CN107590741A - A kind of method and system of predicted pictures popularity - Google Patents

Abstract

The present application discloses a method for predicting the popularity of pictures. The method includes: preprocessing the received source data to obtain feature samples; performing random inactivation processing on feature samples according to a preset ratio to obtain the first input feature; A regression model performs a prediction operation on the input features to obtain an intermediate prediction result, and combines the intermediate prediction result with the first input feature to obtain a second input feature; judge whether the depth stacking regression model converges according to the second input feature; if so, then Generate the corresponding picture popularity according to the intermediate prediction result; if not, combine the intermediate prediction result with the second input feature to obtain the third input feature, and use the third input result as a new feature sample; this method can accurately predict the picture popularity; the application also discloses a system for predicting the popularity of pictures, a computer-readable storage medium and a server, which have the above beneficial effects.

Description

A method and system for predicting the popularity of pictures

technical field

The invention relates to the field of data analysis, in particular to a method and system for predicting picture popularity, a computer-readable storage medium and a server.

Background technique

The rapid development of information technology has promoted the popularity of social media, which has changed the way people interact. Users mainly share their lives and experiences on social media platforms by sending pictures. Therefore, social media has accumulated massive image data. However, the popularity of these images varies. The popularity of pictures sent by users with different popularity is very different, and the popularity of pictures sent by the same user is also different. Applications in many fields, such as the design of news personalized recommendation systems, the placement of online advertisements, etc., all benefit from the research on the popularity prediction of social media pictures.

In the prior art, social network message outbreak detection based on recurrent neural network is to classify and predict historical messages published and forwarded by users in social networks, and to judge whether a message has broken out. This prior art only involves the prediction of text information on social media, and cannot accurately predict the popularity of social pictures.

Therefore, how to accurately predict the popularity of social pictures is a technical problem to be solved by those skilled in the art.

Contents of the invention

The purpose of the present application is to provide a method and system for predicting the popularity of pictures, a computer-readable storage medium and a server, which can accurately predict the popularity of social pictures.

In order to solve the above technical problems, the present application provides a method for predicting the popularity of pictures, the method comprising:

Step 1: Preprocessing the received source data to obtain feature samples; wherein, the feature samples include visual features and social features;

Step 2: Perform random inactivation processing on the feature samples according to a preset ratio to obtain the first input feature;

Step 3: Using several preset regression models to perform prediction operations on the input features to obtain intermediate prediction results, and combine the intermediate prediction results with the first input features to obtain second input features;

Step 4: According to the second input feature, it is judged whether the depth stacking regression model is converged; if not, the intermediate prediction result is combined with the second input feature to obtain a third input feature, and the third input result is Enter step 2 as a new feature sample; if so, enter step 5;

Step 5: Generate the corresponding picture popularity according to the intermediate prediction result.

Optionally, said preprocessing the received source data to obtain feature samples includes:

dividing the source data into image data and social data;

performing feature extraction on the image data using an artificial neural network to obtain the visual features;

Using multitemporal scale and Z-score (standard score) standardization to convert the social data to obtain the social features;

The feature samples are obtained by splicing the visual features and the social features according to preset rules.

Optionally, performing feature extraction on the image data using an artificial neural network to obtain the visual features includes:

performing two-level feature extraction on the image data using an artificial neural network to obtain low-level features and high-level features;

Combining the low-level features and the high-level features to obtain the visual features.

Optionally, performing two-level feature extraction on the image data using an artificial neural network to obtain low-level features and high-level features includes:

ResNeXt model, Xception model and DenseNet model are obtained by training on ImageNet data set and Place365 data set; Wherein, described ResNeXt model, Xception model and DenseNet model are artificial neural networks;

The picture data is input into the ResNeXt model, Xception model and DenseNet model, and the value of the feature map before the last layer of the ResNeXt model, Xception model and DenseNet model is extracted and obtained through feature compression and feature selection. low-level features;

The high-level features are obtained by connecting the scene information and category information predicted by the ResNeXt model, Xception model and DenseNet model to the picture.

The present application also provides a system for predicting the popularity of pictures, the system comprising:

A preprocessing module, configured to preprocess the received source data to obtain feature samples; wherein the feature samples include visual features and social features;

The Dropout module is used to perform random inactivation processing on the feature samples according to a preset ratio to obtain the first input feature;

A Block module, configured to perform a prediction operation on the first input feature using several preset regression models to obtain an intermediate prediction result, and combine the intermediate prediction result with the input feature to obtain a second input feature;

The Detector module is used to judge whether the depth stacking regression model converges according to the second input feature; if not, combine the intermediate prediction result with the second input feature to obtain a third input feature, and use the third The input result is entered into the next layer stacked regression model as the new feature sample; if yes, the corresponding picture popularity is generated according to the intermediate prediction result.

Optionally, the preprocessing module includes:

Classification sub-module for dividing the source data into picture data and social data;

The visual feature extraction sub-module is used to perform feature extraction on the image data using an artificial neural network to obtain the visual features;

The social feature extraction submodule is used to convert the social data to obtain the social feature by using multi-temporal scale and Z-score standardization;

The splicing sub-module is used to splice the visual features and the social features according to preset rules to obtain the feature samples.

Optionally, the visual feature extraction submodule includes:

A two-level feature extraction unit is used to extract low-level features and high-level features from the image data using an artificial neural network for two-level feature extraction;

A combining unit, configured to combine the low-level features and the high-level features to obtain the visual features.

Optionally, the two-stage feature extraction includes:

The model training subunit is used to train the ResNeXt model, the Xception model and the DenseNet model on the ImageNet data set and the Place365 data set; wherein, the ResNeXt model, the Xception model and the DenseNet model are all artificial neural networks;

The low-level feature extraction subunit is used to input the picture data into the ResNeXt model, Xception model and DenseNet model, and extract the value of the feature map before the last layer of the ResNeXt model, Xception model and DenseNet model and Obtaining the low-level features through feature compression and feature selection;

The advanced feature extraction subunit is used to connect the scene information and category information predicted by the ResNeXt model, Xception model and DenseNet model to obtain the advanced features.

The present application also provides a computer-readable storage medium, on which a computer program is stored, and the following steps are implemented when the computer program is executed:

Step 1: Preprocessing the received source data to obtain feature samples; wherein, the feature samples include visual features and social features;

Step 2: Perform random inactivation processing on the feature samples according to a preset ratio to obtain the first input feature;

Step 3: Using several preset regression models to perform prediction operations on the input features to obtain intermediate prediction results, and combine the intermediate prediction results with the first input features to obtain second input features;

Step 4: According to the second input feature, it is judged whether the depth stacking regression model is converged; if not, the intermediate prediction result is combined with the second input feature to obtain a third input feature, and the third input result is Enter step 2 as a new feature sample; if so, enter step 5;

Step 5: Generate the corresponding picture popularity according to the intermediate prediction result.

The present application also provides a server, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the following steps when invoking the computer program in the memory:

Step 1: Preprocessing the received source data to obtain feature samples; wherein, the feature samples include visual features and social features;

Step 2: Perform random inactivation processing on the feature samples according to a preset ratio to obtain the first input feature;

Step 3: Using several preset regression models to perform prediction operations on the input features to obtain intermediate prediction results, and combine the intermediate prediction results with the first input features to obtain second input features;

Step 4: According to the second input feature, it is judged whether the depth stacking regression model is converged; if not, the intermediate prediction result is combined with the second input feature to obtain a third input feature, and the third input result is Enter step 2 as a new feature sample; if so, enter step 5;

Step 5: Generate the corresponding picture popularity according to the intermediate prediction result.

The present invention provides a method for predicting the popularity of a picture. The received source data is preprocessed to obtain feature samples; wherein, the feature samples include visual features and social features; the feature samples are randomized according to a preset ratio. Live processing to obtain the first input feature; using several preset regression models to perform prediction operations on the input feature to obtain an intermediate prediction result, and combine the intermediate prediction result with the first input feature to obtain the second input feature ; Judging whether the depth stacking regression model converges according to the second input feature; if not, combining the intermediate prediction result with the second input feature to obtain a third input feature, and using the third input result as a new the feature samples; if yes, generate the corresponding picture popularity according to the intermediate prediction result.

In this method, the source data is preprocessed to obtain identifiable feature samples, and the random inactivation processing of the feature samples according to a preset ratio can increase the diversity of feature samples and make the picture popularity prediction more accurate. Use the regression model to predict the first input feature of the input to obtain the predicted value, and judge whether the deep stacked regression model converges. If it does not converge, repeat the operations of random deactivation processing, regression model prediction, and convergence judgment until the deep regression model converges. Get image popularity. Since there are multiple regression models, multiple predicted values will be obtained, and the average of the above predicted values is the predicted popularity of the picture. The method can accurately predict the popularity of pictures through the stacking of multi-layer regression models, which is conducive to the development of new media; the application also provides a system for predicting the popularity of pictures, a computer-readable storage medium and a server, which have The above beneficial effects will not be repeated here.

Description of drawings

In order to illustrate the embodiments of the present application more clearly, the following will briefly introduce the accompanying drawings used in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. As far as people are concerned, other drawings can also be obtained based on these drawings on the premise of not paying creative work.

FIG. 1 is a flow chart of a method for predicting picture popularity provided by an embodiment of the present application;

FIG. 2 is a flow chart of another method for predicting picture popularity provided by an embodiment of the present application;

FIG. 3 is a flow chart of another method for predicting picture popularity provided by an embodiment of the present application;

FIG. 4 is a schematic structural diagram of a system for predicting picture popularity provided by an embodiment of the present application.

detailed description

In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments It is a part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the scope of protection of this application.

Please refer to FIG. 1 below. FIG. 1 is a flowchart of a method for predicting picture popularity provided by an embodiment of the present application;

Specific steps can include:

Step S101: Preprocessing the received source data to obtain feature samples; wherein, the feature samples include visual features and social features;

Among them, the purpose of this program is to predict the popularity of the picture, so the source data received in this step is the relevant information of the picture, and the source of the source data can be social platforms such as Flickr, Weibo, QQ space or picture sharing websites. There is no restriction on the source of the source data, as long as it is a public, image-related website. It can be understood that, since the purpose of the present invention is to predict the popularity of pictures, the acquired source data may be pictures with a high degree of attention, or pictures with little attention. However, due to the requirements of legal ethics, some source data with bad information (such as pornography, violence, and political reaction) can be eliminated during preprocessing without popularity prediction.

It is understandable that the source data contains not only the visual information of the picture (the information carried by the content of the picture), but also social information (such as the release time of the picture, the number of comments received by the picture, whether the picture has tags, the number of tags, The length of the picture title, the average number of views of the user who posted the picture, the number of users who joined the group, etc.). Therefore, correspondingly, when preprocessing the source data, it is necessary to preprocess the visual information of the picture, and it is also necessary to preprocess the information of the social media level of the picture. Since it is different from the processed object, the way of preprocessing it is also different. For the visual information of pictures, deep learning techniques (such as ResNeXt, Xception or DenseNet) can be used to extract the visual information in the source data to obtain visual features. Furthermore, autoencoders and random forests can be used for dimensionality reduction and feature selection to obtain visual features with fewer dimensions. Preferably, the time information in the social information can be used to divide the release date of the picture in multiple dimensions, for example, it can be divided into quarters, months, hours and so on.

Of course, there are many operations that can be performed on the source data to obtain feature samples. In addition to the above-mentioned related operations, it can include screening of duplicate source data (that is, multiple identical photos); it can also include source data by image type Carry out classification, such as figure pictures, landscape pictures, animal pictures, etc., that is, first distinguish the picture types and then perform subsequent popularity prediction operations. The above-mentioned classification of pictures exists only as a preferred embodiment, and those skilled in the art can make a comprehensive choice according to the specific scene where the solution is applied.

Step S102: Randomly inactivate the feature samples according to a preset ratio to obtain the first input feature;

Among them, the purpose of this step is to increase the diversity of feature samples. Since this solution realizes the prediction of image popularity by stacking multi-layer regression models, as the complexity of the regression model increases, the prediction of popularity is often limited by overfitting. In order to reduce the limitation caused by overfitting, this scheme performs random deactivation on all feature samples before predicting the popularity of images. For example, assuming that the input feature vector is 4-D and the deactivation ratio is 0.5, a 2-D feature vector is obtained for each base model; assuming there are 4 base models in the block module, based on this deactivation ratio , then the feature size of the input of the next block is 8-D.

It can be understood that the preset ratio in this step is obtained by those skilled in the art after conducting a large number of experiments and demonstrations, and can be selected relatively accurately according to the specific conditions of the source data to improve the accuracy and accuracy of the popularity prediction. efficiency. Generally speaking, when the diversity of feature samples (ie, source data) is insufficient, the preset ratio of random inactivation processing can be appropriately increased; while in the case of relatively high diversity of feature samples, the ratio of random inactivation processing can be appropriately reduced. preset ratio. In a word, the specific value of the preset ratio is not limited here, and there are different preset ratios for different application environments.

Step S103: Using several preset regression models to perform prediction operations on the input features to obtain intermediate prediction results, and combine the intermediate prediction results with the first input features to obtain second input features;

Wherein, the purpose of this step is to predict the input features obtained in step S102 to obtain a preliminary predicted value. There can be many kinds of regression models used in this step, including: Random Forest (RF), Extreme Random Tree (EXRT), XGBoost or Lasso, etc. Of course, there can also be other regression models, and those skilled in the art can make their own Therefore, the specific type of the regression model used in this step is not limited here.

It is understandable that the number of presets mentioned in this step can be 1 or 2 or more, but for the sake of increasing the diversity of output features, using only one regression model will result in low diversity adverse effects, so use as many different types of regression models as possible to increase the diversity of output features. For example, four regression models, Random Forest (RF), Extremely Random Tree (EXRT), XGBoost and Lasso, can be used to simultaneously predict the input features obtained in step S102 to obtain preliminary predicted values. Wherein, since the structures and principles of each type of regression model are different, the prediction values calculated by each regression model are also different, and all the prediction values can be averaged to obtain an intermediate prediction result.

As can be seen from the previous discussion, this solution realizes picture prediction by stacking multi-layer regression models, so in this step, the step of combining the intermediate prediction result with the first input feature to obtain the second input feature is for judging Whether it is necessary to enter the next layer of regression model to predict again.

Step S104: judging whether the deep stack regression model converges according to the second input feature; if not, proceed to step S105; if yes, proceed to step S106;

Wherein, the purpose of this step is to judge whether the deep stack regression model is converged or not based on the second input feature obtained in step S103. Depth (that is, the number of times the regression model is stacked) is very important for the stacked model. The greater the depth within a certain range, the more accurate the prediction of the popularity of the picture is, but blindly increasing the stacking depth when reaching a certain critical value will not increase The accuracy of the forecast will cause a waste of resources. Therefore, this step analyzes whether the stacking depth needs to be increased by judging whether the deep stacking regression model converges.

If the deep stacking model does not converge, you need to continue stacking; if the deep stacking model converges, it means that the prediction of the image popularity has ended. To continue stacking is to use the third input feature obtained in step S105 to repeat the process of step S102, step S103, and step S104 until the convergence of the deep stacking model is judged.

Step S105: Combine the intermediate prediction result with the second input feature to obtain a third input feature, and use the third input result as a new feature sample to enter step S102.

Step S106: Generate the corresponding picture popularity according to the intermediate prediction result.

Please refer to FIG. 2 below. FIG. 2 is a flow chart of another method for predicting the popularity of a picture provided by an embodiment of the present application.

This embodiment is a specific limitation on how to preprocess the source data to obtain feature samples in step S101 in the previous embodiment. Other steps are generally the same as the previous embodiment. For the same parts, please refer to the previous implementation. The relevant part of the example will not be repeated here.

Specific steps can include:

Step S201: dividing the source data into image data and social data;

Among them, the picture data mentioned in this step is the visual information of the picture (the information carried by the picture content), and the social data is the information at the social media level (such as the release time of the picture, the number of comments obtained by the picture, whether the picture has tags, The number of tags, the length of the picture title, the average page views of users who post pictures, the number of users joining groups, etc.).

Step S202: Using an artificial neural network to perform feature extraction on the image data to obtain the visual features;

Step S203: transforming the social data using multitemporal scale and Z-score standardization to obtain the social features;

Step S204: concatenate the visual features and the social features according to preset rules to obtain the feature samples.

Please refer to FIG. 3 below. FIG. 3 is a flow chart of another method for predicting picture popularity provided by an embodiment of the present application.

This embodiment is a specific limitation on how to use the artificial neural network to perform feature extraction on the image data in step S202 in the previous embodiment to obtain the visual features. Other steps are substantially the same as in the previous embodiment. For the same parts, reference may be made to relevant parts of the previous embodiment, and details are not repeated here.

Step S301: ResNeXt model, Xception model and DenseNet model are obtained by training on ImageNet dataset and Place365 dataset; wherein, the ResNeXt model, Xception model and DenseNet model are all artificial neural networks;

Step S302: Input the picture data into the ResNeXt model, Xception model and DenseNet model, and extract the value of the feature map before the last layer of the ResNeXt model, Xception model and DenseNet model and perform feature compression and feature extraction. select to obtain said low-level features;

Step S303: connect the scene information and category information predicted by the ResNeXt model, Xception model and DenseNet model to obtain the high-level features.

Step S304: Combine the low-level features and the high-level features to obtain the visual features.

Please refer to FIG. 4, which is a schematic structural diagram of a system for predicting picture popularity provided by an embodiment of the present application;

The system can include:

A preprocessing module 100, configured to preprocess the received source data to obtain feature samples; wherein, the feature samples include visual features and social features;

The Dropout module 200 is configured to perform random inactivation processing on the feature samples according to a preset ratio to obtain the first input feature;

Block module 300, configured to use several preset regression models to perform prediction operations on the first input features to obtain intermediate prediction results, and combine the intermediate prediction results with the input features to obtain second input features;

The Detector module 400 is used to judge whether the depth stacking regression model converges according to the second input feature; if not, combine the intermediate prediction result with the second input feature to obtain a third input feature, and use the second input feature to obtain a third input feature. The three input results are entered into the next layer stacked regression model as the new feature samples, and if so, the corresponding picture popularity is generated according to the intermediate prediction results.

Among them, in Figure 4, only the modules of the first-layer stacked regression model are marked, and each layer of the stacked regression model has a Dropout module, a Block module, and a Detector module.

In another embodiment of the system for predicting picture popularity provided by the present application, the preprocessing module 100 includes:

Classification sub-module for dividing the source data into picture data and social data;

The visual feature extraction sub-module is used to perform feature extraction on the image data using an artificial neural network to obtain the visual features;

The social feature extraction submodule is used to convert the social data to obtain the social feature by using multi-temporal scale and Z-score standardization;

The splicing sub-module is used to splice the visual features and the social features according to preset rules to obtain the feature samples.

Further, the visual feature extraction submodule includes:

A two-level feature extraction unit is used to extract low-level features and high-level features from the image data using an artificial neural network for two-level feature extraction;

A combining unit, configured to combine the low-level features and the high-level features to obtain the visual features.

Further, the two-level feature extraction includes:

The model training subunit is used to train the ResNeXt model, the Xception model and the DenseNet model on the ImageNet data set and the Place365 data set; wherein, the ResNeXt model, the Xception model and the DenseNet model are all artificial neural networks;

The low-level feature extraction subunit is used to input the picture data into the ResNeXt model, Xception model and DenseNet model, and extract the value of the feature map before the last layer of the ResNeXt model, Xception model and DenseNet model and Obtaining the low-level features through feature compression and feature selection;

The advanced feature extraction subunit is used to connect the scene information and category information predicted by the ResNeXt model, Xception model and DenseNet model to obtain the advanced features.

Since the embodiments of the system part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the system part, and details will not be repeated here.

The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed, the steps provided in the above-mentioned embodiments can be realized. The storage medium may include various media capable of storing program codes such as a U disk, a removable hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk.

The present application also provides a server, which may include a memory and a processor, where a computer program is stored in the memory, and when the processor invokes a computer layer sequence in the memory, the steps provided in the above embodiments can be implemented. Of course, the server may also include various network interfaces, power supplies and other components.

Each embodiment in the description is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other. As for the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for the related information, please refer to the description of the method part. It should be pointed out that those skilled in the art can make some improvements and modifications to the application without departing from the principles of the application, and these improvements and modifications also fall within the protection scope of the claims of the application.

It should also be noted that in this specification, relative terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that these entities or operations There is no such actual relationship or order between the operations. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

Claims (10)

1. A method for predicting picture popularity, comprising: Step 1: Preprocessing the received source data to obtain feature samples; wherein, the feature samples include visual features and social features; Step 2: Perform random inactivation processing on the feature samples according to a preset ratio to obtain the first input feature; Step 3: Using several preset regression models to perform prediction operations on the input features to obtain intermediate prediction results, and combine the intermediate prediction results with the first input features to obtain second input features; Step 4: According to the second input feature, it is judged whether the depth stacking regression model is converged; if not, the intermediate prediction result is combined with the second input feature to obtain a third input feature, and the third input result is Enter step 2 as a new feature sample; if so, enter step 5; Step 5: Generate the corresponding picture popularity according to the intermediate prediction result. 2. The method according to claim 1, wherein said obtaining the feature samples by preprocessing the received source data comprises: dividing the source data into image data and social data; performing feature extraction on the image data using an artificial neural network to obtain the visual features; converting the social data to obtain the social features by using multitemporal scale and Z-score standardization; The feature samples are obtained by splicing the visual features and the social features according to preset rules. 3. The method according to claim 2, wherein said utilizing artificial neural network to perform feature extraction on said picture data to obtain said visual features comprises: performing two-level feature extraction on the image data using an artificial neural network to obtain low-level features and high-level features; Combining the low-level features and the high-level features to obtain the visual features. 4. method according to claim 3, is characterized in that, described utilizing artificial neural network to carry out two-level feature extraction to described picture data and obtain low-level feature and high-level feature comprising: ResNeXt model, Xception model and DenseNet model are obtained by training on ImageNet data set and Place365 data set; Wherein, described ResNeXt model, Xception model and DenseNet model are artificial neural networks; The picture data is input into the ResNeXt model, Xception model and DenseNet model, and the value of the feature map before the last layer of the ResNeXt model, Xception model and DenseNet model is extracted and obtained through feature compression and feature selection. low-level features; The high-level features are obtained by connecting the scene information and category information predicted by the ResNeXt model, Xception model and DenseNet model to the picture. 5. A system for predicting picture popularity, comprising: A preprocessing module, configured to preprocess the received source data to obtain feature samples; wherein the feature samples include visual features and social features; The Dropout module is used to perform random inactivation processing on the feature samples according to a preset ratio to obtain the first input feature; A Block module, configured to perform a prediction operation on the first input feature using several preset regression models to obtain an intermediate prediction result, and combine the intermediate prediction result with the input feature to obtain a second input feature; The Detector module is used to judge whether the depth stacking regression model converges according to the second input feature; if not, combine the intermediate prediction result with the second input feature to obtain a third input feature, and use the third The input result is entered into the next layer stacked regression model as the new feature sample; if yes, the corresponding picture popularity is generated according to the intermediate prediction result. 6. The system according to claim 5, wherein the preprocessing module comprises: Classification sub-module for dividing the source data into picture data and social data; The visual feature extraction sub-module is used to perform feature extraction on the image data using an artificial neural network to obtain the visual features; The social feature extraction submodule is used to convert the social data to obtain the social feature by using multi-temporal scale and Z-score standardization; The splicing sub-module is used to splice the visual features and the social features according to preset rules to obtain the feature samples. 7. system according to claim 6, is characterized in that, described visual feature extraction submodule comprises: A two-level feature extraction unit is used to extract low-level features and high-level features from the image data using an artificial neural network for two-level feature extraction; A combining unit, configured to combine the low-level features and the high-level features to obtain the visual features. 8. The system according to claim 7, wherein the two-stage feature extraction comprises: The model training subunit is used to train the ResNeXt model, the Xception model and the DenseNet model on the ImageNet data set and the Place365 data set; wherein, the ResNeXt model, the Xception model and the DenseNet model are all artificial neural networks; The low-level feature extraction subunit is used to input the picture data into the ResNeXt model, Xception model and DenseNet model, and extract the value of the feature map before the last layer of the ResNeXt model, Xception model and DenseNet model and Obtaining the low-level features through feature compression and feature selection; The advanced feature extraction subunit is used to connect the scene information and category information predicted by the ResNeXt model, Xception model and DenseNet model to obtain the advanced features. 9. A computer-readable storage medium, on which a computer program is stored, wherein the method according to any one of claims 1 to 4 is implemented when the computer program is executed. 10. A server, characterized in that it comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor invokes the computer program in the memory, it realizes any one of claims 1 to 4 Methods.