in "Where is the Chinese video" is the keyword, and "What to do with the connection and division problem?" "Consistently remove the problem" is the key word, and "conscientiously and conscientiously" in "conscientiously and conscientiously" is the key word.

S300 matches the question word with the nodes in the question concept graph to obtain a matching node, wherein the question concept graph is pre-built;

Specifically, a concept map is a graphical method that uses nodes to represent concepts and lines to represent relationships between concepts. Concept maps are generally composed of "nodes", "links" and "related text annotations". The concept map of questions refers to the concept map generated by various questions according to the relationship between the questions. In the concept diagram of the question method, the root node contains multiple sub-nodes, and the sub-nodes are the subordinate concepts of the root node. The sub-nodes can contain multiple sub-sub-nodes, and the sub-sub-nodes are sub-concepts of the sub-nodes.

For example, the root node of the questioning concept graph contains multiple questioning concepts, such as the questioning method for "explaining", the questioning method for "searching", and the questioning method for "adjustment".

Among them, different questioning concepts include multiple questioning methods, for example, the questioning methods expressing "explaining" include "do not understand", "how to understand", "how to do", "what is" and so on. Questions that express "find" include "where", "where", "where" and so on.

Match the question words extracted from the language information with the nodes in the question concept graph, that is, search for the nodes matching the question words in the question concept graph.

S400 acquires the user's intent according to the intent concept corresponding to the matched node and the keyword.

Specifically, after obtaining the matching node, the corresponding intent concept can be obtained according to the matching node, and then the user's intent can be obtained according to the corresponding intent concept and combining keywords.

Exemplarily, the speech information is "what is a linear equation in one variable", the question word extracted from the speech information is "what is", the extracted keyword is "a linear equation in one variable", and the question word is in the concept of question The matching node in the graph is "what is", and the intent concept of the node "what is" in the question concept graph is "explain". Combining the keywords "one-dimensional linear equation" and "explain", the user's intent can be obtained as Search for content related to linear equations in one variable, such as searching for explanations of linear equations in one variable, practice questions for linear equations in one variable, etc.

In this embodiment, various random question methods are used to generate a question method concept map, so that when the user intent is obtained in the generated question method concept map according to the question method words in the voice information, it is easier and more accurate to obtain the user intent, and In the question method concept map, the various question method levels are clear, which makes the search and identification faster.

According to the second embodiment provided by the present invention, as shown in FIG. 2 , a method for recognizing intent based on a concept map, this embodiment is based on the first embodiment, and the method for constructing the concept map is as follows:

S010 collects a large number of user corpora;

Specifically, in the process of the user using the tutoring machine, various corpora generated when the user performs voice interaction with the tutoring machine are collected.

S020 extracts the question words in the corpus;

Specifically, after a large amount of corpus is collected, the corpus can be converted into text information first, and then the converted text information can be segmented and grammatically analyzed, and then the question words related to the question method can be extracted from the converted text information.

For example, the corpus is "what to do with the problem of connecting and removing", "I don't know how to move graphics", "I don't know how to work hard", "where is the semantic video", "help me open the language practice questions", "the volume is too low" , "The volume is too loud", "The font is too small", "I want to preview the language", "Open the history textbook" and so on. The question words extracted from the corpus are "what to do", "I don't know", "I don't understand", "where", "open", "too small", "too big", "preview" "Wait.

S030 obtains the intent concept of the question word with the same intent in the corpus;

Specifically, after extracting various question words from the corpus, classify the question words, classify the question words with the same intention into one category, and then obtain the intent concept of the question words with the same intention.

For example, the French words "what to do", "I don't know" and "I don't understand" all indicate that you don't understand a certain content, don't know how to answer it, don't understand, etc. You need a tutor to search and display the corresponding answers and questions. Lectures or handouts, etc.; therefore, questions such as "what to do", "I don't know", "I don't understand", etc. can be grouped together.

The French question words "where", "open", and "preview" all indicate that a certain content needs to be found, and the corresponding content needs to be opened by the tutor computer. Therefore, the French question words "where", "open", "preview" can be used " etc. are grouped together.

The French words "too small" and "too big" all indicate that an object needs to be adjusted, such as adjusting the volume size, adjusting the font size, adjusting the video display size, etc.; therefore, the French word "too small" can be , "too big" are grouped together.

After classifying all the question words obtained, the intention concept of each type of question word after classification is determined according to the intention expressed by the question word; , the intention concept of "I don't understand" is adjusted to "explain", the intention concept of the question words "where", "open" and "preview" are adjusted to "find the object", and the question word is too small", The intent concept of "too big" is conditioned to 'conditioned'.

S040 establishes a mapping relationship between the question word and the intent concept;

Specifically, after classifying the question words and determining the intent concept of each type of question word, the mapping relationship between the question word and the intent concept is established, that is, the corresponding relationship between the intent concept and the question word is established.

S050 constructs a question concept map according to the mapping relationship between the question word and the intent concept.

Specifically, according to the established correspondence between the question word and the intent concept, the question concept map can be constructed.

In this embodiment, by collecting the corpus of the user when using the intelligent terminal device, a large number of random and colloquial questions can be collected, and then the random questions can be classified and sorted to generate a concept map of the questions, so that the When the concept map is asked for intent recognition, the recognition rate is higher.

According to a third embodiment provided by the present invention, as shown in FIG. 3 , a method for recognizing intent based on a concept map includes:

S100 collects user's voice information;

S200 performs grammatical analysis on the voice information, and extracts question words and keywords in the voice information;

S300 matches the question word with the nodes in the question concept graph to obtain a matching node, wherein the question concept graph is pre-built;

S410 searches for a target word matching the keyword from a preset thesaurus;

Specifically, when acquiring the intent of keywords, a thesaurus can be established first. The keywords in the thesaurus are noun words, and the keywords in the thesaurus can be crawled from the network through the crawler technology, and the words in the thesaurus can be obtained. semantics. Then, the keywords extracted from the speech information are matched with the thesaurus in the thesaurus to find the matching target words.

S420 determines the semantics of the keyword according to the semantics of the target word;

Specifically, after finding the target word, determine the semantics of the keywords in the speech information according to the semantics corresponding to the target words, that is, determine the semantics of the keywords in the speech information, and convert them into a structured data format that can be understood by machines.

S430 acquires the user's intent according to the intent concept corresponding to the matched node and the semantics of the keyword.

Specifically, after obtaining the semantics of the keywords and the intent concept, the user's intent can be acquired according to the intent concept and the semantics of the keywords.

For example, if the voice information is "how to do with the question of consecutive division", the keyword extracted from it is "question of consecutive division", and the question word is "how to do". According to the concept map of the question, the matching word "how to do" is obtained. The corresponding questioning concept of the node is “Explanation”, and the semantics of the keyword “Concatenating and Excluding Questions” is “Concatenating and Excluding Questions”. According to the questioning concept “Explanation” and the keyword “Concatenating and Excluding Questions”, the user’s intention can be obtained. For "explain the problem of concatenation and division".

In this embodiment, the semantics of the keywords are acquired by establishing a thesaurus and using the keyword matching method, which is more convenient and quicker than other semantic recognition methods.

According to a fourth embodiment provided by the present invention, as shown in FIG. 4 , a method for recognizing intent based on a concept map includes:

S100 collects user's voice information;

S200 performs grammatical analysis on the voice information, and extracts question words and keywords in the voice information;

S210, when the keyword contains a specific keyword, receive a touch signal from the user;

Specifically, the specific keywords refer to pronouns such as "this word", "this question", "this sentence", "this", "this" and the like. When the above-mentioned specific keyword is included in the voice information, the user's touch signal on the touch screen is received. The touch signal may be a continuous sliding touch signal by the user on the touch screen, such as a straight line formed by continuous touch, or a circular frame, oval frame, irregular graphic frame, etc. formed by continuous touch.

S220, acquiring image information of the user-selected area according to the touch signal;

Specifically, after receiving the touch signal, the image information of the area selected by the user can be obtained according to the touch signal. For example, after continuous touches form a straight line, a rectangular frame is constructed with the straight line as the diagonal, and the image in the rectangular frame is intercepted. , the image in the rectangular frame is the image information of the area selected by the user.

If the continuous touch forms a circular frame or an oval frame, etc., the image in the formed frame is intercepted, and the image in the frame is the image information of the area selected by the user.

Here, the touch signal is limited to a continuous touch signal, which can prevent erroneous operations and reduce the error rate. For example, it can prevent the user from forming two touch points on the touch screen due to erroneous operations, and obtain the frame formed by the two points. image inside.

When the voice information contains specific keywords, it means that the voice information input by the user is missing, such as "how to solve this problem", the tutor does not know what the problem is, so even if the voice information is semantically parsed Afterwards, the exact intention of the user cannot be obtained. Therefore, when it is determined that the voice information contains a specific keyword, it is necessary to further obtain the image information of the area pointed by the user's finger or framed to obtain the information of a certain question.

S300 matches the question word with the nodes in the question concept graph to obtain a matching node, wherein the question concept graph is pre-built;

S440 acquires the user's intent according to the intent concept corresponding to the matched node, the keyword, and the image information.

Specifically, according to the acquired intent concept and the semantics of the keywords, combined with the acquired image information, the three are integrated to obtain the user's true intent, so as to give corresponding feedback.

According to a fifth embodiment provided by the present invention, as shown in FIG. 5 , a method for recognizing intent based on a concept map includes:

S100 collects user's voice information;

S200 performs grammatical analysis on the voice information, and extracts question words and keywords in the voice information;

S210, when the keyword contains a specific keyword, receive a touch signal from the user;

S220, acquiring image information of the user-selected area according to the touch signal;

S300 matches the question word with the nodes in the question concept graph to obtain a matching node, wherein the question concept graph is pre-built;

S441 identifies text information in the image information;

Specifically, after the image information is obtained, the image is processed and identified to obtain text information in the image information, where the text information is information about a certain question or information about a certain word.

S442 replaces specific keywords in the keywords with the text information;

S443 acquires the user's intent according to the intent concept corresponding to the matched node, the keyword, and the text information.

Specifically, the acquired text information is replaced with a specific keyword, that is, the question is replaced with the acquired text information. Then, according to the intent concept corresponding to the matched node, the semantics of the remaining keywords, and the acquired text information, the real intent of the user is obtained.

For example, when a user reads extracurricular readings on the home teaching machine, when he sees "conscientious and conscientious", he does not recognize "conscientious and conscientious" and wants to ask for help, so the user will say "what do these words read", the home teaching machine collects the voice information, When it is judged that the voice information contains a specific keyword, the image information of the area selected by the user will be obtained according to the touch signal. Combined with the question word "what to read", the real semantics of the user can be obtained as "what to read conscientiously". In this way, words that the user does not know or questions that cannot be answered can be helped through the tutoring machine to assist the user's learning.

According to the sixth embodiment provided by the present invention, as shown in FIG. 6 , a device for recognizing intent based on a concept map includes:

The collection module 100 is used to collect the voice information of the user;

Specifically, the user's voice may be collected through a microphone or other voice collection devices. Microphones and other voice collection devices may be built-in devices in the tutoring machine, or may be external devices. The voice information may be voice input by the user in real time.

Extraction module 200, for performing grammatical analysis on the voice information, and extracting question words and keywords in the voice information;

Specifically, after acquiring the voice information, the voice information may be converted into text information first, and then word segmentation and grammatical analysis are performed on the converted text information to extract question words and keywords from the text information. Converting the voice information into text can be achieved by using various existing technical means. The present invention is not limited to using a certain conversion method, and will not be described in detail here.

The question word here refers to the word expressing various questions in the process of voice interaction, such as "how to do", "how to solve", "I don't understand" and so on. The keywords here refer to the nouns with actual meaning in the voice information, that is, the nouns with the corresponding entities; for example, "Chinese video" in "Where is the Chinese video" is the keyword, and "What to do with the connection and division problem?" "Consistently remove the problem" is the key word, and "conscientiously and conscientiously" in "conscientiously and conscientiously" is the key word.

The matching module 300 is configured to match the question word with the node in the question concept graph to obtain the matched node, wherein the question concept graph is pre-built;

Specifically, a concept map is a graphical method that uses nodes to represent concepts and lines to represent relationships between concepts. Concept maps are generally composed of "nodes", "links" and "related text annotations". The concept map of questions refers to the concept map generated by various questions according to the relationship between the questions. In the concept diagram of the question method, the root node contains multiple sub-nodes, and the sub-nodes are the subordinate concepts of the root node. The sub-nodes can contain multiple sub-sub-nodes, and the sub-sub-nodes are sub-concepts of the sub-nodes.

For example, the root node of the questioning concept graph contains multiple questioning concepts, such as the questioning method for "explaining", the questioning method for "searching", and the questioning method for "adjustment".

Among them, different questioning concepts include multiple questioning methods, for example, the questioning methods expressing "explaining" include "do not understand", "how to understand", "how to do", "what is" and so on. Questions that express "find" include "where", "where", "where" and so on.

Match the question words extracted from the language information with the nodes in the question concept graph, that is, search for the nodes matching the question words in the question concept graph.

The intent acquisition module 400 is configured to acquire the user's intent according to the intent concept corresponding to the matched node and the keyword.

Specifically, after obtaining the matching node, the corresponding intent concept can be obtained according to the matching node, and then the user's intent can be obtained according to the corresponding intent concept and combining keywords.

Exemplarily, the speech information is "what is a linear equation in one variable", the question word extracted from the speech information is "what is", the extracted keyword is "a linear equation in one variable", and the question word is in the concept of question The matching node in the graph is "what is", and the intent concept of the node "what is" in the question concept graph is "explain". Combining the keywords "one-dimensional linear equation" and "explain", the user's intent can be obtained as Search for content related to linear equations in one variable, such as searching for explanations of linear equations in one variable, practice questions for linear equations in one variable, etc.

In this embodiment, various random question methods are used to generate a question method concept map, so that when the user intent is obtained in the generated question method concept map according to the question method words in the voice information, it is easier and more accurate to obtain the user intent, and In the question method concept map, the various question method levels are clear, which makes the search and identification faster.

Preferably, a building module 500 is also included;

Building blocks 500 include:

a corpus collection unit 510, used to collect a large number of user corpora;

Specifically, in the process of the user using the tutoring machine, various corpora generated when the user performs voice interaction with the tutoring machine are collected.

A word extraction unit 520, used for extracting question words in the corpus;

Specifically, after a large amount of corpus is collected, the corpus can be converted into text information first, and then the converted text information can be segmented and grammatically analyzed, and then the question words related to the question method can be extracted from the converted text information.

For example, the corpus is "what to do with the problem of connecting and removing", "I don't know how to move graphics", "I don't know how to work hard", "where is the semantic video", "help me open the language practice questions", "the volume is too low" , "The volume is too loud", "The font is too small", "I want to preview the language", "Open the history textbook" and so on. The question words extracted from the corpus are "what to do", "I don't know", "I don't understand", "where", "open", "too small", "too big", "preview" "Wait.

a concept acquiring unit 530, configured to acquire the intent concept of the question word with the same intent in the corpus;

Specifically, after extracting various question words from the corpus, classify the question words, classify the question words with the same intention into one category, and then obtain the intent concept of the question words with the same intention.

For example, the French words "what to do", "I don't know" and "I don't understand" all indicate that you don't understand a certain content, don't know how to answer it, don't understand, etc. You need a tutor to search and display the corresponding answers and questions. Lectures or handouts, etc.; therefore, questions such as "what to do", "I don't know", "I don't understand", etc. can be grouped together.

The French question words "where", "open", and "preview" all indicate that a certain content needs to be found, and the corresponding content needs to be opened by the tutor computer. Therefore, the French question words "where", "open", "preview" can be used " etc. are grouped together.

The French words "too small" and "too big" all indicate that an object needs to be adjusted, such as adjusting the volume size, adjusting the font size, adjusting the video display size, etc.; therefore, the French word "too small" can be , "too big" are grouped together.

After classifying all the question words obtained, the intention concept of each type of question word after classification is determined according to the intention expressed by the question word; , the intention concept of "I don't understand" is adjusted to "explain", the intention concept of the question words "where", "open" and "preview" are adjusted to "find the object", and the question word is too small", The intent concept of "too big" is conditioned to 'conditioned'.

a relationship establishing unit 540, configured to establish a mapping relationship between the question word and the intent concept;

Specifically, after classifying the question words and determining the intent concept of each type of question word, the mapping relationship between the question word and the intent concept is established, that is, the corresponding relationship between the intent concept and the question word is established.

The constructing unit 550 is configured to construct a question concept map according to the mapping relationship between the question word and the intent concept.

Specifically, according to the established correspondence between the question word and the intent concept, the question concept map can be constructed.

In this embodiment, by collecting the corpus of the user when using the intelligent terminal device, a large number of random and colloquial questions can be collected, and then the random questions can be classified and sorted to generate a concept map of the questions, so that the When the concept map is asked for intent recognition, the recognition rate is higher.

Preferably, the intent acquisition module 400 includes:

A word search unit 410, configured to search for a target word matching the keyword from a preset thesaurus;

Specifically, when acquiring the intent of keywords, a thesaurus can be established first. The keywords in the thesaurus are noun words, and the keywords in the thesaurus can be crawled from the network through the crawler technology, and the words in the thesaurus can be obtained. semantics. Then, the keywords extracted from the speech information are matched with the thesaurus in the thesaurus to find the matching target words.

a semantic determination unit 420, configured to determine the semantics of the keyword according to the semantics of the target word;

Specifically, after finding the target word, determine the semantics of the keywords in the speech information according to the semantics corresponding to the target words, that is, determine the semantics of the keywords in the speech information, and convert them into a structured data format that can be understood by machines.

The intent acquiring unit 430 is configured to acquire the user's intent according to the intent concept corresponding to the matched node and the semantics of the keyword.

Specifically, after obtaining the semantics of the keywords and the intent concept, the user's intent can be acquired according to the intent concept and the semantics of the keywords.

For example, if the voice information is "How to do the problem of continuous division", the keyword extracted from it is "How to do the problem of continuous division", and the question word is "How to do". According to the concept map of the question method, the matching word "How to do" is obtained. The corresponding questioning concept of the node is "explain", and the semantics of the keyword "connection and division question" is "connection and division question". According to the questioning concept "explanation" and the keyword "connection and division question", the user's intention can be obtained. For "explain the problem of concatenation and division".

In this embodiment, the semantics of the keywords are acquired by establishing a thesaurus and using the keyword matching method, which is more convenient and quicker than other semantic recognition methods.

According to a seventh embodiment provided by the present invention, as shown in FIG. 7 , a device for recognizing intent based on a concept map includes:

The collection module 100 is used to collect the voice information of the user;

Extraction module 200, for performing grammatical analysis on the voice information, and extracting question words and keywords in the voice information;

The matching module 300 is configured to match the question word with the node in the question concept graph to obtain the matched node, wherein the question concept graph is pre-built;

The intent acquisition module 400 is configured to acquire the user's intent according to the intent concept corresponding to the matched node and the keyword.

Preferably, it also includes:

a signal receiving module 600, configured to receive a touch signal of the user when the keyword contains a specific keyword;

Specifically, the specific keywords refer to pronouns such as "this word", "this question", "this sentence", "this", "this" and the like. When the above-mentioned specific keyword is included in the voice information, the user's touch signal on the touch screen is received. The touch signal may be a continuous sliding touch signal by the user on the touch screen, such as a straight line formed by continuous touch, or a circular frame, oval frame, irregular graphic frame, etc. formed by continuous touch.

An image acquisition module 700, configured to acquire image information of the user-selected area according to the touch signal;

Specifically, after receiving the touch signal, the image information of the area selected by the user can be obtained according to the touch signal. For example, after continuous touches form a straight line, a rectangular frame is constructed with the straight line as the diagonal, and the image in the rectangular frame is intercepted. , the image in the rectangular frame is the image information of the area selected by the user.

If the continuous touch forms a circular frame or an oval frame, etc., the image in the formed frame is intercepted, and the image in the frame is the image information of the area selected by the user.

Here, the touch signal is limited to a continuous touch signal, which can prevent erroneous operations and reduce the error rate. For example, it can prevent the user from forming two touch points on the touch screen due to erroneous operations, and obtain the frame formed by the two points. image inside.

When the voice information contains specific keywords, it means that the voice information input by the user is missing, such as "how to solve this problem", the tutor does not know what the problem is, so even if the voice information is semantically parsed Afterwards, the exact intention of the user cannot be obtained. Therefore, when it is determined that the voice information contains a specific keyword, it is necessary to further obtain the image information of the area pointed by the user's finger or framed to obtain the information of a certain question.

The intent acquisition module 400 is further configured to acquire the user's intent according to the intent concept corresponding to the matched node, the keyword, and the image information.

Specifically, according to the acquired intent concept and the semantics of the keywords, combined with the acquired image information, the three are integrated to obtain the user's true intent, so as to give corresponding feedback.

Preferably, the intent acquisition module 400 includes:

an identification unit 440, configured to identify text information in the image information;

A replacement unit 450, configured to replace specific keywords in the keywords with the text information;

The intent acquiring unit 430 is configured to acquire the user's intent according to the intent concept corresponding to the matched node, the keyword and the text information.

Specifically, after the image information is obtained, the image is processed and identified to obtain text information in the image information, where the text information is information about a certain question or information about a certain word.

Replace the acquired text information with specific keywords, that is, replace this question with the acquired text information. Then, according to the intent concept corresponding to the matched node, the semantics of the remaining keywords, and the acquired text information, the real intent of the user is obtained.

For example, when a user reads extracurricular readings on the home teaching machine, when he sees "conscientious and conscientious", he does not recognize "conscientious and conscientious" and wants to ask for help, so the user will say "what do these words read", the home teaching machine collects the voice information, When it is judged that the voice information contains a specific keyword, the image information of the area selected by the user will be obtained according to the touch signal. Combined with the question word "what to read", the real semantics of the user can be obtained as "what to read conscientiously". In this way, words that the user does not know or questions that cannot be answered can be helped through the tutoring machine to assist the user's learning.

It should be noted that the above embodiments can be freely combined as required. The above are only the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the principles of the present invention, several improvements and modifications can be made. It should be regarded as the protection scope of the present invention.

Claims (10)

1. An intention recognition method based on a concept graph is characterized by comprising the following steps:

Collecting voice information of a user;

carrying out grammar analysis on the voice information, and extracting question words and key words in the voice information;

matching the question words with nodes in a question concept graph to obtain matched nodes, wherein the question concept graph is constructed in advance;

and acquiring the intention of the user according to the intention concept corresponding to the matched node and the keyword.

2. The method for identifying an intention based on a concept graph according to claim 1, wherein the method for constructing the concept graph of the question method comprises the following steps:

collecting a large amount of user corpora;

extracting the question words in the corpus;

acquiring the intention concepts of the question words with the same intention in the corpus;

establishing a mapping relation between the question words and the intention concepts;

and constructing a question concept graph according to the mapping relation between the question words and the intention concepts.

3. The method as claimed in claim 1, wherein the obtaining the user's intention according to the intention concept corresponding to the matched node and the keyword specifically comprises:

searching a target word matched with the keyword from a preset word bank;

Determining the semantics of the keywords according to the semantics of the target words;

and acquiring the intention of the user according to the intention concept corresponding to the matched node and the semantics of the keyword.

4. The method for recognizing the intention based on the concept graph according to any one of claims 1-3, wherein after parsing the voice information and extracting the question words and the keywords in the voice information, the method further comprises:

when the keywords contain specific keywords, receiving a touch signal of the user;

acquiring image information of the user selection area according to the touch signal;

the obtaining the intention of the user according to the intention concept corresponding to the matched node and the keyword specifically includes:

and acquiring the intention of the user according to the intention concept corresponding to the matched node, the keyword and the image information.

5. The method as claimed in claim 4, wherein the obtaining the user's intention according to the intention concept corresponding to the matched node, the keyword and the image information specifically comprises:

Identifying character information in the image information;

replacing the text information with a specific keyword in the keywords;

and acquiring the intention of the user according to the intention concept corresponding to the matched node, the keyword and the text information.

6. An intention recognition apparatus based on a concept graph, comprising:

the acquisition module is used for acquiring voice information of a user;

the extraction module is used for carrying out grammar analysis on the voice information and extracting the question words and the key words in the voice information;

the matching module is used for matching the question words with nodes in a question concept graph to obtain matched nodes, wherein the question concept graph is constructed in advance;

and the intention acquisition module is used for acquiring the intention of the user according to the intention concept corresponding to the matched node and the keyword.

7. The concept graph-based intention recognition apparatus according to claim 6, further comprising a construction module;

the building module comprises:

the corpus collection unit is used for collecting a large amount of user corpuses;

the word extraction unit is used for extracting the question words in the corpus;

The concept acquisition unit is used for acquiring the intention concepts of the same-intention question words in the corpus;

the relationship establishing unit is used for establishing a mapping relationship between the question and legal terms and the intention concepts;

and the construction unit is used for constructing a question and law concept graph according to the mapping relation between the question and law words and the intention concepts.

8. The concept graph-based intention recognition device according to claim 6, wherein the intention acquisition module comprises:

the word searching unit is used for searching a target word matched with the keyword from a preset word bank;

the semantic determining unit is used for determining the semantics of the keywords according to the semantics of the target words;

and the intention acquisition unit is used for acquiring the intention of the user according to the intention concept corresponding to the matched node and the semantics of the keyword.

9. The concept graph-based intention recognition apparatus according to any one of claims 6 to 8, further comprising:

the signal receiving module is used for receiving a touch signal of the user when the keyword contains a specific keyword;

the image acquisition module is used for acquiring the image information of the user selection area according to the touch signal;

The intention acquisition module is further used for acquiring the intention of the user according to the intention concept corresponding to the matched node, the keyword and the image information.

10. The concept graph-based intention recognition device according to claim 9, wherein the intention acquisition module comprises:

the identification unit is used for identifying character information in the image information;

the replacing unit is used for replacing the text information with a specific keyword in the keywords;

and the intention acquisition unit is used for acquiring the intention of the user according to the intention concept corresponding to the matched node, the keyword and the character information.

Images