CN115563366A - Model training and data analysis method, device, storage medium and equipment - Google Patents
Model training and data analysis method, device, storage medium and equipment Download PDFInfo
- Publication number
- CN115563366A CN115563366A CN202211160077.6A CN202211160077A CN115563366A CN 115563366 A CN115563366 A CN 115563366A CN 202211160077 A CN202211160077 A CN 202211160077A CN 115563366 A CN115563366 A CN 115563366A
- Authority
- CN
- China
- Prior art keywords
- data
- type
- fusion
- gradient information
- analysis
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/906—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The specification discloses a method, a device, a storage medium and equipment for model training and data analysis. The model training method comprises the following steps: inputting sample data into an analysis model to be trained, determining data characteristics of each type of data through a characteristic extraction layer corresponding to each type of data, obtaining analysis results corresponding to each type of data according to the data characteristics of each type of data, determining gradient information corresponding to each type of data during the training of the analysis model, fusing the data characteristics of each type of data through a characteristic fusion layer to obtain fusion characteristics, further obtaining a comprehensive analysis result, determining fusion gradient information generated by all types of data during the training of the analysis model according to the deviation between the comprehensive analysis result and a label corresponding to the sample data, and training the analysis model by taking the fusion gradient information as a gradient label and taking the minimized deviation between the gradient information corresponding to each type of data and the gradient label as an optimization target.
Description
Technical Field
The present disclosure relates to the field of computer technologies, and in particular, to a method, an apparatus, a storage medium, and a device for model training and data analysis.
Background
With the development of artificial intelligence, emotion analysis techniques are also widely applied in various fields, for example, in a human-computer interaction scene of an intelligent robot, image data of a user is usually acquired through an image acquisition device (such as a camera) to extract information such as facial expressions and body movements of the user, meanwhile, a speech tone of a user speaking is also acquired, and conversation content is extracted, so that emotions and emotions currently expressed by the user are correspondingly analyzed according to the extracted information, and then a corresponding interaction strategy is executed according to an analysis result to interact with the user.
However, in practical applications, there are often cases where emotions expressed by some types of data included in the data to be analyzed are contradictory to each other, and in such cases, the current analysis method cannot obtain an accurate analysis result, for example, when the data to be analyzed is relatively flat content spoken by a user in a violent tone, it is difficult for the existing method to determine whether the emotion expressed by the user is a positive emotion, a negative emotion, or a neutral emotion.
Therefore, how to accurately obtain an analysis result of data to be analyzed according to different types of data included in the data to be analyzed is an urgent problem to be solved.
Disclosure of Invention
The present specification provides a method, apparatus, storage medium, and device for model training and data analysis, which partially solve the above problems in the prior art.
The technical scheme adopted by the specification is as follows:
the present specification provides a method of model training, comprising:
acquiring sample data, wherein the sample data comprises at least two types of data;
inputting the sample data into an analysis model to be trained, determining the data characteristics of the type of data through a characteristic extraction layer corresponding to the type of data aiming at each type of data contained in the sample data, obtaining an analysis result corresponding to the type of data according to the data characteristics of the type of data, and determining gradient information generated by the type of data during the analysis model training as the gradient information corresponding to the type of data according to the deviation between the analysis result corresponding to the type of data and a label corresponding to the sample data;
fusing data characteristics of each type of data through a characteristic fusion layer in the analysis model to obtain fusion characteristics, obtaining a comprehensive analysis result according to the fusion characteristics, and determining gradient information generated by all types of data contained in the sample data when the analysis model is trained according to the deviation between the comprehensive analysis result and a label corresponding to the sample data to serve as fusion gradient information;
and taking the fused gradient information as a gradient label, and taking the minimized deviation between the gradient information corresponding to the data of the type and the gradient label as an optimization target to train the analysis model.
Optionally, the data features of each type of data are fused through a feature fusion layer in the analysis model to obtain fusion features, and the method specifically includes:
selecting data of a specified type from various types of data contained in the sample data as first type data, and taking other types of data except the first type data in the sample data as second type data;
for each type of second data, determining a characteristic conversion parameter corresponding to the second type of data according to the data characteristic corresponding to the first type of data and the data characteristic corresponding to the second type of data;
converting the data characteristics corresponding to the second type data through the characteristic conversion parameters to obtain converted characteristics corresponding to the second type data;
and determining the fusion characteristics according to the data characteristics corresponding to the first type data and the converted characteristics corresponding to each type of second type data.
Optionally, determining the fusion feature according to the data feature corresponding to the first type data and the converted feature corresponding to each type of second type data specifically includes:
fusing the converted features corresponding to each second type of data to obtain fused features of each second type of data;
determining the weight corresponding to the fusion feature of each second type of data according to the data feature corresponding to the first type of data and the fusion feature of each second type of data;
and according to the weight, fusing the data characteristics corresponding to the first type data with the fusion characteristics of the second type data, and determining that the fusion characteristics are obtained after the data characteristics of the data of each type are fused.
Optionally, training the analysis model with minimizing a deviation between gradient information corresponding to the type of data and the gradient label as an optimization objective specifically includes:
for each type of data, adjusting the gradient information corresponding to the type of data by fusing the gradient information and aiming at minimizing the deviation between the gradient information corresponding to the type of data and the label gradient information to obtain adjusted gradient information corresponding to the type of data;
and training a feature extraction layer corresponding to each type of data contained in the analysis model according to the adjusted gradient information corresponding to each type of data.
Optionally, the method further comprises:
adjusting the fusion gradient information through the adjusted gradient information corresponding to each type of data to obtain adjusted fusion gradient information;
and training the feature fusion layer according to the adjusted fusion gradient information.
The present specification provides a method of data analysis comprising:
acquiring data to be analyzed, wherein the data to be analyzed comprises at least two types of data;
inputting the data to be analyzed into a pre-trained analysis model, and determining data characteristics corresponding to the data of the type through a characteristic extraction layer corresponding to the data of the type aiming at each type of data contained in the data to be analyzed, wherein the analysis model is obtained by training through the model training method;
determining fusion characteristics corresponding to the data to be analyzed according to the data characteristics corresponding to each type of data through a characteristic fusion layer in the analysis model;
and determining an analysis result of the data to be analyzed according to the fusion characteristics.
The present specification provides an apparatus for model training, comprising:
the acquisition module acquires sample data, wherein the sample data comprises at least two types of data;
the input module is used for inputting the sample data into an analysis model to be trained, determining the data characteristics of the type of data through a characteristic extraction layer corresponding to the type of data aiming at each type of data contained in the sample data, obtaining an analysis result corresponding to the type of data according to the data characteristics of the type of data, and determining gradient information generated by the type of data during the analysis model training as the gradient information corresponding to the type of data according to the deviation between the analysis result corresponding to the type of data and a label corresponding to the sample data;
the fusion module is used for fusing the data characteristics of all types of data through a characteristic fusion layer in the analysis model to obtain fusion characteristics, obtaining a comprehensive analysis result according to the fusion characteristics, and determining gradient information generated by all types of data contained in the sample data when the analysis model is trained according to the deviation between the comprehensive analysis result and a label corresponding to the sample data to serve as fusion gradient information;
and the training module is used for taking the fusion gradient information as a gradient label and training the analysis model by taking the minimized deviation between the gradient information corresponding to the type of data and the gradient label as an optimization target.
The present specification provides an apparatus for data analysis, comprising:
the device comprises an acquisition module, a processing module and a processing module, wherein the acquisition module acquires data to be analyzed, and the data to be analyzed comprises at least two types of data;
the input module is used for inputting the data to be analyzed into a pre-trained analysis model, and determining data characteristics corresponding to the data of the type through a characteristic extraction layer corresponding to the data of the type aiming at each type of data contained in the data to be analyzed, wherein the analysis model is obtained by training through the model training method;
the determining module is used for determining fusion characteristics corresponding to the data to be analyzed according to the data characteristics corresponding to each type of data through a characteristic fusion layer in the analysis model;
and the analysis module is used for determining the analysis result of the data to be analyzed according to the fusion characteristics.
The present specification provides a computer-readable storage medium, which stores a computer program that, when executed by a processor, implements the above-described method of model training and data analysis.
The present specification provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the above-described method of model training and data analysis when executing the program.
The technical scheme adopted by the specification can achieve the following beneficial effects:
in the method for training a model provided in this specification, acquired sample data is input into an analysis model to be trained, so that data features of each type of data are determined through a feature extraction layer corresponding to each type of data, analysis results corresponding to each type of data are obtained, meanwhile, gradient information corresponding to each type of data is determined according to a deviation between the analysis result corresponding to each type of data and a label corresponding to sample data, then, the data features of each type of data are fused through a feature fusion layer in the analysis model to obtain fusion features, a comprehensive analysis result is obtained according to the fusion features, fusion gradient information generated by all types of data included in the sample data during training of the analysis model is determined according to the deviation between the comprehensive analysis result and the label corresponding to the sample data, the fusion gradient information is used as a gradient label, and the analysis model is trained with the goal of minimizing the deviation between the gradient information corresponding to the type of data and the gradient label.
It can be seen from the above method that in the process of training the analysis model, the method can determine the gradient information corresponding to each type of data according to the analysis result corresponding to each type of data, and determine the fusion gradient information according to the comprehensive analysis result obtained by fusing the characteristics, so as to train the analysis model by taking the minimum deviation between the gradient information corresponding to each type of data and the gradient label as the optimization target.
Drawings
The accompanying drawings, which are included to provide a further understanding of the specification and are incorporated in and constitute a part of this specification, illustrate embodiments of the specification and together with the description serve to explain the principles of the specification and not to limit the specification in a limiting sense. In the drawings:
FIG. 1 is a schematic flow chart of a method of model training provided herein;
FIG. 2 is a schematic flow chart diagram of a method of data analysis provided herein;
FIG. 3 is a schematic diagram of an apparatus for model training provided herein;
FIG. 4 is a schematic diagram of an apparatus for data analysis provided herein;
fig. 5 is a schematic diagram of an electronic device corresponding to fig. 1 provided in this specification.
Detailed Description
In order to make the objects, technical solutions and advantages of the present disclosure more clear, the technical solutions of the present disclosure will be clearly and completely described below with reference to the specific embodiments of the present disclosure and the accompanying drawings. It is to be understood that the embodiments described are only a few embodiments of the present disclosure, and not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present specification without making any creative effort belong to the protection scope of the present specification.
The technical solutions provided by the embodiments of the present description are described in detail below with reference to the accompanying drawings.
Fig. 1 is a schematic flowchart of a method for training a model provided in this specification, including the following steps:
s101: acquiring sample data, wherein the sample data comprises at least two types of data;
in many fields such as intelligent customer service, human-computer interaction, and content auditing, data analysis is generally required for different types of acquired data (such as text data, image data, audio data, etc.) to perform corresponding services according to the analysis result. For example, in an application scenario of an intelligent dialogue robot, an interactive system of the robot usually collects facial images, body movements, speaking tones, speech speeds, and text contents included in dialogues of users, and analyzes these data to determine actual emotions or emotions of users who currently interact with the interactive system, so as to enable the intelligent dialogue robot to perform corresponding interactive actions or perform corresponding voice dialogues according to analysis results.
For another example, in a content auditing scenario of a short video platform, corresponding auditing is usually required to be performed on content published by a user, and in this process, corresponding data analysis may be performed on content (such as video, text, images, voice, and the like) published by the user through an analysis model, so as to perform corresponding processing (such as deleting or masking and the like) on content that is too negative or too negative.
In practical applications, it is usually necessary to analyze data by using a corresponding model algorithm to obtain an analysis result, however, in some specific scenarios, situations often occur in which emotions expressed by some types of data included in the data to be analyzed are contradictory, such as a user speaking a relatively flat content in a violent tone or speaking a certain negative content in a normal tone, it is difficult for the model to determine a real emotion expressed by the user, and when the trained model has a high attention to one type of data, the emotion expressed by the data may even affect the final analysis result. In this process, the training mode and the training process of the model may affect the accuracy of the trained model on data analysis.
Based on this, the present specification provides a model training method, so that when training is performed through a model, gradient information corresponding to each type of data is corrected by fusion gradient information generated by all types of data included in sample data, thereby reducing the influence of a single type of data on the overall analysis result of sample data, avoiding the influence on the final analysis result of an analysis model due to mutual contradiction between information corresponding to each type of data, and further improving the accuracy of data analysis.
In practical application, the sample data may be a sample in a specific service scene, for example, in a human-computer interaction scene, the sample data may include collected text data (such as extracted dialog content), audio data, image data, and the like of a user. For another example, in a scene of content audit, the sample data may include characters, videos, audios, and the like published by the user.
The sample data may be user data of one user acquired at the same time, or may be user data acquired at a plurality of times. Of course, the sample data may also include other types of data in other service scenarios, which is not specifically limited in this specification.
In the present specification, the execution subject of the method for implementing model training and the method for data analysis may refer to a designated device such as a server installed on a service platform, and for convenience of description, the present specification only takes the server as the execution subject and describes a method for model training provided in the present specification.
S102: inputting the sample data into an analysis model to be trained, determining the data characteristics of the type of data through a characteristic extraction layer corresponding to the type of data aiming at each type of data contained in the sample data, obtaining an analysis result corresponding to the type of data according to the data characteristics of the type of data, and determining gradient information generated by the type of data during the analysis model training as the gradient information corresponding to the type of data according to the deviation between the analysis result corresponding to the type of data and a label corresponding to the sample data.
The analysis result may be an emotion type actually expressed by each type of data, such as a positive emotion, a negative emotion, and a neutral emotion, or may be some specific emotion types, such as happy emotion, sad emotion, angry, indifference, peaceful emotion, and the like, or may be other types of emotions or other types of analysis results, which is not limited in this specification.
After the server obtains the sample data, the sample data can be input into an analysis model to be trained, the analysis model comprises a feature extraction layer and a feature fusion layer, the feature extraction layer is used for extracting data features corresponding to each type of data contained in the sample data, and the feature fusion layer is used for fusing the data features corresponding to each type of data, so that fusion features corresponding to the sample data are obtained. The process of feature fusion will be described in detail below, and this description will not be repeated here.
It should be noted that the analysis model may include a plurality of feature extraction layers, where the feature extraction method in each feature extraction layer corresponds to one type of data, and after each type of data is input into the respective feature extraction layer, each feature extraction layer extracts the data features of each type of data. Of course, the analysis model may also only include one feature extraction layer, where the feature extraction layer includes a feature extraction method corresponding to each type of data, and after each type of data is input into the feature extraction layer, feature extraction is performed on each type of data by the feature extraction method corresponding to each type of data in the feature extraction layer.
Specifically, the server may extract data features corresponding to each type of data included in the sample data by analyzing the feature extraction layer of each type of data in the model, and since feature dimensions corresponding to each type of feature are different, features corresponding to different types of data may be extracted in different manners in the feature extraction layer.
For example, with respect to image data, features such as facial movements, facial markers, head pose, gaze position, and Histogram of Oriented Gradients (HOG) of a user may be extracted by a Facet tool. For text data, a video or an audio may be transcribed into text data by a corresponding transcription tool, then manually corrected after transcription is completed, and then feature extraction is performed on the transcribed text data by a feature extractor such as a text Embedding feature extractor. For audio data, features such as the first three formants of the audio data, spectral tilt or peak of the wavelet response, etc. can be extracted by a covapre tool, for example.
Of course, in this specification, the feature extraction layer may also extract data features corresponding to each type of data by using other tools or methods, which is not specifically limited in this specification.
Then, for each type of data, the server may obtain an analysis result corresponding to the type of data according to a data feature of the type of data, and determine gradient information generated by the type of data during the training of the analysis model according to a deviation between the analysis result corresponding to the type of data and a label corresponding to sample data and a data feature corresponding to the type of data, as the gradient information corresponding to the type of data, where the gradient information corresponding to each type may be represented by the following formula:
wherein y is an analysis result corresponding to the m-type data, T is a data feature corresponding to the text data, A is a data feature corresponding to the audio data, V is a data feature corresponding to the image data,
gradient information corresponding to data type m, L m (theta) is a loss value of a loss function of the analysis model determined according to the analysis result corresponding to each type of data and the actual label of the sample data, E (m，y)～B The expected value of the sample data label, and B is the sample data.
It should be noted that the gradient information determined by the above formula is the gradient information corresponding to each type, in other words, the text type, the audio type, and the video type all correspond to one gradient information respectively.
S103: and according to the deviation between the comprehensive analysis result and the label corresponding to the sample data, determining gradient information generated by all types of data contained in the sample data when the analysis model is trained as fusion gradient information.
After data features corresponding to various types of data contained in the sample data are extracted, the server can input the features into the feature fusion layer, so that the fusion features of the sample data are obtained.
In this specification, a pre-trained language Representation (BERT) model may be used as a coding tool of the feature extraction layer, and a feature fusion layer is added after the first coding layer.
Specifically, after obtaining the data features corresponding to each type of data, the server may input the data features corresponding to each type of data into the feature fusion layer, select a data of a specified type from the data of each type included in the sample data as a first type of data, use other types of data in the sample data except the first type of data as a second type of data, and determine, for each second type of data, feature conversion parameters (such as a gating vector) corresponding to the second type of data according to the data features corresponding to the first type of data and the data features corresponding to the second type of data.
Taking the data characteristic corresponding to the text data as T i The audio data corresponds to a data characteristic A i The image data corresponds to a data characteristic V i For example, in a normal case, the text type data is used as the designated data, that is, the text data is selected as the first type data, and then a is performed at this time i And V i A data characteristic corresponding to the second type of data, a data characteristic A of the audio data i The corresponding gating vector can be represented by the following formula:
wherein the content of the first and second substances,
data characteristic A for audio data i Corresponding gating vector (characteristic transformation parameter), R is a non-linear function, W ga For determining data characteristics A of audio data i Preset parameters in the corresponding gating vector, b a Are correspondingly biased.
Data characteristic V of video data i The corresponding gating vector can be represented by the following formula:
wherein the content of the first and second substances,
data characteristic V for video data i Corresponding gating vector, R being a non-linear function, W gv For determining data characteristics V of video data i Preset parameters in the case of corresponding gating vectors, b v Are correspondingly biased.
And then the server converts the data characteristics corresponding to the second type data through the gating vectors corresponding to the second type data to obtain the converted characteristics corresponding to the second type data, and further determines the fusion characteristics of the second type data according to the data characteristics corresponding to the first type data and the converted characteristics corresponding to the second type data.
Further, after obtaining the converted features corresponding to each second type of data, the server may fuse the converted features corresponding to each second type of data to obtain the fusion features of each second type of data, and the fusion features of each second type of data may be represented by the following formula:
wherein H i For the fusion characteristics of each second type of data, W a As a data characteristic A i Corresponding preset parameterNumber, W v For data features V i Corresponding preset parameters, b H Are correspondingly biased.
In general, the size of the data feature corresponding to the audio data and the data feature corresponding to the image data is often larger than that of the text data, and if the text data is too small, the text feature may not express a final analysis result, which loses the existing meaning, so to avoid the situation, the server may determine a weight corresponding to each second type of data according to the data feature corresponding to the first type of data and the fusion feature of each second type of data, where the weight may be represented by the following formula:
wherein, alpha is the weight corresponding to each second type data, beta is the corresponding adjustment factor, when the fusion characteristic of each second type data is larger than the characteristic of the data corresponding to the first type data, then
In this case, the larger the initial fusion feature is, the smaller the weight is, that is, the larger the initial fusion feature is, the smaller the weight is.
When the feature value of the initial fusion feature is smaller than the feature value of the data feature corresponding to the first type of data, the weight is 1, in other words, the maximum value of the weight is 1.
Then, the server may fuse the data features corresponding to the first type of data with the fusion features of each second type of data according to the weights, and determine the target fusion features corresponding to the sample data, where the fusion features corresponding to the sample data may be represented by the following formula:
wherein the content of the first and second substances,
the fusion features corresponding to the sample data are also the outputs corresponding to the feature fusion layer.
After the fusion characteristics of the sample data are determined, the server can determine the comprehensive analysis result of the model according to the fusion characteristics, and further determine the gradient information generated by all types of data contained in the sample data when the analysis model is trained as fusion gradient information according to the deviation between the comprehensive analysis result and the label corresponding to the sample data. The fused gradient information can be represented by the following formula:
wherein the data characteristic corresponding to the text data is T, the data characteristic corresponding to the audio data is A, the data characteristic corresponding to the image data is V,
as fusion gradient information of the model, L M (theta) is a loss value of a loss function of the analysis model determined from a deviation between the analysis result and a label corresponding to the sample data, E (T,A,V,y)～B The expected value of the expected analysis result of the model, and B is sample data.
S104: and taking the fused gradient information as a gradient label, and training the analysis model by taking the minimized deviation between the gradient information corresponding to the type of data and the gradient label as an optimization target.
Specifically, the server may adjust the gradient information corresponding to the type of data according to the fused gradient information with a goal of minimizing a deviation between the gradient information corresponding to the type of data and the label gradient information, so as to obtain adjusted gradient information corresponding to the type of data, so that the gradient information corresponding to each type of data is closer to the fused gradient information, thereby avoiding contradiction or conflict between emotions expressed by part of types of data and emotions expressed by other types of data, and further making data features corresponding to each type of data extracted by the feature extraction layer more reliable and accurate. The adjusted gradient information corresponding to each type of data can be represented by the following formula:
wherein, the first and the second end of the pipe are connected with each other,
for adjusted gradient information corresponding to type i, g i For the original gradient information corresponding to type i,
to fuse gradient information.
When the gradient information corresponding to each type is acquired, the server can train the feature extraction layer of the analysis model according to the gradient information, so that parameters corresponding to the feature extraction layer are updated, and the features corresponding to various types of data extracted by the trained feature extraction layer are more accurate.
In addition, the server can adjust the fusion gradient information according to the adjusted gradient information corresponding to each type to obtain the adjusted fusion gradient information, so that the information expressed by the fusion characteristics is more comprehensive. The adjusted fusion gradient information can be represented by the following formula:
wherein the content of the first and second substances,
the gradient information is fused after adjustment.
The fusion gradient information can act on the feature fusion layer, and the server can train the feature fusion layer of the analysis model according to the adjusted fusion gradient information, so that the parameters of the feature fusion layer are updated, and the information obtained by the trained feature fusion layer is richer and more comprehensive.
In this specification, the model may further include a processing layer in addition to the feature extraction layer and the feature fusion layer, so as to obtain a final analysis result according to the fusion feature, and in the process of training the model, the processing layer may also be trained according to the fusion gradient information, so as to update the parameters corresponding to the processing layer.
And then the server can train the analysis model through each sample data until the analysis model meets the training target. The training targets may be: the analysis model converges to a preset threshold range, or reaches a preset training time, so as to ensure that the analysis result obtained by the analysis model is the information (such as the actual expressed emotion type) actually expressed by the sample data. The preset threshold and the preset training times may be set according to actual requirements, and the present specification is not limited specifically.
After the model to be analyzed is trained, the model to be analyzed can be deployed, so that the analysis result of the data to be analyzed is determined through the analysis model, and corresponding business is executed according to the analysis result.
In the above, the method for training the model provided in the present specification is described in terms of model training, and the method for analyzing the data provided in the present specification is described below in terms of practical application of the model.
Fig. 2 is a method of data analysis provided in the present specification.
S201: acquiring data to be analyzed, wherein the data to be analyzed comprises at least two types of data.
In practical application, data to be analyzed may be data actually acquired in an interaction process in different scenes, and an execution subject of the method for implementing data analysis may be a server, and certainly, an interaction system in a terminal device may also be deployed in an actual human-computer interaction scene.
S202: inputting the data to be analyzed into a pre-trained analysis model, and determining the data characteristics corresponding to the data of the type through a characteristic extraction layer corresponding to the data of the type aiming at each type of data contained in the data to be analyzed, wherein the analysis model is obtained by training through the model training method.
After the server acquires the data to be analyzed, the data to be analyzed can be input into a pre-trained analysis model Zhang hong, so that the data features corresponding to each type of data are respectively extracted through the feature extraction layer corresponding to each type of data.
S203: and determining fusion characteristics corresponding to the data to be analyzed according to the data characteristics corresponding to each type of data through the characteristic fusion layer in the analysis model.
After determining the data features corresponding to the various types of data, the server may further input the features into the feature fusion layer, so as to fuse the data features corresponding to the various types of data through the feature fusion layer, thereby obtaining fusion features corresponding to the data to be analyzed.
S204: and determining an analysis result of the data to be analyzed according to the fusion characteristics.
Then, the server may input the fusion feature into the feature extraction layer of the model, so as to obtain an analysis result of the data to be analyzed according to the fusion feature, and after determining the analysis result, the server may perform corresponding business processing according to the analysis result, for example, determine an emotion type and conversation content expressed by the user according to the analysis result of the model, and interact with the user (for example, send a voice or text reply to the user) according to the emotion type and conversation content.
Of course, in this specification, the analysis result of the sample data may also be determined according to the data feature corresponding to each type of data, or the final analysis result may be obtained according to the data feature corresponding to each type of data and the fused data feature (for example, the analysis result determined according to the data feature corresponding to each type of data and the analysis result determined according to the fused feature are combined).
It should be noted that all actions of acquiring signals, information or data in the present application are performed under the premise of complying with the data protection regulation policy responded by the country of the location and obtaining the authorization given by the owner of the corresponding device.
According to the method, in the process of training the analysis model, the gradient information corresponding to each type of data can be determined according to the analysis result corresponding to each type of data, the fusion gradient information can be determined according to the comprehensive analysis result obtained by the fusion characteristics, so that the analysis model is trained by taking the deviation between the gradient information corresponding to each type of data and the gradient label as an optimization target, in the process, the problem of mutual conflict between information expressed by different types of data can be converted to the level of the gradient of the model, the problem of inaccurate data analysis result caused by mutual conflict between information expressed by each type of data is solved by adjusting the gradient of the model, and the accuracy of the trained analysis model for analyzing the data is further improved.
Based on the same idea, the present specification also provides a corresponding apparatus for model training and data analysis, as shown in fig. 3 or fig. 4.
Fig. 3 is a schematic diagram of a model training apparatus provided in the present specification, including:
an obtaining module 301, configured to obtain sample data, where the sample data includes at least two types of data;
an input module 302, configured to input the sample data into an analysis model to be trained, so as to determine, for each type of data included in the sample data, a data feature of the type of data through a feature extraction layer corresponding to the type of data, obtain an analysis result corresponding to the type of data according to the data feature of the type of data, and determine, according to a deviation between the analysis result corresponding to the type of data and a label corresponding to the sample data, gradient information generated by the type of data during training of the analysis model, as gradient information corresponding to the type of data;
a fusion module 303, configured to fuse data features of each type of data through a feature fusion layer in the analysis model to obtain fusion features, obtain a comprehensive analysis result according to the fusion features, and determine, according to a deviation between the comprehensive analysis result and a tag corresponding to the sample data, gradient information generated by all types of data included in the sample data when the analysis model is trained, as fusion gradient information;
a training module 304, configured to use the fusion gradient information as a gradient label, and train the analysis model with a goal of minimizing a deviation between gradient information corresponding to the type of data and the gradient label.
Optionally, the fusion module 303 is specifically configured to select data of a specified type from the types of data included in the sample data, to serve as first type data, and use other types of data in the sample data except the first type data as second type data; for each type of second data, determining a characteristic conversion parameter corresponding to the second type of data according to the data characteristic corresponding to the first type of data and the data characteristic corresponding to the second type of data; converting the data characteristics corresponding to the second type data through the characteristic conversion parameters to obtain converted characteristics corresponding to the second type data; and determining the fusion characteristics according to the data characteristics corresponding to the first type data and the converted characteristics corresponding to each type of second type data.
Optionally, the fusion module 303 is specifically configured to fuse the converted features corresponding to each type of second-type data to obtain a fusion feature of each type of second-type data; determining the weight corresponding to the fusion feature of each second type of data according to the data feature corresponding to the first type of data and the fusion feature of each second type of data; and according to the weight, fusing the data characteristics corresponding to the first type data with the fusion characteristics of the second type data, and determining that the fusion characteristics are obtained after the data characteristics of the data of each type are fused.
Optionally, the training module 304 is specifically configured to, for each type of data, adjust the gradient information corresponding to the type of data by using the fusion gradient information to minimize a deviation between the gradient information corresponding to the type of data and the label gradient information, so as to obtain adjusted gradient information corresponding to the type of data; and training a feature extraction layer corresponding to each type of data contained in the analysis model according to the adjusted gradient information corresponding to each type of data.
Optionally, the training module 304 is further configured to adjust the fusion gradient information according to the adjusted gradient information corresponding to each type of data, so as to obtain adjusted fusion gradient information; and training the feature fusion layer according to the adjusted fusion gradient information.
Fig. 4 is a schematic diagram of a data analysis method provided in the present specification, including:
the acquisition module 401 acquires data to be analyzed, wherein the data to be analyzed includes at least two types of data;
an input module 402, configured to input the data to be analyzed into a pre-trained analysis model, and determine, for each type of data included in the data to be analyzed, a data feature corresponding to the type of data through a feature extraction layer corresponding to the type of data, where the analysis model is obtained by training through the model training method;
a determining module 403, configured to determine, through a feature fusion layer in the analysis model, fusion features corresponding to the data to be analyzed according to data features corresponding to each type of data;
and the analysis module 404 determines an analysis result of the data to be analyzed according to the fusion feature.
The present specification also provides a computer readable storage medium having stored thereon a computer program operable to perform a method of model training and data analysis as provided above with respect to fig. 1 or 2.
The present specification also provides a schematic block diagram of an electronic device corresponding to fig. 1 shown in fig. 5. As shown in fig. 5, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include hardware required by other services. The processor reads a corresponding computer program from the non-volatile memory into the memory and then runs the computer program to implement the method for model training and data analysis described in fig. 1 or fig. 2. Of course, besides the software implementation, the present specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may be hardware or logic devices.
In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical blocks. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as ABEL (Advanced Boolean Expression Language), AHDL (alternate Hardware Description Language), traffic, CUPL (core universal Programming Language), HDCal, jhddl (Java Hardware Description Language), lava, lola, HDL, PALASM, rhyd (Hardware Description Language), and vhigh-Language (Hardware Description Language), which is currently used in most popular applications. It will also be apparent to those skilled in the art that hardware circuitry for implementing the logical method flows can be readily obtained by a mere need to program the method flows with some of the hardware description languages described above and into an integrated circuit.
The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer readable medium that stores computer readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and embedded microcontrollers, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, atmel AT91SAM, microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be regarded as a hardware component and the means for performing the various functions included therein may also be regarded as structures within the hardware component. Or even means for performing the functions may be conceived to be both a software module implementing the method and a structure within a hardware component.
The systems, apparatuses, modules or units described in the above embodiments may be specifically implemented by a computer chip or an entity, or implemented by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being divided into various units by function, respectively. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.
As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and so forth) having computer-usable program code embodied therein.
The description has been presented with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the description. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of other like elements in a process, method, article, or apparatus comprising the element.
As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
This description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
The above description is only an example of the present disclosure, and is not intended to limit the present disclosure. Various modifications and alterations to this description will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.
Claims (10)
1. A method of model training, comprising:
acquiring sample data, wherein the sample data comprises at least two types of data;
inputting the sample data into an analysis model to be trained, determining the data characteristics of the type of data through a characteristic extraction layer corresponding to the type of data aiming at each type of data contained in the sample data, obtaining an analysis result corresponding to the type of data according to the data characteristics of the type of data, and determining gradient information generated by the type of data during the analysis model training as the gradient information corresponding to the type of data according to the deviation between the analysis result corresponding to the type of data and a label corresponding to the sample data;
fusing the data characteristics of each type of data through a characteristic fusion layer in the analysis model to obtain fusion characteristics, obtaining a comprehensive analysis result according to the fusion characteristics, and determining gradient information generated by all types of data contained in the sample data when the analysis model is trained according to the deviation between the comprehensive analysis result and a label corresponding to the sample data as fusion gradient information;
and taking the fused gradient information as a gradient label, and taking the minimized deviation between the gradient information corresponding to the data of the type and the gradient label as an optimization target to train the analysis model.
2. The method according to claim 1, wherein the fusing the data features of each type of data through a feature fusion layer in the analysis model to obtain a fusion feature comprises:
selecting data of a specified type from various types of data contained in the sample data as first type data, and taking other types of data except the first type data in the sample data as second type data;
for each type of second data, determining a characteristic conversion parameter corresponding to the second type of data according to the data characteristic corresponding to the first type of data and the data characteristic corresponding to the second type of data;
converting the data characteristics corresponding to the second type data through the characteristic conversion parameters to obtain converted characteristics corresponding to the second type data;
and determining the fusion characteristics according to the data characteristics corresponding to the first type data and the converted characteristics corresponding to each type of second type data.
3. The method according to claim 2, wherein determining the fusion characteristics according to the data characteristics corresponding to the first type of data and the converted characteristics corresponding to each second type of data specifically comprises:
fusing the converted features corresponding to each second type of data to obtain the fused features of each second type of data;
determining the weight corresponding to the fusion feature of each second type of data according to the data feature corresponding to the first type of data and the fusion feature of each second type of data;
and fusing the data features corresponding to the first type of data with the fusion features of the second type of data according to the weight, and determining to fuse the data features of the types of data to obtain the fusion features.
4. The method of claim 1, wherein training the analytical model with the objective of minimizing a deviation between gradient information corresponding to the type of data and the gradient label comprises:
for each type of data, adjusting the gradient information corresponding to the type of data by fusing the gradient information and aiming at minimizing the deviation between the gradient information corresponding to the type of data and the label gradient information to obtain adjusted gradient information corresponding to the type of data;
and training a feature extraction layer corresponding to each type of data contained in the analysis model according to the adjusted gradient information corresponding to each type of data.
5. The method of claim 4, wherein the method further comprises:
adjusting the fusion gradient information through the adjusted gradient information corresponding to each type of data to obtain adjusted fusion gradient information;
and training the feature fusion layer according to the adjusted fusion gradient information.
6. A method of data analysis, comprising:
acquiring data to be analyzed, wherein the data to be analyzed comprises at least two types of data;
inputting the data to be analyzed into a pre-trained analysis model, and determining data characteristics corresponding to the type of data through a characteristic extraction layer corresponding to the type of data aiming at each type of data contained in the data to be analyzed, wherein the analysis model is obtained by training through the model training method according to any one of claims 1 to 5;
determining fusion characteristics corresponding to the data to be analyzed according to the data characteristics corresponding to each type of data through a characteristic fusion layer in the analysis model;
and determining an analysis result of the data to be analyzed according to the fusion characteristics.
7. An apparatus for model training, comprising:
the acquisition module acquires sample data, wherein the sample data comprises at least two types of data;
the input module is used for inputting the sample data into an analysis model to be trained, determining the data characteristics of the type of data through a characteristic extraction layer corresponding to the type of data aiming at each type of data contained in the sample data, obtaining an analysis result corresponding to the type of data according to the data characteristics of the type of data, and determining gradient information generated by the type of data during the analysis model training as the gradient information corresponding to the type of data according to the deviation between the analysis result corresponding to the type of data and a label corresponding to the sample data;
the fusion module is used for fusing the data characteristics of all types of data through a characteristic fusion layer in the analysis model to obtain fusion characteristics, obtaining a comprehensive analysis result according to the fusion characteristics, and determining gradient information generated by all types of data contained in the sample data when the analysis model is trained according to the deviation between the comprehensive analysis result and a label corresponding to the sample data to be used as fusion gradient information;
and the training module is used for training the analysis model by taking the fused gradient information as a gradient label and taking the minimized deviation between the gradient information corresponding to the type of data and the gradient label as an optimization target.
8. An apparatus for data analysis, comprising:
the device comprises an acquisition module, a processing module and a display module, wherein the acquisition module acquires data to be analyzed, and the data to be analyzed comprises at least two types of data;
an input module, which inputs the data to be analyzed into a pre-trained analysis model, and determines the data characteristics corresponding to the type of data through a characteristic extraction layer corresponding to the type of data for each type of data included in the data to be analyzed, wherein the analysis model is obtained by training through the model training method according to any one of claims 1 to 5;
the determining module is used for determining fusion characteristics corresponding to the data to be analyzed according to the data characteristics corresponding to each type of data through a characteristic fusion layer in the analysis model;
and the analysis module is used for determining the analysis result of the data to be analyzed according to the fusion characteristics.
9. A computer-readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1 to 6.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1 to 6 when executing the program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211160077.6A CN115563366A (en) | 2022-09-22 | 2022-09-22 | Model training and data analysis method, device, storage medium and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211160077.6A CN115563366A (en) | 2022-09-22 | 2022-09-22 | Model training and data analysis method, device, storage medium and equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115563366A true CN115563366A (en) | 2023-01-03 |
Family
ID=84741587
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211160077.6A Pending CN115563366A (en) | 2022-09-22 | 2022-09-22 | Model training and data analysis method, device, storage medium and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115563366A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115828162A (en) * | 2023-02-08 | 2023-03-21 | 支付宝(杭州)信息技术有限公司 | Classification model training method and device, storage medium and electronic equipment |
| CN116028820A (en) * | 2023-03-20 | 2023-04-28 | 支付宝(杭州)信息技术有限公司 | Model training method and device, storage medium and electronic equipment |
| CN117414135A (en) * | 2023-10-20 | 2024-01-19 | 郑州师范学院 | Behavioral and psychological abnormality detection method, system and storage medium |
-
2022
- 2022-09-22 CN CN202211160077.6A patent/CN115563366A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115828162A (en) * | 2023-02-08 | 2023-03-21 | 支付宝(杭州)信息技术有限公司 | Classification model training method and device, storage medium and electronic equipment |
| CN116028820A (en) * | 2023-03-20 | 2023-04-28 | 支付宝(杭州)信息技术有限公司 | Model training method and device, storage medium and electronic equipment |
| CN116028820B (en) * | 2023-03-20 | 2023-07-04 | 支付宝(杭州)信息技术有限公司 | Model training method and device, storage medium and electronic equipment |
| CN117414135A (en) * | 2023-10-20 | 2024-01-19 | 郑州师范学院 | Behavioral and psychological abnormality detection method, system and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115563366A (en) | Model training and data analysis method, device, storage medium and equipment | |
| EP3608822B1 (en) | Method and apparatus for detecting model security and electronic device | |
| CN115952272B (en) | Method, device and equipment for generating dialogue information and readable storage medium | |
| CN111581470B (en) | Multi-mode fusion learning analysis method and system for scene matching of dialogue system | |
| CN115545002B (en) | Model training and business processing method, device, storage medium and equipment | |
| CN115828162B (en) | Classification model training method and device, storage medium and electronic equipment | |
| CN115203394A (en) | Model training method, service execution method and device | |
| CN112860871A (en) | Natural language understanding model training method, natural language understanding method and device | |
| CN115146601A (en) | Method and device for executing language processing task, readable storage medium and equipment | |
| CN112735407B (en) | Dialogue processing method and device | |
| CN113887206B (en) | Model training and keyword extraction method and device | |
| CN116630480B (en) | Interactive text-driven image editing method and device and electronic equipment | |
| WO2024060852A1 (en) | Model ownership verification method and apparatus, storage medium and electronic device | |
| CN116186330B (en) | Video deduplication method and device based on multi-mode learning | |
| CN117409466A (en) | Three-dimensional dynamic expression generation method and device based on multi-label control | |
| CN112199963A (en) | Text processing method and device and text processing device | |
| CN116757208A (en) | Data processing method, device and equipment | |
| CN115019781A (en) | Conversation service execution method, device, storage medium and electronic equipment | |
| CN115545720A (en) | Model training method, business wind control method and business wind control device | |
| CN110728137B (en) | Method and device for word segmentation | |
| CN116434787B (en) | Voice emotion recognition method and device, storage medium and electronic equipment | |
| CN111798849A (en) | Robot instruction identification method and device, electronic equipment and storage medium | |
| CN116501852B (en) | Controllable dialogue model training method and device, storage medium and electronic equipment | |
| CN117116267B (en) | Speech recognition method and device, electronic equipment and storage medium | |
| CN115423485B (en) | Data processing method, device and equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |