CN116340489B

CN116340489B - Japanese teaching interaction method and device based on big data

Info

Publication number: CN116340489B
Application number: CN202310310503.8A
Authority: CN
Inventors: 高红燕; 董青; 翟艳蕾
Original assignee: Qiqihar University
Current assignee: Qiqihar University
Priority date: 2023-03-27
Filing date: 2023-03-27
Publication date: 2023-08-22
Anticipated expiration: 2043-03-27
Also published as: CN116340489A

Abstract

The invention relates to the technical field of teaching interaction, and discloses a Japanese teaching interaction method based on big data, which comprises the following steps: splitting the pre-acquired historical Japanese teaching embedded point data into a plurality of period teaching embedded point data, and extracting an error-prone sound class set, an error-prone word class set and an error-prone grammar class set from the period teaching embedded point data; dividing the periodic teaching embedded data into a plurality of periodic student embedded data, generating student periodic error characteristics corresponding to the periodic student embedded data, clustering all the student periodic error characteristics into a student periodic error characteristic class set, taking Japanese teaching embedded data of a target student as embedded data to be matched, extracting periodic error characteristics to be matched corresponding to the embedded data to be matched, and teaching the target student according to the student periodic error characteristic class set and the periodic error characteristics to be matched. The invention further provides a Japanese teaching interaction device based on the big data. The invention can improve the flexibility of Japanese teaching interaction.

Description

Japanese teaching interaction method and device based on big data

Technical Field

The invention relates to the technical field of teaching interaction, in particular to a Japanese teaching interaction method and device based on big data.

Background

Japanese is used as a language for more than one hundred million people worldwide, is a language which is relatively easy to enter, and also becomes a language which is selected by more and more students in an examination, so that the development of the Japanese teaching industry is further promoted, japanese teaching comprises the ability of training students to listen, speak, read and write to Japanese, and in order to improve the efficiency of Japanese teaching, more and more teaching institutions select online interactive teaching.

The existing Japanese teaching interaction method is mainly based on fixed teaching material problems, preset teaching material problem contents are issued to clients of students, corresponding problem analysis is given to assist the students to learn, in an actual process, the teaching interaction method based on the fixed teaching material problems is relatively solidified, flexibility is lacking, customized interaction cannot be carried out according to learning conditions of each student, and poor flexibility in Japanese teaching interaction can be possibly caused.

Disclosure of Invention

The invention provides a Japanese teaching interaction method and device based on big data, and mainly aims to solve the problem of poor flexibility in Japanese teaching interaction.

In order to achieve the above purpose, the Japanese teaching interaction method based on big data provided by the invention comprises the following steps:

cleaning historical Japanese teaching embedded point data obtained in advance into standard teaching embedded point data, splitting the standard teaching embedded point data into a plurality of period teaching embedded point data according to a teaching period, selecting the period teaching embedded point data one by one as target period teaching embedded point data, and respectively extracting a period pronunciation practice data set, a period word practice data set and a period grammar practice data set from the target period teaching embedded point data;

sequentially performing phoneme recognition and tone recognition on the periodic pronunciation training data set to obtain an incorrect pronunciation data set, sequentially performing word shape recognition on the periodic word training data set to obtain an incorrect word data set, and extracting an incorrect grammar data set from the periodic grammar training data set;

clustering the mispronounced data set into Yi Cuofa sound sets, clustering the mispronounced word data set into error-prone word sets, and clustering the mispronounced grammar data set into error-prone grammar sets;

splitting the target periodic teaching point burying data into a plurality of periodic student point burying data according to student identities, selecting the periodic student point burying data one by one as target student point burying data, generating student periodic error features corresponding to the target student point burying data according to the Yi Cuofa sound class set, the error-prone word class set and the error-prone grammar class set, and clustering all the student periodic error features into student periodic error feature class sets, wherein the clustering all the student periodic error features into the student periodic error feature class sets comprises the following steps: dividing all the periodic error characteristics of the students into a plurality of periodic error characteristic groups, and randomly selecting periodic error center characteristics of the students for each periodic error characteristic group; calculating the feature similarity between each periodic error feature of the trainee and each periodic error center feature of the trainee by using the following feature similarity formula:

Wherein K is the similarity, q is the q-th dimension feature vector, A _q Refers to the q-th dimension characteristic vector in the periodic error characteristic of the learner, wherein A ₁ Refers to the learner pronunciation error feature in the learner cycle error feature, wherein A ₂ Refers to the learner word error feature in the learner cycle error feature, A ₃ Refers to the learner grammar error feature in the learner cycle error feature, B _q Refers to the q-th dimension characteristic vector in the periodic error center characteristic of the learner, wherein B ₁ Refers to the learner pronunciation error feature in the learner cycle error center feature, wherein B ₂ Refers to the error character of the word error of the learner in the error center character of the learner cycle, B ₃ The method is characterized in that the method comprises the step of determining the learner grammar error characteristics in the learner cycle error center characteristics; updating the periodic error characteristic groups of the students into standard error characteristic groups one by one according to the characteristic similarity; calculating the error center characteristics of the standard students of each error characteristic group of the standard students, and calculating the similarity of the center characteristics between the error center characteristics of the standard students and the corresponding periodic error center characteristics of the students one by one; iteratively updating each standard learner error feature group into corresponding learner cycle error feature classes according to all the central feature similarities, and collecting all the learner cycle error feature classes into learner cycles A class of phase error features;

and taking Japanese teaching embedded data of a target student as embedded data to be matched, extracting periodic error features to be matched corresponding to the embedded data to be matched according to the teaching period, selecting a student periodic error feature class corresponding to the periodic error features to be matched from the student periodic error feature class set as a target periodic error feature class, updating a teaching database of the target student according to the target periodic error feature class, and teaching the target student by utilizing the updated teaching database.

Optionally, the cleaning the pre-acquired historical japanese teaching embedded point data into standard teaching embedded point data includes:

performing type detection on the buried point data in the historical Japanese teaching buried point data one by one to obtain buried point type data;

screening out type messy code data and type noise data from the historical Japanese teaching buried point data according to the buried point type data to obtain primary Japanese teaching buried point data;

and carrying out answer association on the primary Japanese teaching embedded point data to obtain standard teaching embedded point data.

Optionally, the sequentially performing phoneme recognition and tone recognition on the periodic pronunciation training data set to obtain a mispronounced data set includes:

Selecting the periodic pronunciation training data in the periodic pronunciation training data set one by one as target pronunciation training data, and sequentially performing audio correction and audio filtering operation on the target pronunciation training data to obtain primary pronunciation training data;

sequentially carrying out voice framing and endpoint detection operation on the primary pronunciation training data to obtain secondary pronunciation training data;

respectively extracting a target phoneme feature set and a target tone feature set of the secondary pronunciation training data;

performing phoneme matching on the target phoneme feature set to obtain a target pronunciation phoneme, performing feature matching on the target tone feature set to obtain a target pronunciation tone, and collecting the target pronunciation phoneme and the target pronunciation tone into target pronunciation identification data;

judging whether the target pronunciation identification data is the same as the real pronunciation answer corresponding to the target pronunciation exercise data;

if yes, returning to the step of selecting the periodic pronunciation exercise data in the periodic pronunciation exercise data set one by one as target pronunciation exercise data;

if not, taking the pronunciation analysis corresponding to the true pronunciation answer as target pronunciation analysis, extracting pronunciation keywords from the target pronunciation analysis, taking the pronunciation keywords as error pronunciation data, and collecting all error pronunciation data into an error pronunciation data set.

Optionally, the sequentially performing audio deviation rectifying and audio filtering operations on the target pronunciation training data to obtain primary pronunciation training data, including:

performing column signal data conversion on the target pronunciation training data to obtain column signal pronunciation training data;

sampling the column signal pronunciation training data at a fixed frequency to obtain a column signal time sequence;

constructing an audio trend item of the target pronunciation exercise data according to the column signal time sequence and the column signal pronunciation exercise data;

performing trend correction on the train signal pronunciation training data according to the audio trend item to obtain correction pronunciation training data;

and performing audio filtering on the deviation rectifying pronunciation training data to obtain primary pronunciation training data.

Optionally, the sequentially performing voice framing and endpoint detection operations on the primary pronunciation training data to obtain secondary pronunciation training data includes:

splitting the primary pronunciation training data into frame pronunciation training data according to a preset framing step length and an overlapping step length;

performing voice windowing on the framing pronunciation exercise data to obtain windowed pronunciation exercise data;

calculating the inter-frame speech energy of the windowed speech exercise data from frame to frame using the inter-frame speech energy formula:

Wherein E refers to the inter-frame speech energy, N refers to the nth frame in the windowed voicing practice data, N is the total frame length of the windowed voicing practice data, S is a signal value function, S (N-1) refers to the signal value of the nth-1 frame in the windowed voicing practice data, S (N-2) refers to the signal value of the nth-2 frame in the windowed voicing practice data, sgn is a symbol function, and |·| is an absolute value symbol;

and performing an end point detection operation on the windowed pronunciation practice data by using the inter-frame voice energy to obtain secondary pronunciation practice data.

Optionally, the sequentially performing word shape recognition on the periodic word training dataset to obtain an error word dataset includes:

selecting the periodic word training data in the periodic word training data set one by one as target periodic word training data, and sequentially performing binarization and median filtering operation on the target periodic word data to obtain primary periodic word data;

performing character cutting operation on the primary periodic word data to obtain secondary periodic word data;

extracting features of the secondary periodic word data to obtain a target word shape feature set;

matching a target training word corresponding to the target periodic word training data according to the target morphological feature set;

Judging whether the real word answers corresponding to the target training word and the target periodic word training data are the same or not;

if yes, returning to the step of selecting the periodic word exercise data in the periodic word exercise data set one by one as target periodic word exercise data;

if not, taking the real word answer corresponding to the target periodic word exercise data as error word data, and collecting all the error word data into an error word data set.

Optionally, the extracting the error grammar data set from the period grammar exercise data set includes:

selecting the periodic grammar training data in the periodic grammar training data set one by one as target grammar training data, and judging whether the real grammar answer corresponding to the target grammar training data is the same as the target grammar training data or not;

if yes, returning to the step of selecting the periodic grammar exercise data in the periodic grammar exercise data set one by one as target grammar exercise data;

if not, using the grammar analysis corresponding to the real grammar answer as target grammar analysis, extracting grammar keywords from the target grammar analysis, using the grammar keywords as error grammar data, and collecting all the error grammar data into an error grammar data set.

Optionally, the clustering the mispronounced data set into a Yi Cuofa sound class set includes:

splitting the mispronounced data set into a plurality of mispronounced data sets, and randomly selecting mispronounced center data for each mispronounced data set;

calculating word vector distances between each piece of mispronounced data and each piece of mispronounced center data in the mispronounced data set, and updating the mispronounced data set into a standard mispronounced data set one by one according to the word vector distances;

calculating standard mispronounced center data of each standard mispronounced data set, and calculating a center word vector distance between the standard mispronounced center data and the corresponding mispronounced center data one by one;

and iteratively updating each standard mispronounced data set into corresponding mispronounced classes according to all the center word vector distances, and collecting all the mispronounced classes into a Yi Cuofa sound class set.

Optionally, the generating the learner cycle error feature corresponding to the target learner buried data according to the Yi Cuofa sound class set, the error-prone word class set and the error-prone grammar class set includes:

Splitting the target student burial point data into target student pronunciation data, target student word data and target word grammar data;

extracting learner error pronunciation data from the target learner pronunciation data, extracting learner error word data from the target learner word data, and extracting learner error grammar data from the target word grammar data;

replacing each piece of error pronunciation data in the error pronunciation data of the learner by using each piece of cluster center data in the Yi Cuofa sound class set to obtain standard error pronunciation data of the learner, and performing normalization operation on the standard error pronunciation data by using the Yi Cuofa sound class set to obtain error pronunciation characteristics of the learner;

replacing each error word data in the error word data of the learner by using each cluster center data in the error-prone word class set to obtain standard error word data, and carrying out normalization operation on the standard error word data by using the error-prone word class set to obtain error word characteristics of the learner;

replacing each error grammar data in the error grammar data of the learner by using each cluster center data in the error grammar class set to obtain standard learner error grammar data, and performing normalization operation on the standard learner error grammar data by using the error grammar class set to obtain learner grammar error characteristics;

And splicing the learner pronunciation error feature, the learner word error feature and the learner grammar error feature into a learner cycle error feature.

In order to solve the above problems, the present invention further provides a japanese teaching interaction device based on big data, the device comprising:

the data splitting module is used for cleaning the pre-acquired historical Japanese teaching embedded point data into standard teaching embedded point data, splitting the standard teaching embedded point data into a plurality of period teaching embedded point data according to a teaching period, selecting the period teaching embedded point data one by one as target period teaching embedded point data, and respectively extracting a period pronunciation exercise data set, a period word exercise data set and a period grammar exercise data set from the target period teaching embedded point data;

the error recognition module is used for sequentially carrying out phoneme recognition and tone recognition on the periodic pronunciation training data set to obtain an error pronunciation data set, sequentially carrying out word shape recognition on the periodic word training data set to obtain an error word data set, and extracting an error grammar data set from the periodic grammar training data set;

the primary clustering module is used for clustering the mispronounced data set into Yi Cuofa sound sets, clustering the mispronounced word data set into error-prone word sets and clustering the error grammar data set into error-prone grammar sets;

The secondary clustering module is configured to split the target periodic teaching point burying data into a plurality of periodic student point burying data according to a student identity, select the periodic student point burying data one by one as target student point burying data, generate student periodic error features corresponding to the target student point burying data according to the Yi Cuofa sound class set, the error prone word class set and the error prone grammar class set, and cluster all the student periodic error features into a student periodic error feature class set, where the clustering of all the student periodic error features into a student periodic error feature class set includes: dividing all the periodic error characteristics of the students into a plurality of periodic error characteristic groups, and randomly selecting periodic error center characteristics of the students for each periodic error characteristic group; calculating the feature similarity between each periodic error feature of the trainee and each periodic error center feature of the trainee by using the following feature similarity formula:

wherein K is the similarity, q is the q-th dimension feature vector, A _q Refers to the q-th dimension characteristic vector in the periodic error characteristic of the learner, wherein A ₁ Refers to the learner pronunciation error feature in the learner cycle error feature, wherein A ₂ Refers to the learner word error feature in the learner cycle error feature, A ₃ Refers to the learner grammar error feature in the learner cycle error feature, B _q Refers to the q-th dimension characteristic vector in the periodic error center characteristic of the learner, wherein B ₁ Refers to the learner pronunciation error feature in the learner cycle error center feature, wherein B ₂ Refers to the error character of the word error of the learner in the error center character of the learner cycle, B ₃ The method is characterized in that the method comprises the step of determining the learner grammar error characteristics in the learner cycle error center characteristics; updating the periodic error characteristic groups of the students into standard error characteristic groups one by one according to the characteristic similarity; calculating the error center characteristics of the standard students of each error characteristic group of the standard students, and calculating the similarity of the center characteristics between the error center characteristics of the standard students and the corresponding periodic error center characteristics of the students one by one; iteratively updating each standard learner error feature group into corresponding learner cycle error feature classes according to all the central feature similarities, and collecting all the learner cycle error feature classes into a learner cycle error feature class set;

the teaching interaction module is used for taking Japanese teaching embedded data of a target student as embedded data to be matched, extracting periodic error features to be matched corresponding to the embedded data to be matched according to the teaching period, selecting the student periodic error feature class corresponding to the periodic error features to be matched from the student periodic error feature class set as a target periodic error feature class, updating a teaching database of the target student according to the target periodic error feature class, and teaching the target student by utilizing the updated teaching database.

According to the embodiment of the invention, the accuracy of historical data can be improved by cleaning the pre-acquired historical Japanese teaching buried point data into standard teaching buried point data, so that the operation accuracy of subsequent clustering and the like is improved, the periodic pronunciation exercise data set, the periodic word exercise data set and the periodic grammar exercise data set are respectively extracted from the target periodic teaching buried point data, the teaching buried point data can be analyzed according to pronunciation, words and grammar three-way, the comprehensiveness of Japanese teaching interaction is ensured, the periodic pronunciation exercise data set is sequentially subjected to phoneme recognition and tone recognition to obtain an error pronunciation data set, the periodic word exercise data set is sequentially subjected to word shape recognition to obtain an error word data set, the error grammar exercise data set is extracted from the periodic grammar exercise data set, keywords of error knowledge points corresponding to each exercise can be selected from the buried point exercise data set, the subsequent error clustering is facilitated, the error pronunciation data set is clustered into error-prone word sets according to Yi Cuofa sound class sets, the error word sets can be clustered into error-prone word sets through the error-prone word sets, the error grammar class sets can be analyzed, and the error word sets can be clustered into error-prone word sets through the error-prone word sets, and the error word sets can be conveniently analyzed after the error-prone word sets are clustered, and the error word sets are conveniently analyzed.

According to the Yi Cuofa sound class set, the error-prone word class set and the error-prone grammar class set, generating the learner cycle error characteristics corresponding to the target learner buried data, clustering all the learner cycle error characteristics into the learner cycle error characteristic class set, classifying and considering each learner according to the error-prone type of the error question in learning interaction, improving the applicability and the matching degree of Japanese teaching, selecting the learner cycle error characteristic class corresponding to the cycle error characteristics to be matched from the learner cycle error characteristic class set as the target cycle error characteristic class, updating a teaching database of the target learner according to the target cycle error characteristic class, teaching the target learner by utilizing the updated teaching database, and generating similar questions of the corresponding type according to the error records of each learner to carry out teaching consolidation, thereby improving the matching degree and the flexibility of Japanese teaching interaction. Therefore, the Japanese teaching interaction method and device based on big data can solve the problem of poor flexibility in Japanese teaching interaction.

Drawings

FIG. 1 is a schematic flow chart of a Japanese teaching interaction method based on big data according to an embodiment of the invention;

FIG. 2 is a flow chart of generating a mispronounced data set according to an embodiment of the present application;

FIG. 3 is a flow chart illustrating the generation of primary pronunciation training data according to an embodiment of the present application;

FIG. 4 is a functional block diagram of a Japanese teaching interaction device based on big data according to an embodiment of the present application;

the achievement of the objects, functional features and advantages of the present application will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.

The embodiment of the application provides a Japanese teaching interaction method based on big data. The execution subject of the Japanese teaching interaction method based on big data comprises at least one of a server, a terminal and the like which can be configured to execute the method provided by the embodiment of the application. In other words, the big data based japanese teaching interaction method may be performed by software or hardware installed in a terminal device or a server device, and the software may be a blockchain platform. The service end includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, and the like. The server may be an independent server, or may be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

Referring to fig. 1, a flow chart of a japanese teaching interaction method based on big data according to an embodiment of the present invention is shown. In this embodiment, the japanese teaching interaction method based on big data includes:

s1, cleaning historical Japanese teaching embedded point data acquired in advance into standard teaching embedded point data, splitting the standard teaching embedded point data into a plurality of period teaching embedded point data according to a teaching period, selecting the period teaching embedded point data one by one as target period teaching embedded point data, and respectively extracting a period pronunciation exercise data set, a period word exercise data set and a period grammar exercise data set from the target period teaching embedded point data.

In the embodiment of the invention, the historical japanese teaching embedded point data refers to embedded point data related to japanese teaching of each client stored in a japanese teaching server in a past period, for example, teaching periods where students corresponding to each client are located, pronunciation practice data of students, word practice data, grammar practice data and the like.

In the embodiment of the invention, the step of cleaning the pre-acquired historical Japanese teaching embedded point data into standard teaching embedded point data comprises the following steps:

In detail, the step of performing type detection on the buried point data in the historical japanese teaching buried point data one by one to obtain buried point type data refers to judging the data type of the buried point data one by one, for example, the buried point data in the japanese word exercise is handwritten word picture data, and the buried point data obtained in the japanese pronunciation exercise is audio data of a learner.

In detail, the type of messy code data refers to messy code data which does not accord with the type of the buried point data, for example, the option of the correct data type of the grammar buried point data is A, B, C, D, the options except A, B, C, D are messy code data, and the type of noise data refers to data which does not accord with the normal data state in the buried point data, for example, the part of the audio data obtained during pronunciation practice, wherein the audio duration is too short and the audio duration is too long.

Specifically, the answer association is performed on the primary japanese teaching embedded data to obtain standard teaching embedded data, which means that the training data of the learner in the primary japanese teaching embedded data is associated with a correct answer corresponding to the training data, for example, a picture 1 of word training of the learner corresponds to the correct japanese answer as a positive body.

In detail, the teaching period refers to a large period of Japanese teaching, such as a section of teaching, and the periodic teaching embedded point data refers to standard teaching embedded point data corresponding to one teaching period.

Specifically, the periodic pronunciation practice data set refers to data related to japanese pronunciation practice, such as pronunciation record data of a learner corresponding to each client, corresponding real japanese pronunciation answer, and the like, in the target periodic teaching buried point data, and the periodic grammar practice data set refers to data related to japanese word practice, such as word filling of a learner corresponding to each client, word implied data, corresponding real japanese word answer, and the like, in the target periodic teaching buried point data, such as grammar selection questions of a learner corresponding to each client, corresponding real japanese grammar answer, and the like.

In the embodiment of the invention, the accuracy of the historical data can be improved by cleaning the pre-acquired historical Japanese teaching buried data into the standard teaching buried data, so that the accuracy of the subsequent clustering operation and the like is improved, and the periodic pronunciation exercise data set, the periodic word exercise data set and the periodic grammar exercise data set are respectively extracted from the target periodic teaching buried data, so that the teaching buried data can be analyzed according to pronunciation, words and grammar, and the comprehensiveness of Japanese teaching interaction is ensured.

S2, sequentially carrying out phoneme recognition and tone recognition on the periodic pronunciation exercise data set to obtain an incorrect pronunciation data set, sequentially carrying out word shape recognition on the periodic word exercise data set to obtain an incorrect word data set, and extracting the incorrect grammar data set from the periodic grammar exercise data set.

In the embodiment of the invention, the wrong pronunciation data set refers to a part of pronunciation data set of the training pronunciation of the learner in the periodic pronunciation training data set, which is inconsistent with the corresponding real japanese pronunciation answer, the wrong word data set refers to a part of word data set of the learner in the periodic word training data set, which is inconsistent with the corresponding real japanese word answer, and the wrong grammar data set refers to a part of grammar data set of the learner in the periodic grammar training data set, which is inconsistent with the real japanese grammar answer, and the grammar gap filling of the learner.

In an embodiment of the present invention, referring to fig. 2, the sequentially performing phoneme recognition and tone recognition on the periodic pronunciation training data set to obtain an mispronounced data set includes:

s21, selecting the periodic pronunciation training data in the periodic pronunciation training data set one by one as target pronunciation training data, and sequentially performing audio correction and audio filtering operation on the target pronunciation training data to obtain primary pronunciation training data;

s22, sequentially carrying out voice framing and endpoint detection operation on the primary pronunciation exercise data to obtain secondary pronunciation exercise data;

s23, respectively extracting a target phoneme feature set and a target tone feature set of the secondary pronunciation training data;

s24, performing phoneme matching on the target phoneme feature set to obtain a target pronunciation phoneme, performing feature matching on the target tone feature set to obtain a target pronunciation tone, and collecting the target pronunciation phoneme and the target pronunciation tone into target pronunciation identification data;

s25, judging whether the target pronunciation identification data and the real pronunciation answer corresponding to the target pronunciation exercise data are the same or not;

s26, if yes, returning to the step of selecting the periodic pronunciation exercise data in the periodic pronunciation exercise data set one by one as target pronunciation exercise data;

And S27, if not, taking the pronunciation analysis corresponding to the true pronunciation answer as target pronunciation analysis, extracting pronunciation keywords from the target pronunciation analysis, taking the pronunciation keywords as error pronunciation data, and collecting all error pronunciation data into an error pronunciation data set.

In detail, by performing phoneme recognition and tone recognition on the periodic pronunciation exercise dataset to obtain an erroneous pronunciation dataset, whether the exercise pronunciation of a learner is standard or not can be judged by combining phonemes in terms of tone, and in japanese pronunciation exercise, due to the existence of special reading methods such as stroking, unvoiced sound, training reading and the like, the accuracy of simple speech recognition is low, and more accurate judgment results can be obtained by judging phonemes and tones.

In the embodiment of the present invention, referring to fig. 3, the steps of sequentially performing audio deviation rectification and audio filtering on the target pronunciation training data to obtain primary pronunciation training data include:

s31, converting the column signal data of the target pronunciation training data to obtain column signal pronunciation training data;

s32, sampling the column signal pronunciation training data at a fixed frequency to obtain a column signal time sequence;

S33, constructing an audio trend item of the target pronunciation exercise data according to the column signal time sequence and the column signal pronunciation exercise data;

s34, carrying out trend correction on the train signal pronunciation training data according to the audio trend item to obtain correction pronunciation training data;

and S35, performing audio filtering on the deviation rectifying pronunciation training data to obtain primary pronunciation training data.

In detail, the matlab algorithm library may be used to convert the column signal data of the target pronunciation training data to obtain column signal pronunciation training data, the least square method may be used to construct an audio trend item of the target pronunciation training data according to the column signal time sequence and the column signal pronunciation training data, and the filter designer may be used to perform audio filtering on the deviation correction pronunciation training data to obtain primary pronunciation training data.

In detail, the sequentially performing the voice framing and the endpoint detection operation on the primary pronunciation training data to obtain secondary pronunciation training data includes:

In the embodiment of the invention, the inter-frame voice energy of the windowed pronunciation training data is calculated by utilizing the inter-frame voice energy formula, and the unvoiced and voiced sounds and the endpoints of the voice signals in the corresponding windows can be judged by analyzing the signal value difference of the head and the tail of the window of each frame.

In detail, the target phoneme feature set and the target tone feature set of the secondary pronunciation training data may be extracted by using mel cepstrum coefficients, the target phoneme feature set may be subjected to phoneme matching by using a pre-trained phoneme neural network to obtain a target pronunciation phoneme, and the target tone feature set may be subjected to feature matching by using a pre-trained tone neural network to obtain a target pronunciation tone.

In detail, the pronunciation analysis refers to answer analysis corresponding to the true pronunciation answer, and includes pronunciation types and intonation types of the pronounced words to be noted, and the pronunciation keywords include voiced sound, unvented sound, unvoiced sound and the like.

In the embodiment of the present invention, the sequentially performing word shape recognition on the periodic word training dataset to obtain an error word dataset includes:

In the embodiment of the invention, the word shape recognition is carried out on the periodic word training data set to obtain the error word data set, whether the training word of the learner is correctly written in a standard way can be judged by combining the feature of the word shape, the word shape of the word shape in Japanese is more complex, the mastery degree of the word is lower by simply selecting word filling, and the word judgment is carried out by the feature of the word shape, so that a more accurate judgment result can be obtained.

In detail, the extracting the erroneous syntax data set from the periodic syntax exercise data set includes:

In detail, the grammar parsing refers to grammar parsing corresponding to the true pronunciation answer, and includes a carrier language, a help word, a salutation, a complement verb, and the like.

In the embodiment of the invention, the wrong pronunciation data set is obtained by sequentially carrying out phoneme recognition and tone recognition on the periodic pronunciation exercise data set, the wrong word data set is obtained by sequentially carrying out word shape recognition on the periodic word exercise data set, the wrong grammar data set is extracted from the periodic grammar exercise data set, and keywords of wrong knowledge points corresponding to each exercise can be screened from buried point exercise data, so that subsequent wrong clustering is facilitated.

S3, clustering the mispronounced data set into Yi Cuofa sound type sets, clustering the mispronounced word data set into error-prone word type sets, and clustering the error grammar data set into error-prone grammar type sets.

In the embodiment of the present invention, the Yi Cuofa sound collection refers to a collection composed of a plurality of error-prone pronunciation classes, each error-prone pronunciation class corresponds to a clustering result in the error-prone pronunciation data set, the error-prone word collection refers to a collection composed of a plurality of error-prone word classes, each error-prone word class corresponds to a clustering result in the error-prone word data set, the error-prone grammar collection refers to a collection composed of a plurality of error-prone grammar classes, and each error-prone pronunciation class corresponds to a clustering result in the error-prone grammar data set.

In an embodiment of the present invention, the clustering the mispronounced data set into Yi Cuofa sound sets includes:

In detail, the calculating the word vector distance between each piece of mispronounced data and each piece of mispronounced center data in the mispronounced data set refers to calculating the euclidean distance of a vector between the mispronounced data after word vectorization and the mispronounced center data after word vectorization, and the updating the mispronounced data set into the standard mispronounced data set one by one according to the word vector distance refers to dividing each piece of mispronounced data into the mispronounced data sets where the mispronounced center data with the word vector closest to each other, so as to obtain the standard mispronounced data set.

Specifically, the calculating of the standard mispronounced center data of each standard mispronounced data set refers to calculating mispronounced data identical to a word vector distance between each mispronounced data in the standard mispronounced data set as standard mispronounced center data, and the center word vector distance refers to a word vector distance between the standard mispronounced center data and the corresponding mispronounced center data.

In detail, the iterative updating of each standard mispronounced data set into a corresponding mispronounced class according to all the center word vector distances refers to a step of calculating a distance sum of all the center word vector distances, when the distance sum is greater than a preset distance sum and a threshold value, returning standard mispronounced center data as mispronounced center data to the step of calculating word vector distances between each mispronounced data and each mispronounced center data in the mispronounced data set until the distance sum is less than or equal to the distance sum and the threshold value, and taking the standard mispronounced data set at the moment as the mispronounced class.

In detail, the method for clustering the error word data set into the error prone word class set is identical to the method for clustering the error pronunciation data set into the Yi Cuofa sound class set in the above step S3, and the method for clustering the error grammar data set into the error prone grammar class set is identical to the method for clustering the error pronunciation data set into the Yi Cuofa sound class set in the above step S3, which is not repeated herein.

In the embodiment of the invention, the error pronunciation data set is clustered into the Yi Cuofa sound set, the error word data set is clustered into the error-prone word set, the error grammar data set is clustered into the error-prone grammar set, and the error-prone knowledge point keywords can be clustered and divided through a clustering algorithm, so that the step of combing and analyzing the error question set is reduced, and the generation of periodic error features of subsequent students is facilitated.

S4, splitting the target period teaching point burying data into a plurality of period student point burying data according to student identities, selecting the period student point burying data as the target student point burying data one by one, generating student period error features corresponding to the target student point burying data according to the Yi Cuofa sound collection, the error-prone word collection and the error-prone grammar collection, and clustering all the student period error features into the student period error feature collection.

In the embodiment of the invention, the student identity can be a student identity account ID used by a student logging on a Japanese teaching platform, and the periodic student buried point data is buried point data corresponding to a student in a teaching period.

In the embodiment of the present invention, the generating, according to the Yi Cuofa sound collection, the error-prone word collection and the error-prone grammar collection, the learner cycle error feature corresponding to the target learner buried data includes:

In the embodiment of the present invention, the replacing each piece of error pronunciation data in the error pronunciation data of the learner with each piece of cluster center data in the Yi Cuofa sound set to obtain standard error pronunciation data refers to selecting error pronunciation data in the error pronunciation data of the learner as target error pronunciation data one by one, selecting a cluster center of error pronunciation class where the target error pronunciation data is located from the Yi Cuofa sound set as target error pronunciation center data, and replacing the target error pronunciation data with the target error pronunciation center data to obtain standard error pronunciation data of the learner.

In the embodiment of the present invention, the clustering all the learner cycle error characteristics into the learner cycle error characteristic class set includes:

dividing all the periodic error characteristics of the students into a plurality of periodic error characteristic groups, and randomly selecting periodic error center characteristics of the students for each periodic error characteristic group;

calculating the feature similarity between each periodic error feature of the trainee and each periodic error center feature of the trainee by using the following feature similarity formula:

Wherein K is the similarity, q is the q-th dimension feature vector, A _q Refers to the q-th dimension characteristic vector in the periodic error characteristic of the learner, wherein A ₁ Refers to the learning in the periodic error feature of the learnerA person pronunciation error feature, wherein A ₂ Refers to the learner word error feature in the learner cycle error feature, A ₃ Refers to the learner grammar error feature in the learner cycle error feature, B _q Refers to the q-th dimension characteristic vector in the periodic error center characteristic of the learner, wherein B ₁ Refers to the learner pronunciation error feature in the learner cycle error center feature, wherein B ₂ Refers to the error character of the word error of the learner in the error center character of the learner cycle, B ₃ The method is characterized in that the method comprises the step of determining the learner grammar error characteristics in the learner cycle error center characteristics;

updating the periodic error characteristic groups of the students into standard error characteristic groups one by one according to the characteristic similarity;

calculating the error center characteristics of the standard students of each error characteristic group of the standard students, and calculating the similarity of the center characteristics between the error center characteristics of the standard students and the corresponding periodic error center characteristics of the students one by one;

and iteratively updating each standard learner error feature group into a corresponding learner cycle error feature class according to all the central feature similarities, and collecting all the learner cycle error feature classes into a learner cycle error feature class set.

In detail, the method for updating the periodic error feature set of the learner into the error feature set of the standard learner according to the feature similarity is consistent with the method for updating the error pronunciation data set into the standard error pronunciation data set according to the word vector distance in the step S3.

Specifically, the method for iteratively updating each standard learner error feature set to a corresponding learner cycle error feature class according to all the central feature similarities is consistent with the method for iteratively updating each standard error pronunciation data set to a corresponding error-prone pronunciation class according to all the central word vector distances in the step S3, which is not described herein.

In the embodiment of the invention, the feature similarity between each periodic error feature of the students and each periodic error center feature of the students is calculated by utilizing the feature similarity formula, so that the similarity of error-prone knowledge points of each student can be calculated in three dimensions of pronunciation, word and grammar, and the accuracy of feature similarity calculation is improved.

In the embodiment of the invention, the learner cycle error characteristics corresponding to the target learner buried data are generated according to the Yi Cuofa sound collection, the error-prone word collection and the error-prone grammar collection, and all the learner cycle error characteristics are clustered into the learner cycle error characteristic collection, so that each learner can be classified and considered according to the error-prone type of the error questions in learning interaction, and the applicability and the matching degree of Japanese teaching are improved.

S5, taking Japanese teaching embedded data of a target student as embedded data to be matched, extracting periodic error features to be matched corresponding to the embedded data to be matched according to the teaching period, selecting student periodic error feature classes corresponding to the periodic error features to be matched from the student periodic error feature classes as target periodic error feature classes, updating a teaching database of the target student according to the target periodic error feature classes, and teaching the target student by utilizing the updated teaching database.

In the embodiment of the present invention, the method for extracting the periodic error feature to be matched corresponding to the buried point data to be matched according to the teaching period is consistent with the method for selecting the periodic student buried point data one by one in the step S4 as the target student buried point data, and generating the student periodic error feature corresponding to the target student buried point data according to the Yi Cuofa sound set, the error-prone word set and the error-prone grammar set, which is not described herein.

Specifically, the selecting the learner cycle error feature class corresponding to the to-be-matched cycle error feature from the learner cycle error feature class set as the target cycle error feature class refers to calculating feature similarity between the to-be-matched cycle error feature and a cluster center of each learner cycle error feature class in the learner cycle error feature class set, and selecting the learner cycle error feature class where the cluster center with the largest feature similarity is located as the target cycle error feature class.

In the embodiment of the present invention, updating the teaching database of the target learner according to the target periodic error feature class refers to collecting error pronunciation data, error word data and error grammar data corresponding to the target periodic error feature class into a target error question set, and updating the teaching database of the target learner according to the target error question set.

In the embodiment of the invention, the learner cycle error feature class corresponding to the cycle error feature to be matched is selected from the learner cycle error feature class set to serve as the target cycle error feature class, the teaching database of the target learner is updated according to the target cycle error feature class, the updated teaching database is utilized to teach the target learner, and similar problem types of corresponding types can be generated according to error records of each learner to carry out teaching consolidation, so that the matching degree and flexibility of Japanese teaching interaction are improved.

According to the Yi Cuofa sound class set, the error-prone word class set and the error-prone grammar class set, generating the learner cycle error characteristics corresponding to the target learner buried data, clustering all the learner cycle error characteristics into the learner cycle error characteristic class set, classifying and considering each learner according to the error-prone type of the error question in learning interaction, improving the applicability and the matching degree of Japanese teaching, selecting the learner cycle error characteristic class corresponding to the cycle error characteristics to be matched from the learner cycle error characteristic class set as the target cycle error characteristic class, updating a teaching database of the target learner according to the target cycle error characteristic class, teaching the target learner by utilizing the updated teaching database, and generating similar questions of the corresponding type according to the error records of each learner to carry out teaching consolidation, thereby improving the matching degree and the flexibility of Japanese teaching interaction. Therefore, the Japanese teaching interaction method based on big data provided by the invention can solve the problem of poor flexibility in Japanese teaching interaction.

Fig. 4 is a functional block diagram of a japanese teaching interaction device based on big data according to an embodiment of the present invention.

The Japanese teaching interaction device 100 based on big data can be installed in electronic equipment. Depending on the functions implemented, the big data based japanese teaching interaction device 100 may include a data splitting module 101, an error recognition module 102, a primary clustering module 103, a secondary clustering module 104, and a teaching interaction module 105. The module of the invention, which may also be referred to as a unit, refers to a series of computer program segments, which are stored in the memory of the electronic device, capable of being executed by the processor of the electronic device and of performing a fixed function.

In the present embodiment, the functions concerning the respective modules/units are as follows:

the data splitting module 101 is configured to clean historical japanese teaching point data obtained in advance into standard teaching point data, split the standard teaching point data into a plurality of periodic teaching point data according to a teaching period, select the periodic teaching point data one by one as target periodic teaching point data, and extract a periodic pronunciation practice data set, a periodic word practice data set and a periodic grammar practice data set from the target periodic teaching point data respectively;

the error recognition module 102 is configured to sequentially perform phoneme recognition and tone recognition on the periodic pronunciation training data set to obtain an erroneous pronunciation data set, sequentially perform word shape recognition on the periodic word training data set to obtain an erroneous word data set, and extract an erroneous grammar data set from the periodic grammar training data set;

The primary clustering module 103 is configured to cluster the mispronounced data set into a Yi Cuofa sound class set, cluster the mispronounced word data set into an error prone word class set, and cluster the error grammar data set into an error prone grammar class set;

the secondary clustering module 104 is configured to split the target periodic teaching point burying data into a plurality of periodic student point burying data according to a student identity, select the periodic student point burying data one by one as target student point burying data, generate a student periodic error feature corresponding to the target student point burying data according to the Yi Cuofa sound set, the error prone word set and the error prone grammar set, cluster all the student periodic error features into a student periodic error feature set, and cluster all the student periodic error features into a student periodic error feature set, where the clustering includes: dividing all the periodic error characteristics of the students into a plurality of periodic error characteristic groups, and randomly selecting periodic error center characteristics of the students for each periodic error characteristic group; calculating the feature similarity between each periodic error feature of the trainee and each periodic error center feature of the trainee by using the following feature similarity formula:

The teaching interaction module 105 is configured to take japanese teaching embedded data of a target learner as embedded data to be matched, extract periodic error features to be matched corresponding to the embedded data to be matched according to the teaching period, select a learner periodic error feature class corresponding to the periodic error features to be matched from the learner periodic error feature class set as a target periodic error feature class, update a teaching database of the target learner according to the target periodic error feature class, and use the updated teaching database to teach the target learner.

In detail, each module in the japanese teaching interaction device 100 based on big data in the embodiment of the present invention adopts the same technical means as the japanese teaching interaction method based on big data described in fig. 1 to 3, and can generate the same technical effects, which is not described herein.

In the several embodiments provided in the present invention, it should be understood that the disclosed apparatus, device and method may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the modules is merely a logical function division, and there may be other manners of division when actually implemented.

The modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical units, may be located in one place, or may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional module in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units can be realized in a form of hardware or a form of hardware and a form of software functional modules.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof.

The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference signs in the claims shall not be construed as limiting the claim concerned.

The embodiment of the application can acquire and process the related data based on the artificial intelligence technology. Among these, artificial intelligence (Artificial Intelligence, AI) is the theory, method, technique and application system that uses a digital computer or a digital computer-controlled machine to simulate, extend and extend human intelligence, sense the environment, acquire knowledge and use knowledge to obtain optimal results.

Furthermore, it is evident that the word "comprising" does not exclude other elements or steps, and that the singular does not exclude a plurality. Multiple units or means as set forth in the system embodiments may also be implemented by one unit or means in software or hardware. The terms first, second, etc. are used to denote a name, but not any particular order.

Finally, it should be noted that the above-mentioned embodiments are merely for illustrating the technical solution of the present application and not for limiting the same, and although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made to the technical solution of the present application without departing from the spirit and scope of the technical solution of the present application.

Claims

1. The Japanese teaching interaction method based on big data is characterized by comprising the following steps:

S1: cleaning historical Japanese teaching embedded point data obtained in advance into standard teaching embedded point data, splitting the standard teaching embedded point data into a plurality of period teaching embedded point data according to a teaching period, selecting the period teaching embedded point data one by one as target period teaching embedded point data, and respectively extracting a period pronunciation practice data set, a period word practice data set and a period grammar practice data set from the target period teaching embedded point data;

s2: sequentially performing phoneme recognition and tone recognition on the periodic pronunciation training data set to obtain an incorrect pronunciation data set, sequentially performing word shape recognition on the periodic word training data set to obtain an incorrect word data set, and extracting an incorrect grammar data set from the periodic grammar training data set;

s3: clustering the mispronounced data set into Yi Cuofa sound sets, clustering the mispronounced word data set into error-prone word sets, and clustering the mispronounced grammar data set into error-prone grammar sets;

s4: splitting the target periodic teaching point burying data into a plurality of periodic student point burying data according to student identities, selecting the periodic student point burying data one by one as target student point burying data, generating student periodic error features corresponding to the target student point burying data according to the Yi Cuofa sound class set, the error-prone word class set and the error-prone grammar class set, and clustering all the student periodic error features into student periodic error feature class sets, wherein the clustering all the student periodic error features into the student periodic error feature class sets comprises the following steps:

S41: dividing all the periodic error characteristics of the students into a plurality of periodic error characteristic groups, and randomly selecting periodic error center characteristics of the students for each periodic error characteristic group;

s42: calculating the feature similarity between each periodic error feature of the trainee and each periodic error center feature of the trainee by using the following feature similarity formula:

wherein K is the similarity, q is the q-th dimension feature vector, A _q Refers to the q-th dimension characteristic vector in the periodic error characteristic of the learner, wherein A ₁ Refers to the learner pronunciation error feature in the learner cycle error feature, wherein A ₂ Refers to the learner word error feature in the learner cycle error feature, A ₃ Refers to the learner grammar error feature in the learner cycle error feature, B _q Refers to the q-th dimension characteristic vector in the periodic error center characteristic of the learner, wherein B ₁ Refers to the learner pronunciation error feature in the learner cycle error center feature, wherein B ₂ Refers to the periodic errors of the traineesLearner word error feature in center feature, B ₃ The method is characterized in that the method comprises the step of determining the learner grammar error characteristics in the learner cycle error center characteristics;

s43: updating the periodic error characteristic groups of the students into standard error characteristic groups one by one according to the characteristic similarity;

S44: calculating the error center characteristics of the standard students of each error characteristic group of the standard students, and calculating the similarity of the center characteristics between the error center characteristics of the standard students and the corresponding periodic error center characteristics of the students one by one;

s45: iteratively updating each standard learner error feature group into corresponding learner cycle error feature classes according to all the central feature similarities, and collecting all the learner cycle error feature classes into a learner cycle error feature class set;

s5: and taking Japanese teaching embedded data of a target student as embedded data to be matched, extracting periodic error features to be matched corresponding to the embedded data to be matched according to the teaching period, selecting a student periodic error feature class corresponding to the periodic error features to be matched from the student periodic error feature class set as a target periodic error feature class, updating a teaching database of the target student according to the target periodic error feature class, and teaching the target student by utilizing the updated teaching database.

2. The big data-based japanese teaching interaction method according to claim 1, wherein the cleaning the pre-acquired historical japanese teaching embedded point data into standard teaching embedded point data comprises:

3. The big data based japanese teaching interactive method according to claim 1, wherein said sequentially performing phoneme recognition and tone recognition on said periodic pronunciation training data set to obtain a mispronounced data set comprises:

4. The big data-based japanese teaching interaction method according to claim 3, wherein the sequentially performing audio correction and audio filtering operations on the target pronunciation practice data to obtain primary pronunciation practice data comprises:

5. The big data based japanese tutorial interaction method according to claim 3, wherein the sequentially performing the voice framing and the endpoint detection operation on the primary pronunciation training data to obtain secondary pronunciation training data comprises:

6. The big data-based japanese teaching interaction method according to claim 1, wherein the sequentially performing word shape recognition on the periodic word training dataset to obtain an erroneous word dataset comprises:

7. The big data based japanese tutorial interaction method of claim 1, wherein the extracting the erroneous grammar data set from the periodic grammar practice data set comprises:

8. The big data based japanese teaching interactive method according to claim 1, wherein said clustering said mispronounced data set into Yi Cuofa sound class sets comprises:

9. The big data based japanese teaching interaction method according to claim 8, wherein the generating the learner cycle error feature corresponding to the target learner buried data based on the Yi Cuofa class set, the error prone word class set, and the error prone grammar class set comprises:

10. A japanese teaching interactive device based on big data, the device comprising:

wherein K is the similarity, q is the q-th dimension feature vector, A _q Refers to the q-th dimension characteristic vector in the periodic error characteristic of the learner, wherein A ₁ Refers to the periodic error characteristics of the trainee A learner pronunciation error feature, wherein A ₂ Refers to the learner word error feature in the learner cycle error feature, A ₃ Refers to the learner grammar error feature in the learner cycle error feature, B _q Refers to the q-th dimension characteristic vector in the periodic error center characteristic of the learner, wherein B ₁ Refers to the learner pronunciation error feature in the learner cycle error center feature, wherein B ₂ Refers to the error character of the word error of the learner in the error center character of the learner cycle, B ₃ The method is characterized in that the method comprises the step of determining the learner grammar error characteristics in the learner cycle error center characteristics; updating the periodic error characteristic groups of the students into standard error characteristic groups one by one according to the characteristic similarity; calculating the error center characteristics of the standard students of each error characteristic group of the standard students, and calculating the similarity of the center characteristics between the error center characteristics of the standard students and the corresponding periodic error center characteristics of the students one by one; iteratively updating each standard learner error feature group into corresponding learner cycle error feature classes according to all the central feature similarities, and collecting all the learner cycle error feature classes into a learner cycle error feature class set;