CN111966840B

CN111966840B - Man-machine interaction management method and management system for language teaching

Info

Publication number: CN111966840B
Application number: CN202010830397.2A
Authority: CN
Inventors: 刘金龙; 王亮; 赵薇; 何苏; 柳景明; 郭常圳
Original assignee: Beijing Ape Power Future Technology Co Ltd
Current assignee: Beijing Ape Power Future Technology Co Ltd
Priority date: 2020-08-18
Filing date: 2020-08-18
Publication date: 2024-07-02
Anticipated expiration: 2040-08-18
Also published as: CN111966840A

Abstract

The invention provides a man-machine interaction management method and a man-machine interaction management system for language teaching. The method comprises the following steps: outputting prompt material content; acquiring voice input of a user and converting the voice input into a user text; acquiring a preset answer with a corresponding relation with the prompt material from a first database, and matching the user text with the preset answer; according to a semantic similarity algorithm, searching whether a second database contains a first text meeting the condition of the semantic similarity algorithm with the user text, and if so, acquiring content with a corresponding relation with the first text; and outputting the content with the corresponding relation with the first text according to a preset first rule or outputting the prompt material content according to the matching degree of the user text and the preset answer. The invention improves the experience and interest of the user learning.

Description

Man-machine interaction management method and management system for language teaching

Technical Field

The invention belongs to the technical field of artificial intelligence, and particularly relates to a man-machine dialogue management method and system for language teaching.

Background

In the existing online learning software, the method for managing the dialogue with the user is mainly divided into manual-based and neural network-based generation technologies. Based on manual dialogue management, for example, a teacher performs one-to-one dialogue with students, and the teacher performs questions or answers with the students according to learning and answering conditions of the students. Based on the dialogue management generated by the neural network, replies to children are generated by the neural network according to the response history of the students.

Based on manual dialogue management, a great deal of labor cost is required, and the teaching method and knowledge reserve of each teacher are different, so that the knowledge grasping degree of students in the dialogue cannot be quantified. Dialog management generated based on a neural network often generates meaningless replies, such as a system reply "I donot know" in the english dialog process, objectively forces the dialog to end, which can lead to reduced interestingness of the dialog and too few dialog rounds; in addition, the generated replies may contain words and phrases that are difficult to understand, so that children cannot understand the meaning of the replies. The occurrence of the situation reduces the interest of students in continuing to learn, and especially in learning languages such as Chinese, english and the like, the promotion of the oral expression of the students is very important for learning language contents.

Disclosure of Invention

The application provides a man-machine interaction management method and a man-machine interaction management system for language teaching. The method comprises the following steps: outputting prompt material content; acquiring voice input of a user and converting the voice input into a user text; acquiring a preset answer with a corresponding relation with the prompt material from a first database, and matching the user text with the preset answer; according to a semantic similarity algorithm, searching whether a second database contains a first text meeting the condition of the semantic similarity algorithm with the user text, and if so, acquiring content with a corresponding relation with the first text; and outputting the content with the corresponding relation with the first text according to a preset first rule or outputting the prompt material content according to the matching degree of the user text and the preset answer.

In the above method, outputting the content having the corresponding relation with the first text according to the preset first rule or outputting the prompt material according to the matching degree of the user text and the preset answer specifically includes: if the user text and the preset answer meet the minimum matching degree requirement, outputting new prompt materials or/and output scores according to a preset second rule; otherwise, searching the first text according to the semantic similarity algorithm, and outputting the content with the corresponding relation with the first text.

In the above method, if the user text and the preset answer meet the minimum matching degree requirement, outputting new prompt materials or/and outputting scores according to a preset second rule includes: if the user text contains the least number of keywords in the preset answers, and the sentence pattern of the preprocessed user text is inconsistent with the sentence pattern of the preset answers; and outputting sentence pattern prompt materials and/or outputting scores.

Or if the user text and the preset answer meet the minimum matching degree requirement, outputting new prompt materials or/and outputting scores according to a preset second rule comprises: if the user text does not contain the minimum number of keywords in the preset answer, and the sentence pattern of the preprocessed user text is consistent with the sentence pattern of the preset answer; then, outputting word prompt materials or/and outputting scores.

In the above method, the minimum number of keywords are all keywords in the preset answer.

Based on the method, the method further comprises the following steps: if the voice input of the user is not acquired within the first time after the prompt material is output, the prompt material is output again or a new prompt material is output.

Further, the method comprises the following steps: and if the preset second time length is reached after the prompt material is output for the first time, stopping outputting the prompt material.

The prompt materials specifically comprise: voice or animation or pictures or a combination thereof;

And the content of the output corresponding to the first text is specifically output in the form of voice or animation or pictures or any combination thereof.

In the above method, according to the semantic similarity algorithm, searching whether the second database contains the first text meeting the similarity algorithm condition with the user text specifically includes: and acquiring a first text which is semantically similar to the text of the user by using a text matching algorithm of the long and short term memory LSTM and the Attention mechanism Attention.

The application also provides a man-machine conversation management system, which comprises:

A processor; and

A memory having executable code stored thereon which, when executed by the processor, causes the processor to perform the method as described above.

The present application also provides a non-transitory machine-readable storage medium having stored thereon executable code which, when executed by a processor of an electronic device, causes the processor to perform the method as described above.

The technical scheme provided by the application can comprise the following beneficial effects: in the process of language learning, if the user voice input content is not the answer of the preset question, searching the sentence which is the closest to the user voice input content by using a semantic similarity algorithm, so that the content replying to the sentence can be obtained. When the content is a sentence, the voice of the sentence is output to form a man-machine dialogue with the voice input of the user, and even when the input of the user is not the correct answer of the question, the method can promote the language interaction with the user to continue without interruption, and the user is guided to return to the learning content and the question on the basis of continuing the interaction with the user, so that the learning experience and interest of the user are improved.

In the method, whether the user input answer is correct is judged, on one hand, whether the user input answer contains all keywords in the preset answer is judged, and on the other hand, whether the user voice input sentence is in a correct sentence pattern is judged, so that the method is more suitable for being applied to language teaching.

In the method, the text matching algorithm of the long-term memory LSTM and the Attention mechanism Attention is utilized to acquire the content similar to the text semantics of the user, the accuracy of the algorithm avoids the situation that the content output by the program is irrelevant to the voice input content of the user, and therefore the human-computer interaction experience is prevented from being reduced; on the other hand, the algorithm can find out that the content output of the dialogue with the user can be maintained, and the user experience is further improved.

In the method, the preset question forms of program output can be various, such as pictures, voice, short video or a combination thereof; similarly, the program output for maintaining interaction with the user may be in the form of speech, a combination of speech and images, or an animation. The experience of interaction with the user is improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application as claimed.

Drawings

The foregoing and other objects, features and advantages of the application will be apparent from the following more particular descriptions of exemplary embodiments of the application as illustrated in the accompanying drawings wherein like reference numbers generally represent like parts throughout the exemplary embodiments of the application.

FIG. 1 is a schematic flow chart of a man-machine interaction management method for language teaching according to an embodiment of the application;

FIG. 2 is an example of a problem database in an embodiment of the application.

Detailed Description

Preferred embodiments of the present application will be described in more detail below with reference to the accompanying drawings. While the preferred embodiments of the present application are shown in the drawings, it should be understood that the present application may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the application to those skilled in the art.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or all possible combinations of one or more of the associated listed items.

It should be understood that although the terms "first," "second," "third," etc. may be used herein to describe various information, these information should not be limited by these terms. These terms are only used to distinguish one type of information from another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the application. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present application, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

The main form of online learning is that users learn by using intelligent terminals, such as smart phones, computers and other devices. One way of online learning is that students and teachers interact online through intelligent devices connected with the Internet at different places respectively; another way of online learning is for online learning software to output planned learning content for the user to view learning. For example, the software is used for helping students to memorize by outputting English words on an interface or simultaneously matching and outputting pictures consistent with the meaning of the English words; and, by outputting a question using the intelligent terminal, waiting for a student to answer to check the learning effect of the student, so-called on-line test.

In language learning, a mode mainly based on voice interaction is often adopted to help students complete online learning. For example, the intelligent terminal outputs a picture of sheep and simultaneously outputs a voice question "What is it? ". Further, in the prior art, the software program generally waits for the voice input of the student, then determines whether the content of the voice input of the student is correct, and outputs the determination result as the end.

The problem with the existing practice is that the learning interest of the students is reduced.

On the one hand, compared with the real teaching, the teaching mode of the learning content of the software output plan reduces the interactive process of teachers and students, and the students cannot be guided to complete the learning task according to the current learning situation of the students, especially the current performance like a teacher.

On the other hand, in language learning (e.g., english learning) of low age, a child may speak another sentence, such as "Whoareyou? ". According to the existing practice, the software program judges that sentences of children are not correct answers and gives a conclusion that the answers are incorrect, so that the interest of the children and the software program in continuing to speak English is hit, and even the learning of English is bored.

The invention provides a method for managing man-machine interaction, which is particularly used for a scene of learning by a user through an intelligent terminal.

The following describes in detail a technical solution of a preferred embodiment of the present application, taking an english learning scenario as an example. The application may be used in other language learning scenarios.

In the embodiment, the AI technology is adopted to guide the user to conduct multi-round dialogue with the dialogue robot, and the reply of the dialogue robot is dynamically selected according to the learning condition and the reply history of the user while the interestingness is ensured, so that the aim of continuously guiding the user to learn English is fulfilled.

In order to enable children to communicate better in English, the embodiment of the application designs a one-to-one conversation mode, wherein the conversation mechanism is as follows.

First, a question is asked and a user's voice input is awaited to be received.

And the software in the intelligent terminal is used for asking questions. The questioning can be performed in a voice output mode, a voice combined image mode or an image mode.

For example, a sheep is displayed on the interactive page and a voice question "What is it? ".

After the problem is posed, waiting for the children to speak for answering, and after the voice input of the user is obtained, outputting corresponding materials according to the method of the invention, so as to complete knowledge learning and improve the learning experience of the children.

Secondly, the system timer judges whether the maximum duration of the dialogue exercise is reached

Presetting learning duration, for example, 20 minutes, starting timing from the first question posed by the conversation robot;

if the preset duration is not reached, the dialogue robot outputs materials such as questions and the like according to the answering situation of the children and the history information, and continuously guides the children to learn;

if the preset duration is reached, the interaction with the children is ended, and the learning is ended; or the child actively closes the conversation.

In the whole interactive process of English teaching, the pre-set problem database and the user dialogue material library are utilized to ask questions and replies to children, and the specific method is as follows, referring to FIG. 1.

11 Outputting the prompt material content.

The problem database consists of questions and answers corresponding to the questions, wherein the answers can be split into sentence patterns and keywords. Therefore, when children learn English, the children can grasp the corresponding sentence patterns and keywords.

The problem database stores the corresponding relation of questions, answers, sentence patterns and keywords. For example, referring to fig. 2, materials, questions, answers, sentences, words, notes, etc. are included in the data format. Referring to the first line shown in fig. 2, the material is "sheep" and the problem is "What is it? "answer is" It's a quench ", sentence" It's a ", word (i.e. keyword) is" quench ", and" It is=it's "in remarks.

The questions are selected from the problem database to ask the user, and the questions are in various forms, such as voice, pictures and the like. For example, a sheep is displayed on the interactive page and a voice question "What is it? ".

12A voice input of the user is acquired and the acquired voice input content is converted into a file in text form.

After outputting the content of the question material, waiting for the user to answer the question by voice. As a preferred implementation manner, if the voice input of the user is not obtained within the preset first duration, the content of the prompt material can be output again, or new prompt material content can be output according to the related business rule.

And obtaining voice input of a user, performing voice recognition and converting the voice input into a text form.

13 Obtaining a preset answer with a corresponding relation with the prompt material from a problem database, and matching the user text with the preset answer.

131 According to the output questions, searching answers corresponding to the questions in the problem database. See above, the output question is "What is it? ", the corresponding preset answer is expected to be" it's a streep ".

132 Pre-processing the user answer converted into text form, the pre-processing mainly comprising the following aspects:

a) Meaningless punctuation marks are removed.

B) Case-to-case conversion is performed.

C) The abbreviations for the words are expanded, such as converting It's to It is.

D) The single complex number of the word is converted, such as apples and apples.

The text mismatch caused by different language expression forms can be greatly reduced by preprocessing. The present invention is not limited to the method of pretreatment nor does it require that all of the pretreatment steps described above be performed in their entirety. For example, the word singular/plural is not converted.

133 And matching the preprocessed user answers with answers corresponding to the questions.

In this embodiment, for example, the english teaching is taken as an example, and the correct answer of the english sentence generally includes two parts, namely, a keyword and a sentence pattern.

The keywords are words that require the user to learn mastery, such as the word "strep" in the example above, i.e., the words that children know "sheep" in english.

The sentence pattern is also known as grammar of English. The child has knowledge of the grammar and can speak a complete sentence instead of just answering the word.

In order to achieve the purpose of matching the preprocessed user answers with answers corresponding to the questions, the embodiment adopts a fine-grained text matching method to match keywords and sentence patterns in preset answers respectively.

And for keyword matching, judging whether the answers of the children contain all keywords by adopting a longest public subsequence algorithm. If so, the content of the voice input of the child is considered to be matched with the keyword, otherwise, the content is not considered to be matched.

For sentence pattern matching, judging whether the answers of the children contain continuous and complete sentence patterns, if so, considering that the answers of the children are matched with the sentence patterns, otherwise, not matching.

And for matching of the answers, judging that the answers of the children are matched with the keywords and the sentence patterns, if the answers are matched with the keywords and the sentence patterns, considering that the answers of the children are matched with the answers, otherwise, not matching.

By way of example as shown in fig. 2, material output: a sheep is displayed on the interactive page and a voice question "What is it? ". The answer corresponding to the problem database is "It's a quench", the sentence is "It's a", the word (i.e. keyword) is "quench", and "It is=it's" in the remark.

After converting the voice input of the child into a text form, judging whether a keyword 'quench' is contained or not by using a longest public subsequence algorithm, and if so, matching words; and, it is judged whether the sentence pattern "It's a" is contained, and if so, the sentence pattern is matched.

14 Searching whether a first text meeting the similarity algorithm condition with the user text is contained in a user dialogue material library according to a semantic similarity algorithm, and if so, acquiring content with a corresponding relation with the first text;

In a preferred embodiment, sentences which can form a dialogue with the user voice input are preset in the user dialogue material library.

For example, "Whoareyou" is the user dialogue material library saved? "and" Iam zack, "and that the two words have correspondence, i.e.," Whoareyou "if user speech input is recognized? "at time, the system outputs" Iam zack.

The dialogue material saved by the user dialogue material library is shown in the following table. For example, if the user voice input is identified as "I like you," the system outputs "I like you, to.

Problem(s)	System output
		What's your hobby？	I like singing.
What do you like to eat？	I like to eat apples.
		What are you doing？	I'm talking with you.
Nice to meet you.	Nice to meet you,too！
		I love you.	I love you,too.
I like you.	I like you,too.
		How old are you？	I'm six years old.
How are you？	I'm happy.
		Good night	Good night！
Good morning！	Good morning！
		Good evening！	Good evening！
Do you like dogs	Yes,I like dogs.
		Do you like cats	Yes,I like cats.
Can you swim？	No,I can't.
		Are you sad？	No,I'm happy.
Are you happy？	Yes,I'm happy.
		Are you angry？	No,I'm happy.
Are you angry？	No,I'm happy！

The invention does not limit the preset sentences in the user speech material library, and the skilled person can set and continuously add dialogue corpus aiming at specific business scenes.

The invention does not limit the output content form in the user dialogue material library. As a preferred implementation, the user dialogue material library stores dialogue sentences for completing man-machine interaction, for example, "Whoareyou? "and" Iam zack "voice input forms. However, the material library can also comprise materials such as pictures or animations, so that the form of interaction with the user is enriched, and the user experience is enhanced. For example, when outputting voice and user dialogue, the interactive page outputs corresponding animation or picture or sentence characters to assist the user in understanding the meaning of the voice output by the system.

The invention does not limit the data format and the searching mode of the user dialogue material library, and only needs to realize the formation of corresponding material output according to the user statement input.

The following embodiment of the invention provides a better semantic similarity algorithm, which is used for searching whether sentences similar to the input semantics of a user exist in a user dialogue material library. The better semantic similarity algorithm can avoid that the materials output by the system are irrelevant to the voice input content of the user, and avoid reducing the human-computer interaction experience; on the other hand, materials which can maintain the dialogue with the user can be obtained through a semantic similarity algorithm, so that the dialogue is prevented from being forcedly interrupted.

The semantic similarity algorithm is used for judging whether two sentences have the same semantic meaning.

Firstly, screening out a plurality of preset questions most similar to the answers of children by adopting a Lucene search engine,

Then, a text matching algorithm based on Long Short Term Memory (LSTM) and Attention mechanism (Attention) is used to screen out the people's questions most similar to the responses of children.

Long Short-Term Memory (LSTM) is a type of time-Recurrent Neural Network (RNN) suitable for processing and predicting very Long-spaced and delayed events of importance in a time series. Attention mechanism (Attention) is to calculate the matching degree of the current input sequence and the output vector, and the matching degree is high, namely the relative score of the Attention concentration point is higher, wherein the matching degree weight calculated by the Attention is limited to the current sequence pair.

The text matching algorithm based on LSTM and Attention is to consider text matching as a classification problem, i.e. whether two texts express the same semantic. Two texts are defined and represented by question and answer, respectively. The whole text matching algorithm is shown in fig. 2 and consists of an input layer, an attention layer and an output layer.

And an input layer for encoding and representing the text information. The following two features are selected for one text.

(1) Global vector of word representations (Global Vectors for Word Representation).

(2) Whether a word in one text appears in co-occurrence features in another text.

These two features encode two texts, which are input to the Attention (Attention) layer below.

And the attention layer is responsible for carrying out interaction of two text word levels, and can fuse more context information and learn semantic information among different texts through new features obtained by continuous interaction. Specific:

first, the word vector feature of question is operated on seq_ attention with the word vector feature of answer to obtain a new representation of answer.

The seq_ attention operation multiplies the word vector feature matrix of question and the answer, calculates the alignment feature between their vectors, and then re-represents the feature of the answer with the alignment feature and the word vector feature matrix of question to obtain a new feature of the answer.

Then, the word vector feature of answer, the manual feature (such as co-occurrence feature, etc.) and the new feature are spliced together, and input to one bidirectional LSTM to obtain a new feature ha, and simultaneously the word vector feature of question and the manual feature are input to the other bidirectional LSTM to obtain a new feature hq. Through learning features through the attention layer, the interaction features of question and answer can be learned, and rich semantic and contextual features in the text can be captured by adding the LSTM.

The basic idea of the bi-directional LSTM is that each training sequence is forwarded and backward respectively two Recurrent Neural Networks (RNNs), and both are connected to one output layer. This structure provides the output layer with complete past and future context information for each point in the input sequence.

And the output layer is used for carrying out further interaction and outputting a result to predict the semantic similarity of the two sentences.

For the features hq and ha, using self-attention to obtain new features hq1 and ha1;

Wherein self-attention models the degree of association of different words in the sentence, and then represents the current word with other words, so that the context information can be better captured. The feature [ ha1, hq1, ha 1] is then constructed using hq1 and ha1, and this feature is passed through a full join layer to predict whether the two texts are similar.

15 Outputting content with corresponding relation with the first text according to a preset first rule or outputting prompt material content according to the matching degree of the user text and a preset answer.

After the user voice input is converted into the user text, the user text is matched with the preset answers in the problem database, and whether the user dialogue material library contains the text with the same semantic as the user text is searched according to the semantic similarity algorithm. And further, executing subsequent processing according to the matching result of the problem database and the searching result of the user dialogue material library. The method specifically comprises the following steps:

If the user text is matched with the keywords and sentence patterns in the preset answer, a new question is selected from the problem library, and a prompt material of the question is output, for example, refer to fig. 2: a picture of a puppy Is input and a voice "Is it a dog? ";

if the user text is matched with the sentence pattern in the preset answer but is not matched with the keyword in the preset answer, outputting the keyword content, such as the text of "streep" or/and the voice of "streep"; further, as a better implementation manner, if the user does the question for the first time, after outputting the keyword content (text of "quench" or/and voice of "quench"), the prompt material of the question is repeatedly output once again, otherwise, a material of a new question is selected to be output.

If the user text is matched with the keyword in the preset answer but is not matched with the sentence pattern in the preset answer, outputting prompt contents of the sentence pattern, for example, the text of ' It's a ' or/and the voice of ' It's a ' or the text of ' It's a streep '. Further, as a better implementation manner, if the user does the question for the first time, after outputting the sentence-based content, the prompt material of the question is repeatedly output once again, otherwise, the material of a new question is selected to be output.

If the user input is not matched with the keywords and the sentence patterns in the preset answer, further judging whether sentences with the same meaning as the user text are found in the user dialogue material library, and if so, outputting the content corresponding to the sentences in the user dialogue material library.

For example, "Whoareyou" in a user dialogue material library, through a semantic similarity algorithm, is found with a sentence input by a user's voice? "have the same semantics, and thus, output corresponding material, including, for example, speech or/and" Iam zack ".

If the user input does not match the keywords and the sentence patterns in the preset answer, and the user dialogue material library does not have sentences with the same meaning as the user text. The content of the preset answer, for example, a voice and/or text of "It's a quench," is output. Then, as a preferred embodiment, if the user is judged to be the first question, the prompt material of the question is repeatedly output once again, otherwise, a new question is selected to be output.

The foregoing describes a specific implementation of the method of the present invention. In the above embodiment, according to the user text obtained after the user voice is input, the preset answer matching of the problem database is performed, the user dialogue material database is searched by using the semantic similarity algorithm, and according to the matching result of the problem database and the searching result of the user dialogue material database, the corresponding processing is performed when different conditions are satisfied according to the preset rules.

In another implementation manner, first, matching a preset answer according to the user text, and if the user text cannot be matched with the preset answer, that is, neither the keyword nor the sentence pattern of the preset answer is matched, searching whether a sentence meeting the algorithm condition exists in a user dialogue database by using a semantic similarity algorithm.

In the above embodiment, sentences with the same meaning as the text of the user in the user dialogue material library are obtained by using a semantic similarity algorithm. In other implementations, the application does not exclude sentences obtained using semantic similarity algorithms that have similar semantics to the user text. Thus, a person skilled in the art can set the requirements of the algorithm according to the requirements of the business, and the obtained sentences fulfil the conditions of the semantic similarity algorithm, which may have substantially the same or similar semantics as the user text.

In the above-described embodiment, the preset answer has a keyword and a sentence pattern, for example: when the answer is "It's a quench", the sentence is "It's a", and the keyword is "quench". And when the text of the user is not matched with the keywords and the sentence patterns, outputting materials which maintain the dialogue interactive relation with the user through a semantic similarity algorithm. When the preset answer includes a plurality of keywords, a person skilled in the art needs to set a requirement of minimum matching degree, so that when the user text does not include all the keywords, what is the case reaches a conclusion that the user text matches or does not match the keywords. The preferred implementation is to require the user text to match all keywords before reaching the conclusion that the user text matches the keywords. In other implementations, the person skilled in the art achieves the purpose of executing the semantic similarity algorithm under what conditions by setting the lowest keyword matching number or the rule of the lowest matching degree of other keywords.

The above embodiments only use english language teaching as an example for the implementation of the present invention, but the present invention is not limited to the scenario of application to other language teaching. Under the condition that grammar and words are used as language teaching, a semantic similarity algorithm can be utilized to maintain a dialogue with a user, and the effect of the language teaching is improved based on the continuation of the dialogue.

The method according to the present disclosure may also be implemented as a man-machine dialogue management system for language teaching, including: a processor; and a memory having executable code stored thereon that, when executed by the processor, causes the processor to perform part or all of the steps of the above-described methods of the present disclosure.

The present disclosure may also be embodied as a non-transitory machine-readable storage medium (or computer-readable storage medium, or machine-readable storage medium) having stored thereon executable code (or computer program, or computer instruction code) that, when executed by a processor of an electronic device (or computing device, server, etc.), causes the processor to perform some or all of the steps of the above-described methods according to the present disclosure.

The aspects of the present application have been described in detail hereinabove with reference to the accompanying drawings. In the foregoing embodiments, the descriptions of the embodiments are focused on, and for those portions of one embodiment that are not described in detail, reference may be made to the related descriptions of other embodiments. Those skilled in the art will also appreciate that the acts and steps referred to in the specification are not necessarily required for the present application. In addition, it can be understood that the steps in the method of the embodiment of the present application may be sequentially adjusted, combined and pruned according to actual needs, and the modules in the device of the embodiment of the present application may be combined, divided and pruned according to actual needs.

The foregoing description of embodiments of the application has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the various embodiments described. The terminology used herein was chosen in order to best explain the principles of the embodiments, the practical application, or the improvement of technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A man-machine interaction management method for language teaching is characterized by comprising the following steps:

outputting prompt material content to ask questions;

acquiring voice input of a user and converting the voice input into a user text;

acquiring a preset answer with a corresponding relation with the prompt material from a first database, and matching the user text with the preset answer; wherein the first database is a problem database;

If so, outputting prompt material content according to the matching degree of the user text and the preset answer;

If the text is not matched with the user text, searching whether the second database contains a first text meeting the condition of the semantic similarity algorithm according to the semantic similarity algorithm, if so, acquiring the content with a corresponding relation with the first text, and outputting the content with the corresponding relation with the first text; the second database is a user dialogue material library;

Otherwise, outputting the content of the preset answer; the method comprises the following steps: judging that the user is doing the topic for the first time, repeatedly outputting the prompting material of the topic once again, otherwise, selecting a new topic for outputting.

2. The method of claim 1, wherein outputting the content corresponding to the first text according to the preset first rule or outputting the prompt material according to the matching degree of the user text and the preset answer specifically comprises:

If the user text and the preset answer meet the minimum matching degree requirement, outputting new prompt materials or/and output scores according to a preset second rule;

otherwise, searching the first text according to the semantic similarity algorithm, and outputting the content with the corresponding relation with the first text.

3. The method of claim 2, wherein outputting new prompt material or/and output score according to a preset second rule if the user text meets a minimum matching requirement with the preset answer comprises:

If the user text contains the least number of keywords in the preset answers, and the sentence pattern of the preprocessed user text is inconsistent with the sentence pattern of the preset answers;

And outputting sentence pattern prompt materials and/or outputting scores.

4. The method of claim 2, wherein outputting new prompt material or/and output score according to a preset second rule if the user text meets a minimum matching requirement with the preset answer comprises:

If the user text does not contain the minimum number of keywords in the preset answer, and the sentence pattern of the preprocessed user text is consistent with the sentence pattern of the preset answer;

Then, outputting word prompt materials or/and outputting scores.

5. The method of claim 3 or 4, wherein the minimum number of keywords are all keywords in a preset answer.

6. The method according to one of claims 1 to 4, further comprising:

if the voice input of the user is not acquired within the first time after the prompt material is output, the prompt material is output again or a new prompt material is output.

7. The method as recited in claim 6, further comprising:

and if the preset second time length is reached after the prompt material is output for the first time, stopping outputting the prompt material.

8. The method of claim 1, wherein the step of determining the position of the substrate comprises,

9. The method according to claim 1, wherein searching whether the second database contains the first text satisfying the similarity algorithm condition with the user text according to a semantic similarity algorithm is specifically:

And acquiring a first text which is semantically similar to the text of the user by using a text matching algorithm of the long and short term memory LSTM and the Attention mechanism Attention.

10. A human-computer interaction management system for language teaching, comprising:

A processor; and

A memory having executable code stored thereon, which when executed by the processor, causes the processor to perform the method of any of claims 1-9.

11. A computer readable storage medium having stored thereon executable code which when executed by a processor of a computing device causes the processor to perform the method of any of claims 1-9.