CN114707010A - Model training and medium information processing method, device, equipment and storage medium - Google Patents
Model training and medium information processing method, device, equipment and storage mediumDownload PDFInfo
- Publication number
- CN114707010A CN114707010ACN202210163051.0ACN202210163051ACN114707010ACN 114707010 ACN114707010 ACN 114707010ACN 202210163051 ACN202210163051 ACN 202210163051ACN 114707010 ACN114707010 ACN 114707010A
- Authority
- CN
- China
- Prior art keywords
- media information
- sample
- original
- negative
- target
- 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 OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
- G06F16/45—Clustering; Classification
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Translated fromChinese
Description
Translated fromChinese技术领域technical field
本公开涉及人工智能技术领域，尤其涉及深度学习和计算机视觉技术，具有涉及一种模型训练和媒介信息处理方法、装置、设备及存储介质。The present disclosure relates to the field of artificial intelligence technologies, in particular to deep learning and computer vision technologies, and relates to a method, apparatus, device and storage medium for model training and media information processing.
背景技术Background technique
随着互联网技术的不断发展，媒介信息的数量和种类越来越多。如何向用户提供高质量的媒介信息物料至关重要。With the continuous development of Internet technology, the quantity and types of media information are increasing. How to provide users with high-quality media information materials is very important.
发明内容SUMMARY OF THE INVENTION
本公开提供了一种模型训练和媒介信息处理方法、装置、设备及存储介质。The present disclosure provides a method, apparatus, device and storage medium for model training and media information processing.
根据本公开的一方面，提供了一种模型训练方法，该方法包括：According to an aspect of the present disclosure, a model training method is provided, the method comprising:
根据样本媒介信息物料，构建三元组；其中，所述三元组包括正样本物料、负样本物料和预测样本物料；According to the sample media information material, a triplet is constructed; wherein, the triplet includes a positive sample material, a negative sample material and a predicted sample material;
采用所述三元组，对神经网络模型进行训练，得到目标质量分类模型；所述神经网络模型包括三个结构相同的子质量分类模型。Using the triplet, a neural network model is trained to obtain a target quality classification model; the neural network model includes three sub-quality classification models with the same structure.
根据本公开的另一方面，提供了一种媒介信息处理方法，该方法包括：According to another aspect of the present disclosure, there is provided a media information processing method, the method comprising:
获取目标媒介信息物料；Obtain target media information materials;
采用目标质量分类模型对所述目标媒介信息物料进行分类；其中，所述目标质量分类模型通过本公开任一所述的模型训练方法训练得到；A target quality classification model is used to classify the target media information material; wherein, the target quality classification model is obtained by training any one of the model training methods described in the present disclosure;
根据分类结果，对所述目标媒介信息物料进行处理。According to the classification result, the target media information material is processed.
根据本公开的另一方面，提供了一种电子设备，该电子设备包括：According to another aspect of the present disclosure, there is provided an electronic device comprising:
至少一个处理器；以及at least one processor; and
与所述至少一个处理器通信连接的存储器；其中，a memory communicatively coupled to the at least one processor; wherein,
所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行本公开任一实施例所述的模型训练方法或媒介信息处理方法。The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the model training of any embodiment of the present disclosure method or media information processing method.
根据本公开的另一方面，提供了一种存储有计算机指令的非瞬时计算机可读存储介质，其中，所述计算机指令用于使计算机执行本公开任一实施例所述的模型训练方法或媒介信息处理方法。According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute the model training method or medium described in any embodiment of the present disclosure Information processing method.
根据本公开的技术，能够提高目标质量分类模型的识别精度，进而实现了高效且精准向用户提供高质量的媒介信息物料。According to the technology of the present disclosure, the recognition accuracy of the target quality classification model can be improved, thereby realizing efficient and accurate provision of high-quality media information materials to users.
应当理解，本部分所描述的内容并非旨在标识本公开的实施例的关键或重要特征，也不用于限制本公开的范围。本公开的其它特征将通过以下的说明书而变得容易理解。It should be understood that what is described in this section is not intended to identify key or critical features of embodiments of the disclosure, nor is it intended to limit the scope of the disclosure. Other features of the present disclosure will become readily understood from the following description.
附图说明Description of drawings
附图用于更好地理解本方案，不构成对本公开的限定。其中：The accompanying drawings are used for better understanding of the present solution, and do not constitute a limitation to the present disclosure. in:
图1A是根据本公开实施例提供的一种模型训练方法的流程图；1A is a flowchart of a model training method provided according to an embodiment of the present disclosure;
图1B是根据本公开实施例提供的一种神经网络模型的结构示意图；1B is a schematic structural diagram of a neural network model provided according to an embodiment of the present disclosure;
图1C是根据本公开实施例提供的一种子质量分类模型的结构示意图；1C is a schematic structural diagram of a sub-quality classification model provided according to an embodiment of the present disclosure;
图2是根据本公开实施例提供的另一种模型训练方法的流程图；2 is a flowchart of another model training method provided according to an embodiment of the present disclosure;
图3是根据本公开实施例提供的一种媒介信息处理方法的流程图；3 is a flowchart of a method for processing media information provided according to an embodiment of the present disclosure;
图4是根据本公开实施例提供的另一种媒介信息处理方法的流程图；4 is a flowchart of another media information processing method provided according to an embodiment of the present disclosure;
图5是根据本公开实施例提供的一种模型训练装置的结构示意图；5 is a schematic structural diagram of a model training apparatus provided according to an embodiment of the present disclosure;
图6是根据本公开实施例提供的一种媒介信息处理装置的结构示意图；6 is a schematic structural diagram of a medium information processing apparatus provided according to an embodiment of the present disclosure;
图7是用来实现本公开实施例的数据处理方法或媒介信息处理的电子设备的框图。FIG. 7 is a block diagram of an electronic device for implementing the data processing method or media information processing according to the embodiment of the present disclosure.
具体实施方式Detailed ways
以下结合附图对本公开的示范性实施例做出说明，其中包括本公开实施例的各种细节以助于理解，应当将它们认为仅仅是示范性的。因此，本领域普通技术人员应当认识到，可以对这里描述的实施例做出各种改变和修改，而不会背离本公开的范围和精神。同样，为了清楚和简明，以下的描述中省略了对公知功能和结构的描述。Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding and should be considered as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted from the following description for clarity and conciseness.
图1A是根据本公开实施例提供的一种模型训练方法的流程图。该方法适用于如何精准确定媒介信息质量的情况。该方法可以由模型训练装置来执行，该装置可以采用软件和/或硬件的方式实现，并可集成配置有模型训练功能的电子设备中。如图1A所示，本实施例的模型训练方法可以包括：FIG. 1A is a flowchart of a model training method provided according to an embodiment of the present disclosure. This method is suitable for how to accurately determine the quality of media information. The method may be performed by a model training apparatus, which may be implemented in software and/or hardware, and may be integrated into an electronic device configured with a model training function. As shown in FIG. 1A , the model training method of this embodiment may include:
S101，根据样本媒介信息物料，构建三元组。S101 , constructing a triplet according to the sample media information material.
其中，媒介信息可以包括广告、新闻和书籍等，在本实施例中媒介信息优选为敏感类型的广告，比如植发广告、口腔美容广告等；媒介信息物料为媒介信息展现的内容，可以是文字、图片或Flash类型的物料等。所谓样本媒介信息物料可以为从互联网中获取的媒介信息物料，例如可以是经过图像处理后满足神经网络输入要求的图像。Among them, the media information may include advertisements, news, books, etc. In this embodiment, the media information is preferably a sensitive type of advertisement, such as a hair transplant advertisement, an advertisement for oral beauty, etc.; the media information material is the content displayed by the media information, which can be text, Pictures or Flash-type materials, etc. The so-called sample media information material may be the media information material obtained from the Internet, for example, may be an image that meets the input requirements of the neural network after image processing.
所谓三元组可以包括正样本物料、负样本物料和预测样本物料。其中，正样本物料为高质物料，负样本物料为低质物料，预测样本物料为预测时所需的样本物料，可以为预测正样本物料或预测负样本物料。需要说明的是，正样本物料、负样本物料和预测样本物料的数量相同。The so-called triples can include positive sample materials, negative sample materials and predicted sample materials. Among them, the positive sample materials are high-quality materials, the negative sample materials are low-quality materials, and the forecast sample materials are the sample materials required for forecasting, which can be forecast positive sample materials or forecast negative sample materials. It should be noted that the quantities of positive sample materials, negative sample materials and forecasted sample materials are the same.
具体的，可以从样本媒介信息物料中选择设定数量的媒介信息物料作为正样本物料，选择设定数量的媒介信息物料作为负样本物料，再选则设定数量的媒介信息物料作为预测样本物料。Specifically, a set number of media information materials can be selected from the sample media information materials as positive sample materials, a set number of media information materials can be selected as negative sample materials, and then a set number of media information materials can be selected as predicted sample materials .
S102，采用三元组，对神经网络模型进行训练，得到目标质量分类模型。S102, using triples to train a neural network model to obtain a target quality classification model.
本实施例中，神经网络模型包括三个结构相同的子质量分类模型，例如图1B所示，神经网络模型包括子质量分类模型A、子质量分类模型B和子质量分类模型C，且子质量分类模型A、子质量分类模型B和子质量分类模型C的网络结构完全相同。In this embodiment, the neural network model includes three sub-quality classification models with the same structure. For example, as shown in FIG. 1B , the neural network model includes a sub-quality classification model A, a sub-quality classification model B, and a sub-quality classification model C, and the sub-quality classification model The network structures of model A, sub-quality classification model B and sub-quality classification model C are exactly the same.
子质量分类模型用于识别媒介信息物料的类别，可以包括一个卷积神经网络和一个全连接网络；其中，卷积神经网络用于特征提取，可以包括至少一个卷积层、至少一个激活层和至少两个池化层；全连接网络用于分类。The sub-quality classification model is used to identify the categories of media information materials, and can include a convolutional neural network and a fully connected network; wherein, the convolutional neural network is used for feature extraction, and can include at least one convolutional layer, at least one activation layer and At least two pooling layers; a fully connected network is used for classification.
需要说明的是，卷积层的卷积核可以根据训练数据(即样本媒介信息物料)进行配置，例如，当训练集数据小的时候，使用多个卷积核，可以得到多个降维新矩阵，等效于增加训练集，对模型进行数据增量训练；当训练集数据大的时候，可选择少量卷积核。It should be noted that the convolution kernel of the convolution layer can be configured according to the training data (that is, the sample medium information material). For example, when the training set data is small, multiple convolution kernels can be used to obtain multiple new dimensionality reduction matrices. , which is equivalent to increasing the training set and performing incremental data training on the model; when the training set data is large, a small number of convolution kernels can be selected.
此外，还需要说明的是，子质量分类模型是预先采用样本媒介信息物料训练好的。可选的，子质量分类模型中的卷积神经网络可以包括一个卷积层、一个激活层和两个池化层，具体训练过程如下：将样本媒介信息物料输入至质量分类模型中的卷积神经网络的卷积层C1，经过卷积层处理，输出特征图谱；将卷积层C1层输出的特征图谱输入至一个激活层S2，该激活层S2对卷积层C1输出的特征图谱中的像素进行求和、加权重、加偏置，通过Sigmod函数得到该特征图谱在激活层S2的输出，即一个特征映射图谱；将激活层S2输出的特征映射图谱输入至池化层C3，该池化层C3对激活层S2输出的特征映射图谱进行滤波(比如平均池化或最大池化操作)，得到池化层C3的输出，即一个特征图谱；将池化层C3输出的特征图谱中的像素进行求和、加权重、加偏置，通过Sigmod函数得到该特征图谱在激活层S4的输出，即一个特征映射图谱；将激活层S4输出的特征映射图谱输入至全连接网络，得到预测结果；根据预测结果和样本监督数据，确定子质量分类模型的训练损失值；进而根据训练损失值，对子质量分类模型进行训练，直到迭代次数达到设定次数，或者训练损失值达到设定数值，则停止训练，将停止训练时的模型作为训练好的子质量分类模型。其中，设定次数和设定数值可以由本领域技术人员根据实际情况设定；训练损失值可以基于Focal Loss+Lovasz的损失函数确定；训练过程可以采用Adam优化算法，以及采用在epoch逐渐增加时学习率减小的Learning Rate策略。In addition, it should be noted that the sub-quality classification model is pre-trained using sample media information materials. Optionally, the convolutional neural network in the sub-quality classification model may include one convolution layer, one activation layer, and two pooling layers, and the specific training process is as follows: input the sample media information material into the convolutional layer in the quality classification model. The convolutional layer C1 of the neural network is processed by the convolutional layer to output a feature map; the feature map output by the convolutional layer C1 layer is input to an activation layer S2, and the activation layer S2 outputs the feature map of the convolutional layer C1. The pixels are summed, weighted, and biased, and the output of the feature map in the activation layer S2 is obtained through the sigmod function, that is, a feature map; the feature map output by the activation layer S2 is input to the pooling layer C3, the pool The pooling layer C3 filters the feature map output by the activation layer S2 (such as average pooling or maximum pooling) to obtain the output of the pooling layer C3, that is, a feature map; The pixels are summed, weighted, and biased, and the output of the feature map in the activation layer S4 is obtained through the Sigmod function, that is, a feature map map; the feature map map output by the activation layer S4 is input to the fully connected network to obtain the prediction result. ; Determine the training loss value of the sub-quality classification model according to the prediction results and sample supervision data; and then train the sub-quality classification model according to the training loss value until the number of iterations reaches the set number of times, or the training loss value reaches the set value, Then stop training, and use the model at the time of stopping training as the trained sub-quality classification model. Among them, the set times and the set value can be set by those skilled in the art according to the actual situation; the training loss value can be determined based on the loss function of Focal Loss+Lovasz; the training process can use the Adam optimization algorithm, and use the learning when the epoch gradually increases. Learning Rate strategy with reduced rate.
所谓目标质量分类模型用于对目标媒介信息物料进行质量分类(即高质物料或低质物料)。The so-called target quality classification model is used to classify the quality of target media information materials (ie high-quality materials or low-quality materials).
具体的，可以采用三元组，对神经网络模型进行度量训练，例如，结合图1B，可以将三元组中的正样本物料输入至子质量分类模型A，将三元组中的负样本物料输入至子质量分类模型B，将三元组中的预测样本物料输入至子质量分类模型C，进行训练，直到训练次数达到预设次数，或者目标损失值达到预设值，停止对神经网络模型训练，将停止训练时的神经网络模型作为目标质量分类模型。其中，预设次数和预设值可以由本领域技术人员根据实际情况设定；目标损失值可以基于三元组损失函数(TripletLoss)确定，训练过程可以采用Adam优化算法，以及采用在epoch逐渐增加时学习率减小的Learning Rate策略。Specifically, triples can be used to train the neural network model. For example, with reference to Figure 1B, the positive sample materials in the triplet can be input into the sub-quality classification model A, and the negative sample materials in the triplet can be input into the sub-quality classification model A. Input to the sub-quality classification model B, input the predicted sample materials in the triplet to the sub-quality classification model C, and train until the number of training times reaches the preset number of times, or the target loss value reaches the preset value, stop the neural network model. For training, the neural network model at the time of stopping training is used as the target quality classification model. The preset number of times and the preset value can be set by those skilled in the art according to the actual situation; the target loss value can be determined based on the triplet loss function (TripletLoss), and the training process can use the Adam optimization algorithm, and when the epoch gradually increases Learning Rate policy with reduced learning rate.
需要说明的是，采用度量学习机制对神经网络模型进行训练得到的目标质量分类模型，可以在面对新的未见过的媒介信息物料时，也能够准确的对该媒介信息物料进行分类。It should be noted that the target quality classification model obtained by using the metric learning mechanism to train the neural network model can accurately classify the media information materials when faced with new unseen media information materials.
本公开实施例提供的技术方案，通过根据样本媒介信息物料，构建三元组；其中，三元组包括正样本物料、负样本物料和预测样本物料，之后采用三元组，对神经网络模型进行训练，得到目标质量分类模型；神经网络模型包括三个结构相同的子质量分类模型。上述技术方案，引入三元组，对三个结构相同的子质量分类模型进行度量训练，所得到的目标质量分类模型能够对媒介信息物料的质量进行精准识别，进而实现了高效且精准地向用户提供高质量的媒介信息物料。According to the technical solution provided by the embodiments of the present disclosure, a triplet is constructed according to the sample media information materials; wherein, the triplet includes a positive sample material, a negative sample material, and a predicted sample material, and then the triplet is used to perform the neural network model. After training, the target quality classification model is obtained; the neural network model includes three sub-quality classification models with the same structure. In the above technical solution, triples are introduced to perform measurement training on three sub-quality classification models with the same structure, and the obtained target quality classification model can accurately identify the quality of media information materials, thereby realizing efficient and accurate information to users. Provide high-quality media information materials.
在上述实施例的基础上，为提高目标质量分类模型的准确性，作为本公开的一种可选方式，子质量分类模型包括至少两个不同的卷积神经网络、一个注意力机制模块和一个全连接网络；至少两个不同的卷积神经网络并行连接于注意力机制模块的一端，注意力机制模块的另一端连接于全连接网络。On the basis of the above embodiment, in order to improve the accuracy of the target quality classification model, as an optional method of the present disclosure, the sub-quality classification model includes at least two different convolutional neural networks, an attention mechanism module and a Fully connected network; at least two different convolutional neural networks are connected in parallel to one end of the attention mechanism module, and the other end of the attention mechanism module is connected to the fully connected network.
其中，卷积神经网络包括至少一个卷积层、至少一个激活层和至少两个池化层。不同的卷积神经网络中卷积层的卷积核不同。注意力机制模型用于对卷积神经网络的输出特征的权值进行调整。Wherein, the convolutional neural network includes at least one convolution layer, at least one activation layer and at least two pooling layers. The convolution kernels of the convolutional layers in different convolutional neural networks are different. The attention mechanism model is used to adjust the weights of the output features of the convolutional neural network.
优选的，如图1C所示，子质量分类模型可以包括三个卷积神经网络和NN网络(由一个注意力机制模块和一个全连接层组成)，每个卷积神经网络可以包括一个卷积层C1，一个激活层S2，一个池化层C3和一个激活层S4。具体的，子质量分类模型的训练过程如下：将每个样本媒介信息物料分别输入至子质量分类模型的三个并行的卷积神经网络的卷积层C1，经过卷积层C1处理后，得到三个特征图谱；对于每一特征图谱，对该特征图谱中每组的四个像素进行加权值，加偏置，通过一个Sigmoid函数得到该特征图片在激活层S2层的输出。对激活层S2层输出的每一映射特征图谱再经过滤波(比如平均池化或最大池化)得到池化层C3层的输出。对于C3层的每一输出，可以对该输出中每组的四个像素进行求和、加权值、加偏置，并通过sigmoid函数，产出该输出在S4层的输出。通过注意力机制模块确定激活层S4输出的三个特征的权值；基于权值，将激活层S4输出的三个特征拼接成一个向量；该向量通过全连接网络输出该图像属于某一类别的预测概率(即预测结果)。进而基于样本媒介信息物料的预测结果和样本监督数据，确定训练损失值，并根据损失值对子质量分类模型进行训练，直到训练次数达到设定次数，或者训练损失值达到设定值，停止模型训练，得到训练好的子质量分类模型。Preferably, as shown in Figure 1C, the sub-quality classification model may include three convolutional neural networks and NN networks (composed of an attention mechanism module and a fully connected layer), and each convolutional neural network may include a convolutional neural network Layer C1, one activation layer S2, one pooling layer C3 and one activation layer S4. Specifically, the training process of the sub-quality classification model is as follows: each sample media information material is respectively input into the convolutional layer C1 of the three parallel convolutional neural networks of the sub-quality classification model, and after the convolutional layer C1 is processed, the obtained Three feature maps; for each feature map, the four pixels of each group in the feature map are weighted and biased, and a sigmoid function is used to obtain the output of the feature image in the activation layer S2. The output of the pooling layer C3 is obtained by filtering (such as average pooling or maximum pooling) for each mapped feature map output by the activation layer S2. For each output of the C3 layer, the four pixels of each group in the output can be summed, weighted, and biased, and the output of the output in the S4 layer can be produced through the sigmoid function. The weights of the three features output by the activation layer S4 are determined by the attention mechanism module; based on the weights, the three features output by the activation layer S4 are spliced into a vector; the vector outputs the image belonging to a certain category through the fully connected network Predicted probabilities (i.e. predicted outcomes). Then, based on the prediction results of the sample media information materials and the sample supervision data, the training loss value is determined, and the sub-quality classification model is trained according to the loss value until the number of training times reaches the set number of times, or the training loss value reaches the set value, and the model is stopped. Training to get the trained sub-quality classification model.
可以理解的是，引入多个卷积神经网络和注意力机制，可以使得子质量分类模型的分类效果更加准确。It is understandable that the introduction of multiple convolutional neural networks and attention mechanisms can make the classification effect of the sub-quality classification model more accurate.
进一步的，全连接网络基于预先训练的残差网络进行迁移学习得到。例如，可以基于预先训练的resnet18网络的参数和输出特征对全连接网络进行训练。Further, the fully connected network is obtained by transfer learning based on the pre-trained residual network. For example, a fully connected network can be trained based on the parameters and output features of a pre-trained resnet18 network.
可以理解的是，通过预先训练的残差网络迁移学习得到全连接网络，可以提高子质量分类模型的泛化能力。It is understandable that the fully connected network obtained by transfer learning of the pre-trained residual network can improve the generalization ability of the sub-quality classification model.
在上述实施例的基础上，作为本公开的一种可选方式，根据样本媒介信息物料，构建三元组还可以是，从样本媒介信息物料中提取第一物料集和第二物料集；从第一物料集中抽取正样本物料和负样本物料；从第二物料集中抽取预测样本物料；根据正样本物料、负样本物料和预测样本物料，构建三元组。On the basis of the above embodiment, as an optional method of the present disclosure, constructing a triplet according to the sample media information material may also be: extracting the first material set and the second material set from the sample media information material; The positive sample material and the negative sample material are extracted from the first material set; the predicted sample material is extracted from the second material set; the triplet is constructed according to the positive sample material, the negative sample material and the predicted sample material.
具体的，可以从样本媒介信息物料的每类媒介信息物料中抽取一定数量的媒介信息物料，作为第一物料集，从样本媒介信息物料的每类媒介信息物料中抽取一定数量的媒介信息物料，作为第二物料集；之后从第一物料集的每类媒介信息物料的正样本物料中抽取设定数量的媒介信息物料，作为正样本物料，从第一物料集的每类媒介信息物料的负样本物料中抽取设定数量的媒介信息物料，作为负样本物料；从第二物料集中每类媒介信息物料的正样本物料或负样本物料中随机抽取设定数量的媒介信息物料，作为预测样本物料；进而将正样本物料、负样本物料和预测样本物料，作为三元组。其中，设定数量可以由本领域技术人员根据实际情况设定。Specifically, a certain amount of media information materials can be extracted from each type of media information materials in the sample media information materials, as the first material set, a certain amount of media information materials can be extracted from each type of media information materials in the sample media information materials, As the second material set; then extract a set amount of media information materials from the positive sample materials of each type of media information materials in the first material set, as positive sample materials, from the negative sample materials of each type of media information materials in the first material set A set number of media information materials are selected from the sample materials as negative sample materials; a set number of media information materials are randomly selected from the positive sample materials or negative sample materials of each type of media information materials in the second material set as predicted sample materials ; and then take the positive sample material, the negative sample material and the predicted sample material as a triple. Wherein, the set number can be set by those skilled in the art according to the actual situation.
可以理解的是，从样本媒介信息物料中抽取媒介信息物料来构建三元组，可以确定合适数量的三元组训练样本，避免由于三元组中训练样本过多或过少导致目标质量分类模型不准确的问题。It can be understood that by extracting media information materials from sample media information materials to construct triples, an appropriate number of training samples of triples can be determined, so as to avoid the target quality classification model caused by too many or too few training samples in the triples. inaccurate question.
图2是根据本公开实施例提供的另一种模型训练方法的流程图。在上述实施例的基础上，进一步优化，提供一种可选实施方案。如图2，本实施例提供的模型训练方法可以包括：FIG. 2 is a flowchart of another model training method provided according to an embodiment of the present disclosure. On the basis of the above embodiment, further optimization is provided to provide an optional embodiment. As shown in Figure 2, the model training method provided in this embodiment may include:
S201，对原始媒介信息物料中的原始负物料进行数据扩充，得到扩充负物料。S201, data expansion is performed on the original negative material in the original media information material to obtain the expanded negative material.
其中，原始媒介信息物料为从互联网中获取的媒介信息物料。Among them, the original media information material is the media information material obtained from the Internet.
具体的，对原始媒介信息物料中的原始负物料进行数据扩充，即进行数据增强处理，得到扩充负物料。Specifically, data expansion is performed on the original negative material in the original media information material, that is, data enhancement processing is performed to obtain the expanded negative material.
一种可选方式，可以对原始媒介信息物料中的原始负物料进行颜色变化增强，例如进行噪声、模糊或者颜色变换等操作，得到扩充负物料。In an optional way, the original negative material in the original media information material can be enhanced by color change, for example, by performing operations such as noise, blur, or color transformation, to obtain the expanded negative material.
另一种可选方式，还可以对原始媒介信息物料中的原始负物料进行几何变化增强，例如进行旋转、剪裁处理，得到扩充负物料。In another optional way, the original negative material in the original media information material can also be enhanced by geometric change, such as rotation and clipping processing, to obtain the expanded negative material.
又一种可选方式，还可以对原始媒介信息物料中的原始负物料进行风格迁移，或者通过GAN网络进行处理，得到扩充负物料。In another optional way, the original negative material in the original media information material can also be style-transferred, or processed through the GAN network to obtain the expanded negative material.
可以理解的是，通过对原始负物料进行扩充，避免了正负物料不平衡，导致目标质量分类模型不准确的问题。It is understandable that by expanding the original negative material, the problem of the imbalance between the positive and negative materials and the inaccuracy of the target quality classification model is avoided.
S202，根据扩充负物料、原始负物料和原始媒介信息物料中的原始正物料，确定样本媒介信息物料。S202: Determine the sample media information material according to the original positive material in the expanded negative material, the original negative material, and the original medium information material.
具体的，可以将所有的扩充负物料、原始负物料和原始正物料，作为样本媒介信息物料。Specifically, all extended negative materials, original negative materials, and original positive materials can be used as sample medium information materials.
S203，根据样本媒介信息物料，构建三元组。S203, construct a triplet according to the sample media information material.
其中，三元组包括正样本物料、负样本物料和预测样本物料。Among them, the triplet includes positive sample material, negative sample material and predicted sample material.
S204，采用三元组，对神经网络模型进行训练，得到目标质量分类模型。S204, using triples to train the neural network model to obtain a target quality classification model.
本公开实施例提供的技术方案，通过对原始媒介信息物料中的原始负物料进行数据扩充，得到扩充负物料，之后根据扩充负物料、原始负物料和原始媒介信息物料中的原始正物料，确定样本媒介信息物料。进而根据样本媒介信息物料，构建三元组，并采用三元组，对神经网络模型进行训练，得到目标质量分类模型。上述技术方案，通过对原始负物料进行扩充，避免了真实媒介信息中由于媒介信息物料不平衡导致模型训练不准确的问题，从而提升了目标质量分类模型的准确度。In the technical solution provided by the embodiments of the present disclosure, the expanded negative material is obtained by performing data expansion on the original negative material in the original media information material, and then, according to the expanded negative material, the original negative material and the original positive material in the original media information material, determine Sample media information material. Then, according to the sample media information materials, a triplet is constructed, and the triplet is used to train the neural network model to obtain the target quality classification model. The above technical solution, by expanding the original negative material, avoids the problem of inaccurate model training in real media information due to unbalanced media information material, thereby improving the accuracy of the target quality classification model.
在上述实施例的基础上，作为本公开的一种可选方式，根据扩充负物料、原始负物料和原始媒介信息物料中的原始正物料，确定样本媒介信息物料还可以是，根据扩充负物料与原始负物料之间的相似度，对扩充负物料进行筛选；将筛选后的扩充负物料、原始负物料和原始媒介信息物料中的原始正物料，作为样本媒介信息物料。On the basis of the above-mentioned embodiment, as an optional method of the present disclosure, according to the expanded negative material, the original negative material, and the original positive material in the original media information material, it is determined that the sample media information material may also be, according to the expanded negative material Based on the similarity with the original negative material, the expanded negative material is screened; the filtered expanded negative material, the original negative material and the original positive material in the original media information material are used as the sample medium information material.
具体的，可以结合原始负物料的数量、原始正物料的数量、以及原始负物料和扩充负物料之间的相似度，从扩充负物料中选择一部分负物料。例如，可以根据原始正物料的数量和原始负物料的数量，确定需要选择的扩充负物料的扩充数量，进而根据原始负物料和扩充负物料之间的相似度对扩充负物料进行从小到大排序，将排序在前的扩充数量个扩充负物料保留，且删除剩余扩充负物料；将筛选后的扩充负物料、原始负物料和原始媒介信息物料中的原始正物料，作为样本媒介信息物料。Specifically, a part of negative materials may be selected from the expanded negative materials in combination with the quantity of the original negative materials, the number of the original positive materials, and the similarity between the original negative materials and the expanded negative materials. For example, the expansion quantity of the expanded negative material to be selected can be determined according to the quantity of the original positive material and the quantity of the original negative material, and then the expanded negative material can be sorted from small to large according to the similarity between the original negative material and the expanded negative material , keep the number of expanded negative materials in the first order, and delete the remaining expanded negative materials; take the filtered expanded negative materials, original negative materials, and original positive materials among the original media information materials as sample media information materials.
可以理解的是，引入相似度，对扩充负物料进行筛选，可以剔除掉一些无用的扩充负物料，保证了样本媒介信息物料更加丰富准确。It is understandable that by introducing similarity and screening expanded negative materials, some useless expanded negative materials can be eliminated, which ensures that the sample media information materials are more abundant and accurate.
图3是根据本公开实施例提供的一种媒介信息处理方法的流程图。该方法适用于如何精准确定媒介信息质量的情况。该方法可以由媒介信息处理来执行，该装置可以采用软件和/或硬件的方式实现，并可集成配置有媒介信息处理功能的电子设备中。如图3所示，本实施例提供的媒介信息处理方法可以包括：FIG. 3 is a flowchart of a method for processing media information according to an embodiment of the present disclosure. This method is suitable for how to accurately determine the quality of media information. The method may be performed by media information processing, and the apparatus may be implemented in software and/or hardware, and may be integrated into an electronic device configured with a media information processing function. As shown in FIG. 3 , the media information processing method provided in this embodiment may include:
S301，获取目标媒介信息物料。S301, acquiring target media information materials.
本实施例中，目标媒介信息物料为需要进行质量评估的媒介信息物料。In this embodiment, the target media information material is the media information material that needs to be evaluated for quality.
具体的，可以从业务端获取目标媒介信息物料。Specifically, the target media information material can be obtained from the business end.
S302，采用目标质量分类模型对目标媒介信息物料进行分类。S302, classifying the target media information material by adopting the target quality classification model.
本实施例中，目标质量分类模型通过上述实施例提供的任一的模型训练方法训练得到。In this embodiment, the target quality classification model is obtained by training any of the model training methods provided in the above embodiments.
具体的，将目标媒介信息物料输入至目标质量分类模型中，经过模型分类处理，得到目标媒介信息物料的分类结果。Specifically, the target media information material is input into the target quality classification model, and the classification result of the target media information material is obtained through model classification processing.
S303，根据分类结果，对目标媒介信息物料进行处理。S303, according to the classification result, process the target media information material.
本实施例中，分类结果为目标媒介信息物料为高质物料或者低质物料的概率。In this embodiment, the classification result is the probability that the target media information material is a high-quality material or a low-quality material.
一种可选方式，若根据分类结果确定目标媒介信息物料属于高质物料，则对目标媒介信息物料进行首屏展示。具体的，若分类结果中概率大于设定概率值，则确定目标媒介信息物料属于高质物料，进而在用户的检索端对目标媒介信息物料进行首屏展示。进一步的，还可以从高质物料的目标媒介信息物料中抽取一部分进行人工审核，若人工审核通过，则对人工审核通过的目标媒介信息物料进行首屏展示。In an optional way, if the target media information material is determined to be a high-quality material according to the classification result, the first screen display of the target media information material is performed. Specifically, if the probability in the classification result is greater than the set probability value, it is determined that the target media information material is a high-quality material, and then the target media information material is displayed on the first screen at the user's retrieval end. Further, a part of the target media information materials of the high-quality materials can be extracted for manual review. If the manual review is passed, the first screen display of the target media information materials that have passed the manual review will be displayed.
又一种可选方式，若根据分类结果确定目标媒介信息物料属于低质物料，则对目标媒介信息物料进行滤除。具体的，若分类结果中概率小于设定概率值，则确定目标媒介信息物料属于低质物料，进而将目标媒介信息物料进行滤除。In another optional manner, if the target media information material is determined to be a low-quality material according to the classification result, the target media information material is filtered out. Specifically, if the probability in the classification result is less than the set probability value, it is determined that the target media information material is a low-quality material, and then the target media information material is filtered out.
本公开实施例提供的技术方案，通过获取目标媒介信息物料，之后采用目标质量分类模型对目标媒介信息物料进行分类，进而根据分类结果，对目标媒介信息物料进行处理。上述技术方案，通过目标质量分类模型，可以对媒介信息物料的质量进行精准识别，进而为向用户提供高质量的媒介信息物料提供了保障。In the technical solution provided by the embodiments of the present disclosure, the target media information material is obtained by obtaining the target media information material, and then the target media information material is classified by the target quality classification model, and then the target media information material is processed according to the classification result. The above technical solution can accurately identify the quality of media information materials through the target quality classification model, thereby providing a guarantee for providing users with high-quality media information materials.
图4是根据本公开实施例提供的另一种媒介信息处理方法的流程图。在上述实施例的基础上，对“获取目标媒介信息物料”进一步优化，提供一种可选实施方案。如图4所示，本实施例提供的媒介信息处理方法可以包括：FIG. 4 is a flowchart of another media information processing method provided according to an embodiment of the present disclosure. On the basis of the above-mentioned embodiment, the "obtaining target media information material" is further optimized, and an optional implementation solution is provided. As shown in FIG. 4 , the media information processing method provided in this embodiment may include:
S401，对待处理媒介信息物料进行合规审核。S401, a compliance review is performed on the media information material to be processed.
本实施例中，待处理媒介信息为从业务端获取的原始的媒介信息。In this embodiment, the media information to be processed is the original media information obtained from the service end.
具体的，可以对待处理媒介信息物料进行去重和过滤处理，例如，删除掉重复的待处理媒介信息物料，并根据业务逻辑(待处理媒介信息物料所属行业等)对待处理媒介信息物料进行过滤。Specifically, the media information materials to be processed may be deduplicated and filtered, for example, duplicate media information materials to be processed are deleted, and the media information materials to be processed are filtered according to business logic (the industry to which the media information materials to be processed belong, etc.).
进一步的，还可以对去重和过滤后的待处理媒介信息物料进行格式或者黄反检测，得到合规的待处理媒介信息物料。Further, the deduplicated and filtered media information materials to be processed can also be formatted or reverse-detected to obtain compliant media information materials to be processed.
S402，从合规审核通过的待处理媒介信息物料中获取目标媒介信息物料。S402 , acquiring the target media information material from the to-be-processed media information material that has passed the compliance review.
一种可选方式，可以从合规审核通过的待处理媒介信息物料中随机选择一个待处理媒介信息物料，作为目标媒介信息物料。As an optional method, a to-be-processed media information material may be randomly selected from the to-be-processed media information materials that have passed the compliance review as the target media information material.
又一种可选方式，还可以根据待处理媒介信息物料的所属行业的优先级，从合规审核通过的待处理媒介信息物料中获取目标媒介信息物料。具体的，可以将合规审核通过的待处理媒介信息物料中所属行业优先级高的待处理媒介信息物料，作为目标媒介信息物料。In yet another optional manner, the target media information material may be obtained from the to-be-processed media information material that has passed the compliance review according to the priority of the industry to which the to-be-processed media information material belongs. Specifically, among the media information materials to be processed that have passed the compliance review, the media information materials to be processed that belong to a higher industry priority may be used as the target media information materials.
另一种可选方式，还可以根据待处理媒介信息物料的获取时间的先后顺序，从合规审核通过的待处理媒介信息物料中获取目标媒介信息物料。具体的，可以将合规审核通过的待处理媒介信息物料中获取时间靠前的待处理媒介信息物料，作为目标媒介信息物料。In another optional manner, the target media information material may also be acquired from the pending media information materials that have passed the compliance review according to the order of acquisition time of the to-be-processed media information materials. Specifically, the to-be-processed media information materials that are obtained earlier in the time-to-be-processed media information materials that have passed the compliance review may be used as the target media information materials.
S403，采用目标质量分类模型对目标媒介信息物料进行分类。S403, classifying the target media information material by adopting the target quality classification model.
S404，根据分类结果，对目标媒介信息物料进行处理。S404, according to the classification result, process the target media information material.
本公开实施例提供的技术方案，通过对待处理媒介信息物料进行合规审核，从合规审核通过的待处理媒介信息物料中获取目标媒介信息物料，之后采用目标质量分类模型对目标媒介信息物料进行分类，进而根据分类结果，对目标媒介信息物料进行处理。上述技术方案，对待处理媒介信息物料进行合规处理，能够使得向用户所提供的媒介信息物料的质量更高。According to the technical solution provided by the embodiments of the present disclosure, the target media information material is obtained from the to-be-processed media information material that has passed the compliance review by performing compliance review on the media information material to be processed, and then the target media information material is subjected to a target quality classification model. Classification, and then process the target media information material according to the classification result. In the above technical solution, the media information material to be processed can be processed in compliance with the regulations, so that the quality of the media information material provided to the user is higher.
图5是根据本公开实施例提供的一种模型训练装置的结构示意图。本实施例的技术方案适用于如何精准确定媒介信息质量的情况。该装置可以采用软件和/或硬件的方式实现，并可集成配置有模型训练功能的电子设备中。如图5所示，本实施例的模型训练装置500可以包括：FIG. 5 is a schematic structural diagram of a model training apparatus provided according to an embodiment of the present disclosure. The technical solution of this embodiment is applicable to the situation of how to accurately determine the quality of media information. The apparatus can be implemented in software and/or hardware, and can be integrated into an electronic device configured with a model training function. As shown in FIG. 5 , the
三元组构建模块501，用于根据样本媒介信息物料，构建三元组；其中，三元组包括正样本物料、负样本物料和预测样本物料；The
训练模块502，用于采用三元组，对神经网络模型进行训练，得到目标质量分类模型；神经网络模型包括三个结构相同的子质量分类模型。The
本公开实施例提供的技术方案，通过根据样本媒介信息物料，构建三元组；其中，三元组包括正样本物料、负样本物料和预测样本物料，之后采用三元组，对神经网络模型进行训练，得到目标质量分类模型；神经网络模型包括三个结构相同的子质量分类模型。上述技术方案，引入三元组，对三个结构相同的子质量分类模型进行度量训练，所得到的目标质量分类模型能够对媒介信息物料的质量进行精准识别，进而实现了高效且精准地向用户提供高质量的媒介信息物料。According to the technical solution provided by the embodiments of the present disclosure, a triplet is constructed according to the sample media information materials; wherein, the triplet includes a positive sample material, a negative sample material, and a predicted sample material, and then the triplet is used to perform the neural network model. After training, the target quality classification model is obtained; the neural network model includes three sub-quality classification models with the same structure. In the above technical solution, triples are introduced to perform measurement training on three sub-quality classification models with the same structure, and the obtained target quality classification model can accurately identify the quality of media information materials, thereby realizing efficient and accurate information to users. Provide high-quality media information materials.
进一步地，子质量分类模型包括至少两个不同的卷积神经网络、一个注意力机制模块和一个全连接网络；至少两个不同的卷积神经网络并行连接于注意力机制模块的一端，注意力机制模块的另一端连接于全连接网络。Further, the sub-quality classification model includes at least two different convolutional neural networks, an attention mechanism module and a fully connected network; at least two different convolutional neural networks are connected in parallel to one end of the attention mechanism module, and the attention The other end of the mechanism module is connected to the fully connected network.
进一步地，全连接网络基于预先训练的残差网络进行迁移学习得到。Further, the fully connected network is obtained by transfer learning based on the pre-trained residual network.
进一步地，三元组构建模块501具体用于：Further, the
从样本媒介信息物料中提取第一物料集和第二物料集；extracting the first material set and the second material set from the sample media information material;
从第一物料集中抽取正样本物料和负样本物料；Extract positive sample materials and negative sample materials from the first material set;
从第二物料集中抽取预测样本物料；Extract the forecast sample material from the second material set;
根据正样本物料、负样本物料和预测样本物料，构建三元组。Construct triples based on positive sample material, negative sample material, and predicted sample material.
进一步地，该装置还包括：Further, the device also includes:
扩充负物料确定模块，用于对原始媒介信息物料中的原始负物料进行数据扩充，得到扩充负物料；The expansion negative material determination module is used for data expansion of the original negative material in the original media information material to obtain the expanded negative material;
样本物料确定模块，用于根据扩充负物料、原始负物料和原始媒介信息物料中的原始正物料，确定样本媒介信息物料。The sample material determination module is used to determine the sample medium information material according to the original positive material in the expanded negative material, the original negative material and the original medium information material.
进一步地，样本物料确定模块具体用于：Further, the sample material determination module is specifically used for:
根据扩充负物料与原始负物料之间的相似度，对扩充负物料进行筛选；Screen the expanded negative materials according to the similarity between the expanded negative materials and the original negative materials;
将筛选后的扩充负物料、原始负物料和原始媒介信息物料中的原始正物料，作为样本媒介信息物料。Use the filtered expanded negative material, the original negative material, and the original positive material in the original media information material as the sample medium information material.
进一步地，扩充负物料确定模块具体用于：Further, the expanded negative material determination module is specifically used for:
对原始媒介信息物料中的原始负物料进行旋转、剪裁和风格迁移中的至少一项处理，得到扩充负物料。At least one of rotation, trimming and style transfer is performed on the original negative material in the original media information material to obtain the expanded negative material.
图6是根据本公开实施例提供的一种媒介信息处理装置的结构示意图。本公开的技术方案适用于如何精准确定媒介信息质量的情况。该装置可以采用软件和/或硬件的方式实现，并可集成配置有媒介信息处理功能的电子设备中。如图6所示，本实施例提供的媒介信息处理装置600可以包括：FIG. 6 is a schematic structural diagram of a medium information processing apparatus provided according to an embodiment of the present disclosure. The technical solutions of the present disclosure are applicable to the situation of how to accurately determine the quality of media information. The apparatus can be implemented in software and/or hardware, and can be integrated into an electronic device configured with a media information processing function. As shown in FIG. 6 , the media
目标物料获取模块601，用于获取目标媒介信息物料；The target
分类模块602，用于采用目标质量分类模型对目标媒介信息物料进行分类；其中，目标质量分类模型通过上述实施例提供的模型训练方法训练得到；The
物料处理模块603，用于根据分类结果，对目标媒介信息物料进行处理。The
本公开实施例提供的技术方案，通过获取目标媒介信息物料，之后采用目标质量分类模型对目标媒介信息物料进行分类，进而根据分类结果，对目标媒介信息物料进行处理。上述技术方案，通过目标质量分类模型，可以提高目标媒介信息物料的分类准确度，进而可以精准的对目标媒介信息物料进行处理。In the technical solution provided by the embodiments of the present disclosure, the target media information material is obtained by obtaining the target media information material, and then the target media information material is classified by the target quality classification model, and then the target media information material is processed according to the classification result. The above technical solution, through the target quality classification model, can improve the classification accuracy of the target media information materials, and then can accurately process the target media information materials.
进一步地，目标物料获取模块601具体用于：Further, the target
对待处理媒介信息物料进行合规审核；Conduct compliance audits on the media information materials to be processed;
从合规审核通过的待处理媒介信息物料中获取目标媒介信息物料。Obtain the target media information materials from the pending media information materials that have passed the compliance review.
进一步地，物料处理模块603具体用于：Further, the
若根据分类结果确定目标媒介信息物料属于高质物料，则对目标媒介信息物料进行首屏展示。If it is determined according to the classification result that the target media information material is a high-quality material, the first screen display of the target media information material is performed.
本公开的技术方案中，所涉及的媒介信息物料数据等的获取，存储和应用等，均符合相关法律法规的规定，且不违背公序良俗。In the technical solution of the present disclosure, the acquisition, storage and application of the media information, material data, etc. involved are in compliance with the provisions of relevant laws and regulations, and do not violate public order and good customs.
根据本公开的实施例，本公开还提供了一种电子设备、一种可读存储介质和一种计算机程序产品。According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.
图7是用来实现本公开实施例的数据处理方法和媒介信息处理的电子设备的框图。图7示出了可以用来实施本公开的实施例的示例电子设备700的示意性框图。电子设备旨在表示各种形式的数字计算机，诸如，膝上型计算机、台式计算机、工作台、个人数字助理、服务器、刀片式服务器、大型计算机、和其它适合的计算机。电子设备还可以表示各种形式的移动装置，诸如，个人数字处理、蜂窝电话、智能电话、可穿戴设备和其它类似的计算装置。本文所示的部件、它们的连接和关系、以及它们的功能仅仅作为示例，并且不意在限制本文中描述的和/或者要求的本公开的实现。FIG. 7 is a block diagram of an electronic device for implementing the data processing method and media information processing according to the embodiment of the present disclosure. FIG. 7 shows a schematic block diagram of an example
如图7所示，电子设备700包括计算单元701，其可以根据存储在只读存储器(ROM)702中的计算机程序或者从存储单元708加载到随机访问存储器(RAM)703中的计算机程序，来执行各种适当的动作和处理。在RAM 703中，还可存储电子设备700操作所需的各种程序和数据。计算单元701、ROM 702以及RAM 703通过总线704彼此相连。输入/输出(I/O)接口705也连接至总线704。As shown in FIG. 7 , the
电子设备700中的多个部件连接至I/O接口705，包括：输入单元706，例如键盘、鼠标等；输出单元707，例如各种类型的显示器、扬声器等；存储单元708，例如磁盘、光盘等；以及通信单元709，例如网卡、调制解调器、无线通信收发机等。通信单元709允许电子设备700通过诸如因特网的计算机网络和/或各种电信网络与其他设备交换信息/数据。Various components in the
计算单元701可以是各种具有处理和计算能力的通用和/或专用处理组件。计算单元701的一些示例包括但不限于中央处理单元(CPU)、图形处理单元(GPU)、各种专用的人工智能(AI)计算芯片、各种运行机器学习模型算法的计算单元、数字信号处理器(DSP)、以及任何适当的处理器、控制器、微控制器等。计算单元701执行上文所描述的各个方法和处理，例如模型训练或媒介信息处理方法。例如，在一些实施例中，模型训练或媒介信息处理方法可被实现为计算机软件程序，其被有形地包含于机器可读介质，例如存储单元708。在一些实施例中，计算机程序的部分或者全部可以经由ROM 702和/或通信单元709而被载入和/或安装到电子设备700上。当计算机程序加载到RAM703并由计算单元701执行时，可以执行上文描述的模型训练或媒介信息处理方法的一个或多个步骤。备选地，在其他实施例中，计算单元701可以通过其他任何适当的方式(例如，借助于固件)而被配置为执行模型训练或媒介信息处理方法。
本文中以上描述的系统和技术的各种实施方式可以在数字电子电路系统、集成电路系统、现场可编程门阵列(FPGA)、专用集成电路(ASIC)、专用标准产品(ASSP)、芯片上系统的系统(SOC)、复杂可编程逻辑设备(CPLD)、计算机硬件、固件、软件、和/或它们的组合中实现。这些各种实施方式可以包括：实施在一个或者多个计算机程序中，该一个或者多个计算机程序可在包括至少一个可编程处理器的可编程系统上执行和/或解释，该可编程处理器可以是专用或者通用可编程处理器，可以从存储系统、至少一个输入装置、和至少一个输出装置接收数据和指令，并且将数据和指令传输至该存储系统、该至少一个输入装置、和该至少一个输出装置。Various implementations of the systems and techniques described herein above can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips system (SOC), complex programmable logic device (CPLD), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpretable on a programmable system including at least one programmable processor that The processor, which may be a special purpose or general-purpose programmable processor, may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device an output device.
用于实施本公开的方法的程序代码可以采用一个或多个编程语言的任何组合来编写。这些程序代码可以提供给通用计算机、专用计算机或其他可编程数据处理装置的处理器或控制器，使得程序代码当由处理器或控制器执行时使流程图和/或框图中所规定的功能/操作被实施。程序代码可以完全在机器上执行、部分地在机器上执行，作为独立软件包部分地在机器上执行且部分地在远程机器上执行或完全在远程机器或服务器上执行。Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer or other programmable data processing apparatus, such that the program code, when executed by the processor or controller, performs the functions/functions specified in the flowcharts and/or block diagrams. Action is implemented. The program code may execute entirely on the machine, partly on the machine, partly on the machine and partly on a remote machine as a stand-alone software package or entirely on the remote machine or server.
在本公开的上下文中，机器可读介质可以是有形的介质，其可以包含或存储以供指令执行系统、装置或设备使用或与指令执行系统、装置或设备结合地使用的程序。机器可读介质可以是机器可读信号介质或机器可读储存介质。机器可读介质可以包括但不限于电子的、磁性的、光学的、电磁的、红外的、或半导体系统、装置或设备，或者上述内容的任何合适组合。机器可读存储介质的更具体示例会包括基于一个或多个线的电气连接、便携式计算机盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦除可编程只读存储器(EPROM或快闪存储器)、光纤、便捷式紧凑盘只读存储器(CD-ROM)、光学储存设备、磁储存设备、或上述内容的任何合适组合。In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in connection with the instruction execution system, apparatus or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), fiber optics, compact disk read only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the foregoing.
为了提供与用户的交互，可以在计算机上实施此处描述的系统和技术，该计算机具有：用于向用户显示信息的显示装置(例如，CRT(阴极射线管)或者LCD(液晶显示器)监视器)；以及键盘和指向装置(例如，鼠标或者轨迹球)，用户可以通过该键盘和该指向装置来将输入提供给计算机。其它种类的装置还可以用于提供与用户的交互；例如，提供给用户的反馈可以是任何形式的传感反馈(例如，视觉反馈、听觉反馈、或者触觉反馈)；并且可以用任何形式(包括声输入、语音输入或者、触觉输入)来接收来自用户的输入。To provide interaction with a user, the systems and techniques described herein may be implemented on a computer having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user ); and a keyboard and pointing device (eg, a mouse or trackball) through which a user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (eg, visual feedback, auditory feedback, or tactile feedback); and can be in any form (including acoustic input, voice input, or tactile input) to receive input from the user.
可以将此处描述的系统和技术实施在包括后台部件的计算系统(例如，作为数据服务器)、或者包括中间件部件的计算系统(例如，应用服务器)、或者包括前端部件的计算系统(例如，具有图形用户界面或者网络浏览器的用户计算机，用户可以通过该图形用户界面或者该网络浏览器来与此处描述的系统和技术的实施方式交互)、或者包括这种后台部件、中间件部件、或者前端部件的任何组合的计算系统中。可以通过任何形式或者介质的数字数据通信(例如，通信网络)来将系统的部件相互连接。通信网络的示例包括：局域网(LAN)、广域网(WAN)和互联网。The systems and techniques described herein may be implemented on a computing system that includes back-end components (eg, as a data server), or a computing system that includes middleware components (eg, an application server), or a computing system that includes front-end components (eg, a user's computer having a graphical user interface or web browser through which a user may interact with implementations of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system. The components of the system may be interconnected by any form or medium of digital data communication (eg, a communication network). Examples of communication networks include: Local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.
计算机系统可以包括客户端和服务器。客户端和服务器一般远离彼此并且通常通过通信网络进行交互。通过在相应的计算机上运行并且彼此具有客户端-服务器关系的计算机程序来产生客户端和服务器的关系。服务器可以是云服务器，也可以为分布式系统的服务器，或者是结合了区块链的服务器。A computer system can include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a distributed system server, or a server combined with blockchain.
人工智能是研究使计算机来模拟人的某些思维过程和智能行为(如学习、推理、思考、规划等)的学科，既有硬件层面的技术也有软件层面的技术。人工智能硬件技术一般包括如传感器、专用人工智能芯片、云计算、分布式存储、大数据处理等技术；人工智能软件技术主要包括计算机视觉技术、语音识别技术、自然语言处理技术及机器学习/深度学习技术、大数据处理技术、知识图谱技术等几大方向。Artificial intelligence is the study of making computers to simulate certain thinking processes and intelligent behaviors of people (such as learning, reasoning, thinking, planning, etc.), both hardware-level technology and software-level technology. AI hardware technologies generally include technologies such as sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing; AI software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, and machine learning/depth Learning technology, big data processing technology, knowledge graph technology and other major directions.
云计算(cloud computing)，指的是通过网络接入弹性可扩展的共享物理或虚拟资源池，资源可以包括服务器、操作系统、网络、软件、应用和存储设备等，并可以按需、自服务的方式对资源进行部署和管理的技术体系。通过云计算技术，可以为人工智能、区块链等技术应用、模型训练提供高效强大的数据处理能力。Cloud computing refers to accessing elastically scalable shared physical or virtual resource pools through the network. Resources can include servers, operating systems, networks, software, applications and storage devices, etc., and can be self-service on demand and on demand. A technical system for deploying and managing resources in a way. Through cloud computing technology, it can provide efficient and powerful data processing capabilities for artificial intelligence, blockchain and other technical applications and model training.
应该理解，可以使用上面所示的各种形式的流程，重新排序、增加或删除步骤。例如，本发公开中记载的各步骤可以并行地执行也可以顺序地执行也可以不同的次序执行，只要能够实现本公开公开的技术方案所期望的结果，本文在此不进行限制。It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, the steps described in the present disclosure can be executed in parallel, sequentially, or in different orders. As long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, there is no limitation herein.
上述具体实施方式，并不构成对本公开保护范围的限制。本领域技术人员应该明白的是，根据设计要求和其他因素，可以进行各种修改、组合、子组合和替代。任何在本公开的精神和原则之内所作的修改、等同替换和改进等，均应包含在本公开保护范围之内。The above-mentioned specific embodiments do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present disclosure should be included within the protection scope of the present disclosure.
Claims (23)
Translated fromChinesePriority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210163051.0ACN114707010A (en) | 2022-02-22 | 2022-02-22 | Model training and medium information processing method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210163051.0ACN114707010A (en) | 2022-02-22 | 2022-02-22 | Model training and medium information processing method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114707010Atrue CN114707010A (en) | 2022-07-05 |
Family
ID=82167527
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210163051.0APendingCN114707010A (en) | 2022-02-22 | 2022-02-22 | Model training and medium information processing method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114707010A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020037960A1 (en)* | 2018-08-21 | 2020-02-27 | 深圳大学 | Sar target recognition method and apparatus, computer device, and storage medium |
| CN111783590A (en)* | 2020-06-24 | 2020-10-16 | 西北工业大学 | A Multi-Class Small Object Detection Method Based on Metric Learning |
| CN112734031A (en)* | 2020-12-31 | 2021-04-30 | 珠海格力电器股份有限公司 | Neural network model training method, neural network model recognition method, storage medium, and apparatus |
| CN112860848A (en)* | 2021-01-20 | 2021-05-28 | 平安科技（深圳）有限公司 | Information retrieval method, device, equipment and medium |
| CN112883990A (en)* | 2019-11-29 | 2021-06-01 | 京东安联财产保险有限公司 | Data classification method and device, computer storage medium and electronic equipment |
| CN114022739A (en)* | 2021-11-17 | 2022-02-08 | 南京邮电大学 | A zero-shot learning method based on the combination of aligned variational autoencoder and triples |
- 2022
- 2022-02-22CNCN202210163051.0Apatent/CN114707010A/enactivePending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020037960A1 (en)* | 2018-08-21 | 2020-02-27 | 深圳大学 | Sar target recognition method and apparatus, computer device, and storage medium |
| CN112883990A (en)* | 2019-11-29 | 2021-06-01 | 京东安联财产保险有限公司 | Data classification method and device, computer storage medium and electronic equipment |
| CN111783590A (en)* | 2020-06-24 | 2020-10-16 | 西北工业大学 | A Multi-Class Small Object Detection Method Based on Metric Learning |
| CN112734031A (en)* | 2020-12-31 | 2021-04-30 | 珠海格力电器股份有限公司 | Neural network model training method, neural network model recognition method, storage medium, and apparatus |
| CN112860848A (en)* | 2021-01-20 | 2021-05-28 | 平安科技（深圳）有限公司 | Information retrieval method, device, equipment and medium |
| CN114022739A (en)* | 2021-11-17 | 2022-02-08 | 南京邮电大学 | A zero-shot learning method based on the combination of aligned variational autoencoder and triples |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113326764B (en) | Method and device for training image recognition model and image recognition | |
| CN113361578B (en) | Training method and device for image processing model, electronic equipment and storage medium | |
| CN114648676B (en) | Point cloud processing model training and point cloud instance segmentation method and device | |
| CN114612749B (en) | Neural network model training method and device, electronic device and medium | |
| JP7357114B2 (en) | Training method, device, electronic device and storage medium for living body detection model | |
| CN114490998B (en) | Text information extraction method and device, electronic equipment and storage medium | |
| CN115082740B (en) | Target detection model training method, target detection device and electronic equipment | |
| CN113051911B (en) | Methods, devices, equipment, media and program products for extracting sensitive words | |
| US20220374678A1 (en) | Method for determining pre-training model, electronic device and storage medium | |
| WO2023207028A1 (en) | Image retrieval method and apparatus, and computer program product | |
| CN112580733A (en) | Method, device and equipment for training classification model and storage medium | |
| CN113656587A (en) | Text classification method and device, electronic equipment and storage medium | |
| CN114494776A (en) | A model training method, device, equipment and storage medium | |
| CN115248890A (en) | Method, device, electronic device and storage medium for generating user interest portrait | |
| CN115511779A (en) | Image detection method, device, electronic device and storage medium | |
| CN114090601A (en) | A data screening method, apparatus, device and storage medium | |
| CN113706705B (en) | Image processing method, device, equipment and storage medium for high-precision map | |
| CN113989569B (en) | Image processing method, device, electronic equipment and storage medium | |
| CN114973333A (en) | Human interaction detection method, device, device and storage medium | |
| CN114842541A (en) | Model training and face recognition method, device, equipment and storage medium | |
| CN114330576A (en) | Model processing method, device, image recognition method and device | |
| CN114281990A (en) | Document classification method and device, electronic equipment and medium | |
| CN117576376A (en) | Object detection methods and devices, electronic equipment | |
| CN114580448B (en) | A sign language interpretation method, device, electronic device and storage medium | |
| CN116226353A (en) | Image-text retrieval model training method, image-text retrieval 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 |