User action trajectory prediction method, device and storage medium based on social network

Abstract

The present invention discloses a method, device and storage medium for predicting user action trajectory based on social network, the method comprising: obtaining dynamic information posted by users in social network, and extracting text features and picture features from the dynamic information; analyzing the correlation between the text features and the picture features, and obtaining a condition set according to the correlation; training a user action trajectory prediction model according to the condition set to obtain a trained user action trajectory prediction model; wherein the user action trajectory prediction model comprises a generator and a discriminator; inputting text features and picture features extracted from dynamic information posted by users in social network into the trained user action trajectory prediction model to obtain a user action trajectory prediction result. The present invention can effectively improve the prediction accuracy of user action trajectory by using a generative adversarial network to train the features extracted from dynamic information posted by users in social network.

Description

Technical Field

[0001] The present invention relates to the technical field of crowd flow prediction, and in particular to a method, device and storage medium for predicting user movement trajectory based on a social network. Background Art

[0002] As a branch of traffic forecasting and spatiotemporal data mining, crowd flow prediction plays a key role in traffic regulation and prediction. Deploying a crowd flow monitoring and early warning platform in the city based on the prediction results can effectively improve the city's operating efficiency and more effectively protect the city's public safety. In addition, when an epidemic occurs, the flow of people may cause the virus to spread wildly, seriously endangering people's lives and health. Through the historical trends of crowd flow and the predicted future trends, relevant departments have the ability to estimate hidden infection groups and the number of potential infected people and future high-incidence areas of infection.

[0003] Existing technologies have systematically analyzed the spatial mobility patterns of the entire Chinese population and the decisive factors that determine the population mobility in various regions of China based on open source map data sets. However, the above methods ignore the credibility of the historical trajectory of the crowd. Since human movements are much more flexible than vehicles and other means of transportation, their variability also fully demonstrates that the population trajectory data without data evaluation must have many errors or offsets, resulting in strong noise in machine learning and low prediction accuracy. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method, device and storage medium for predicting user action trajectories based on social networks, which uses a generative adversarial network to train features extracted from dynamic information posted by users in social networks, thereby effectively improving the prediction accuracy of user action trajectories.

[0005] In order to achieve the above object, an embodiment of the present invention provides a method for predicting user action trajectories based on a social network, comprising:

[0006] Obtain dynamic information posted by users in social networks, and extract text features and image features from the dynamic information;

[0007] Analyzing the correlation between the text feature and the image feature, and obtaining a condition set according to the correlation;

[0008] The user action trajectory prediction model is trained according to the condition set to obtain a trained user action trajectory prediction model; wherein the user action trajectory prediction model includes a generator and a discriminator;

[0009] The text features and image features extracted from the dynamic information posted by users in the social network are input into the trained user action trajectory prediction model to obtain the user action trajectory prediction result.

[0010] As an improvement of the above solution, the step of extracting text features and picture features from the dynamic information specifically includes:

[0011] Extracting text features from the dynamic information through an Encoder-Decoder model;

[0012] Image features are extracted from the dynamic information through a convolutional neural network.

[0013] As an improvement of the above solution, the analyzing the correlation between the text feature and the image feature and obtaining a condition set according to the correlation specifically includes:

[0014] Analyze the correlation between the text features and the image features through the Attention mechanism;

[0015] The text features and picture features corresponding to the correlation degree being greater than a preset value are used as the condition set.

[0016] As an improvement of the above solution, the method further includes:

[0017] Generate the user location when the user posts the dynamic information according to the correlation between the text feature and the image feature through the Encoder-Decoder model;

[0018] Determine whether the generated user position deviates from the actual position, and correct the user position with the deviate.

[0019] As an improvement of the above solution, the user action trajectory prediction model is trained according to the condition set to obtain a trained user action trajectory prediction model, specifically including:

[0020] Inputting the condition set and the noise vector into a generator in the user action trajectory prediction model, and having the generator predict the user's future action trajectory;

[0021] The discriminator analyzes the error between the action trajectory predicted by the generator in a noise interference environment and the actual action trajectory;

[0022] The generator and the discriminator are repeatedly trained according to the error until the error meets a preset condition, thereby obtaining a trained user action trajectory prediction model.

[0023] As an improvement of the above solution, the network of the generator includes two layers: the first layer is a network layer that only uses the condition set for prediction; the second layer is a network layer that uses both the condition set and the noise vector for prediction.

[0024] The embodiment of the present invention further provides a user action trajectory prediction device based on a social network, comprising:

[0025] An acquisition module is used to acquire dynamic information posted by users in social networks and extract text features and image features from the dynamic information;

[0026] An analysis module, used for analyzing the correlation between the text features and the image features, and obtaining a condition set according to the correlation;

[0027] A training module, used to train the user action trajectory prediction model according to the condition set to obtain a trained user action trajectory prediction model; wherein the user action trajectory prediction model includes a generator and a discriminator;

[0028] The prediction module is used to input the text features and image features extracted from the dynamic information posted by users in the social network into the trained user action trajectory prediction model to obtain the user action trajectory prediction result.

[0029] Furthermore, the acquisition module extracts text features and image features from the dynamic information, specifically including:

[0030] Extracting text features from the dynamic information through an Encoder-Decoder model;

[0031] Image features are extracted from the dynamic information through a convolutional neural network.

[0032] An embodiment of the present invention also provides a user action trajectory prediction device based on a social network, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements any of the above-mentioned user action trajectory prediction methods based on a social network.

[0033] An embodiment of the present invention further provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any of the above-mentioned methods for predicting user action trajectories based on social networks.

[0034] Compared with the prior art, the user action trajectory prediction method, device and storage medium provided by the embodiment of the present invention based on social network have the following beneficial effects: by obtaining dynamic information posted by users in social network, and extracting text features and picture features from the dynamic information; analyzing the correlation between the text features and the picture features, and obtaining a condition set according to the correlation; training the user action trajectory prediction model according to the condition set to obtain a trained user action trajectory prediction model; wherein the user action trajectory prediction model includes a generator and a discriminator; inputting the text features and picture features extracted from the dynamic information posted by users in social network into the trained user action trajectory prediction model to obtain the user action trajectory prediction result. The embodiment of the present invention can effectively improve the prediction accuracy of user action trajectory by using a generative adversarial network to train the features extracted from the dynamic information posted by users in social network. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flow chart of a preferred embodiment of a method for predicting user action trajectory based on a social network provided by the present invention;

[0036] Figure 2 It is a structural schematic diagram of a user action trajectory prediction model in a user action trajectory prediction method based on a social network provided by the present invention;

[0037] Figure 3 It is a schematic diagram of the structure of a generator of a user action trajectory prediction model in a user action trajectory prediction method based on a social network provided by the present invention;

[0038] Figure 4 It is a schematic diagram of the discriminator structure of a user action trajectory prediction model in a user action trajectory prediction method based on a social network provided by the present invention;

[0039] Figure 5 It is a structural schematic diagram of a preferred embodiment of a user action trajectory prediction device based on a social network provided by the present invention;

[0040] Figure 6 It is a structural schematic diagram of another preferred embodiment of a user action trajectory prediction device based on a social network provided by the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] See also Figure 1 , Figure 1 1 is a flow chart of a preferred embodiment of a method for predicting user action trajectories based on a social network provided by the present invention. The method for predicting user action trajectories based on a social network comprises:

[0043] S1, obtaining dynamic information posted by users in a social network, and extracting text features and image features from the dynamic information;

[0044] S2, analyzing the correlation between the text feature and the image feature, and obtaining a condition set according to the correlation;

[0045] S3, training the user action trajectory prediction model according to the condition set to obtain a trained user action trajectory prediction model; wherein the user action trajectory prediction model includes a generator and a discriminator;

[0046] S4, inputting text features and image features extracted from dynamic information posted by users in the social network into the trained user action trajectory prediction model to obtain a user action trajectory prediction result.

[0047] Specifically, in today's social context, sharing has become another important function in mobile phones in addition to the communication function. People share their experiences on their mobile phones and record their travel moods and footprints through text and pictures. This information can be used to effectively predict the next move of each person, and then it is possible to know the possible movement of people within a specific area. In this embodiment, the dynamic information posted by users in the social network is first obtained, for example: the travel check-in record posted on Weibo, and the text features and picture features are extracted from the dynamic information. Secondly, the correlation between the text features and the picture features is analyzed, and the condition set is obtained according to the correlation. It should be noted that the condition set in this embodiment is a vector decoded in the Encoder-decoder model, and each bit in the vector indicates whether the position or time information in the text is related to the picture. The stronger the correlation, the larger the value on the bit. Then, the user action trajectory prediction model is trained according to the condition set to obtain a trained user action trajectory prediction model. Among them, the user action trajectory prediction model in this embodiment is a model constructed based on a generative adversarial network, so the user action trajectory prediction model at least includes a generator and a discriminator. Finally, the text features and image features extracted from the dynamic information posted by users on the social network are input into the trained user action trajectory prediction model to obtain the user action trajectory prediction result. The additional information mainly used in this embodiment is the records of people's travel check-ins (such as Weibo dynamics, including text and pictures). This information can contain important location information. This embodiment uses this information to allow the model to learn the most accurate trajectory information.

[0048] The embodiment of the present invention uses the dynamic information posted by the user to obtain effective information, accurately estimates the location of the user, and fully retains the correlation between the information of the text and the picture, which enables the subsequent GAN model to have the characteristic of identifying this correlation. In the embodiment of the present invention, the condition set is used to represent this correlation. The condition set dynamically represents the dynamic correlation of the user's footprint in space when the user posts different dynamics, and at the same time improves the learning ability of the generator and the discriminator in the GAN. In addition, the embodiment of the present invention selects the condition GAN with time series data processing capability, which uses RNN as the basic model in both the generator and the discriminator, and can be combined with the condition set in the generator. This correlation is so that even if the noise added to the generator can make the model have a strong learning ability, the model has a strong anti-interference ability to the noise, effectively improves the prediction accuracy of the model, and then improves the prediction accuracy of the user's action trajectory.

[0049] In another preferred embodiment, extracting text features and image features from the dynamic information in S1 specifically includes:

[0050] Extracting text features from the dynamic information through an Encoder-Decoder model;

[0051] Image features are extracted from the dynamic information through a convolutional neural network.

[0052] Specifically, since text and pictures are not features that can be directly learned by the model, this embodiment adopts the Encoder-Decoder model, which is used to encode text into features related to prediction. In layman's terms, it is to extract the features related to location information that are hidden in the text or can be directly extracted from the words for later use. In this embodiment, the dynamic information posted by users obtained from the social network is input into the Encoder-Decoder model in a chronological sequence, and a string of vectors V with the same length as the sequence length in the encoder is obtained through the Encoder-decoder model. The vector V contains data that can be trained by the user's action trajectory prediction model, and the dimension of the vector V is equal to the number of all possible attractions in the city where the user is located. It should be noted that the Encoder-Decoder model needs to be trained, and the trained Encoder-Decoder can give the element value that best matches the location of the user in the data set, that is, the latitude and longitude of the user's location. For pictures, a convolutional neural network is used, and the extracted information is compared with the text to give an approximate geographical location of the user.

[0053] In another preferred embodiment, the step S2, analyzing the correlation between the text feature and the image feature, and obtaining a condition set according to the correlation, specifically includes:

[0054] S201, analyzing the correlation between the text feature and the image feature through an Attention mechanism;

[0055] S202: Taking the text features and picture features corresponding to the correlation degree being greater than a preset value as the condition set.

[0056] Specifically, in this embodiment, the condition set is generated by the text features and image features in the feature extraction stage. Specifically, the correlation of these features is analyzed. Of course, the correlation is not determined by a variety of text and image features. It is only necessary to observe the features with obvious relationships. However, this recognition ability requires the model to obtain it through the learning of the Attention mechanism. Therefore, this embodiment introduces the use of the Attention mechanism in the feature extraction stage.

[0057] In another preferred embodiment, the method further comprises:

[0058] Generate the user location when the user posts the dynamic information according to the correlation between the text feature and the image feature through the Encoder-Decoder model;

[0059] Determine whether the generated user position deviates from the actual position, and correct the user position with the deviate.

[0060] Specifically, after extracting text features and image features from the dynamic information posted by users in the social network, this embodiment generates the user location at the time when the user posted the dynamic information based on the correlation between the text features and the image features through the Encoder-Decoder model. Then, it is determined whether there is a deviation between the generated user location and the actual location, and the user location with the deviation is corrected, and the user location can be corrected to a compromise position between the two. This embodiment verifies and corrects people's locations through additional information that may reflect people's whereabouts, which is more in line with the characteristics that the movement of people in the city is more flexible, thereby improving the prediction accuracy of the model.

[0061] In another preferred embodiment, the step S3, training the user action trajectory prediction model according to the condition set to obtain a trained user action trajectory prediction model, specifically includes:

[0062] S301, inputting the condition set and the noise vector into a generator in the user action trajectory prediction model, and having the generator predict the user's future action trajectory;

[0063] S302, the discriminator analyzes the error between the action trajectory predicted by the generator in a noise interference environment and the actual action trajectory;

[0064] S303, repeatedly training the generator and the discriminator according to the error until the error meets a preset condition, thereby obtaining a trained user action trajectory prediction model.

[0065] For details, please refer to Figure 2 , Figure 2It is a structural diagram of a user action trajectory prediction model in a user action trajectory prediction method based on a social network provided by the present invention. The model for prediction in the embodiment of the present invention is condition GAN (generative adversarial network), which is a GAN model with time series data processing, and mainly consists of two parts: a generator and a discriminator. Unlike the traditional GAN, in the process of generating the generator and the discriminator, the present embodiment needs to add a condition set (i.e., the condition window in the condition GAN) to achieve the purpose of improving the accuracy of the generator and the discriminator. The training process is specifically as follows: first, a preliminary generator is obtained by adding a noise vector to the condition set. At this time, the condition window is equivalent to adding additional contextual features to the user data set to increase the prediction accuracy. The generator will predict the user's future action trajectory, for example: the place where the user will check in in the future, or the place to go at the next time point. Then the generated discriminator still contains the degree of association in the condition set, and the discriminator with this degree of association recognition ability can make the prediction result of the generator as close to the real data as possible. The discriminator analyzes the error between the action trajectory predicted by the generator in a noisy environment and the actual action trajectory. The generator and the discriminator are repeatedly trained according to the error until the error meets the preset conditions, and a trained user action trajectory prediction model is obtained.

[0066] As a preferred solution, the network of the generator includes two layers: the first layer is a network layer that only uses the condition set for prediction; the second layer is a network layer that uses both the condition set and the noise vector for prediction.

[0067] For details, please refer to Figure 3 , Figure 3 The figure is a schematic diagram of the generator structure of the user action trajectory prediction model in the user action trajectory prediction method based on social network provided by the present invention. In this embodiment, the network of the generator includes two layers: the first layer is a network layer that only uses the condition set for prediction; the second layer is a network layer that combines the condition set and the noise vector into one and uses them for prediction at the same time.

[0068] See also Figure 4 , Figure 4 This is a schematic diagram of the discriminator structure of the user action trajectory prediction model in the user action trajectory prediction method based on a social network provided by the present invention. In this embodiment, the discriminator has only one RNN time series prediction layer, because the discriminator mainly determines how large the error of the prediction result of the generator is in a noise interference environment, and repeats the training based on this error until the extracted condition set can greatly improve the model accuracy.

[0069] It should be noted that this embodiment is to ensure that all data sets involved in the experiment can be safely trained to protect data sets of different departments. A federated learning framework is used here. The core is that the server can receive updated parameters sent by independent model servers in real time, and the final parameters allow the joint server to update a better learning parameter. There are many ways to implement federated learning, among which Pysyft can well complete various processes of federated learning. The general process is as follows:

[0070] First, each data set is trained on its own server (here it is assumed that the server can withstand the entire training process). Second, the servers send the parameters they obtain in each cycle to the federated server. Finally, the federated server updates the two different parameters to obtain new parameters. Of course, the update here is a trade-off between the correlation of the two data sets, and finally the parameters are sent back to different servers.

[0071] The whole process can also be described by the following pseudo code:

[0072] Algorithm 1Federated Learning

[0073] Require: A serial graphic record of user

[0074] Ensure: Where the Location is of this user and which the next POI it will go to with a high possibility.

[0075] 1: Input the content into the Encoder-decoder framework and generate aprofit Vector representation of this Content.

[0076] 2: Extract the feature of picture via convolution operation.

[0077] 3: Evaluate the degree of the relationship between content Vector and picture, then generate a condition window which could denote correlation among all features.

[0078] 4: With the condition window and noise Vector, produce a generator and a Discrimination

[0079] 5: Predict a possible location of user U

[0080] 6: Retrain the model and tune that how condition window should be represented.

[0081] In the feature extraction stage (i.e. the stage of generating conditional sets), the Encoder-Decoder generates an N-dimensional vector, where N refers to the number of possible attractions in the city where the user is located. If a text vector V is more likely to indicate that the user is near attraction n1, then the value of the index 1 element in the V vector is higher. Similarly, if there are K dynamics created by user U, k different text vectors will be generated.

[0082] For pictures, this embodiment uses CNN to extract features and uses softmax to give the most likely scenic spots in the picture. In order not to observe too many information points in the picture, the Attention mechanism is used to allow the model to adaptively select useful parts. Finally, the Pearson coefficient is used to generate the most likely location of the user at different dynamic time points, and then generate a condition set.

[0083] Correspondingly, the present invention also provides a user action trajectory prediction device based on a social network, which can implement all the processes of the user action trajectory prediction method based on a social network in the above embodiment.

[0084] See also Figure 5 , Figure 5 1 is a schematic diagram of a preferred embodiment of a user action trajectory prediction device based on a social network provided by the present invention. The user action trajectory prediction device based on a social network comprises:

[0085] An acquisition module 501 is used to acquire dynamic information posted by users in a social network and extract text features and image features from the dynamic information;

[0086] An analysis module 502 is used to analyze the correlation between the text feature and the image feature, and obtain a condition set according to the correlation;

[0087] A training module 503 is used to train the user action trajectory prediction model according to the condition set to obtain a trained user action trajectory prediction model; wherein the user action trajectory prediction model includes a generator and a discriminator;

[0088] The prediction module 504 is used to input the text features and image features extracted from the dynamic information posted by users in the social network into the trained user action trajectory prediction model to obtain the user action trajectory prediction result.

[0089] Preferably, the acquisition module 501 extracts text features and image features from the dynamic information, specifically including:

[0090] Extracting text features from the dynamic information through an Encoder-Decoder model;

[0091] Image features are extracted from the dynamic information through a convolutional neural network.

[0092] Preferably, the analysis module 502 is specifically used for:

[0093] Analyze the correlation between the text features and the image features through the Attention mechanism;

[0094] The text features and picture features corresponding to the correlation degree being greater than a preset value are used as the condition set.

[0095] Preferably, the device further comprises a correction module, wherein the correction module is used for:

[0096] Generate the user location when the user posts the dynamic information according to the correlation between the text feature and the image feature through the Encoder-Decoder model;

[0097] Determine whether the generated user position deviates from the actual position, and correct the user position with the deviate.

[0098] Preferably, the training module 503 is specifically used for:

[0099] Inputting the condition set and the noise vector into a generator in the user action trajectory prediction model, and having the generator predict the user's future action trajectory;

[0100] The discriminator analyzes the error between the action trajectory predicted by the generator in a noise interference environment and the actual action trajectory;

[0101] The generator and the discriminator are repeatedly trained according to the error until the error meets a preset condition, thereby obtaining a trained user action trajectory prediction model.

[0102] Preferably, the network of the generator comprises two layers: the first layer is a network layer that uses only the condition set for prediction; the second layer is a network layer that uses both the condition set and the noise vector for prediction.

[0103] In the specific implementation, the working principle, control process and technical effect of the user action trajectory prediction device based on social network provided by the embodiment of the present invention are the same as those of the user action trajectory prediction method based on social network in the above embodiment, and will not be repeated here.

[0104] See also Figure 6 , Figure 6 6 is a schematic diagram of another preferred embodiment of a user action trajectory prediction device based on a social network provided by the present invention. The user action trajectory prediction device based on a social network includes a processor 601, a memory 602, and a computer program stored in the memory 602 and configured to be executed by the processor 601. When the processor 601 executes the computer program, the user action trajectory prediction method based on a social network described in any of the above embodiments is implemented.

[0105] Preferably, the computer program can be divided into one or more modules / units (such as computer program 1, computer program 2, ...), and the one or more modules / units are stored in the memory 602 and executed by the processor 601 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in the user action trajectory prediction device based on social network.

[0106] The processor 601 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor, or the processor 601 may be any conventional processor. The processor 601 is the control center of the user action trajectory prediction device based on social networks, and various interfaces and lines are used to connect various parts of the user action trajectory prediction device based on social networks.

[0107] The memory 602 mainly includes a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function, etc., and the data storage area can store related data, etc. In addition, the memory 602 can be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, and a flash card (Flash Card), etc., or the memory 602 can also be other volatile solid-state storage devices.

[0108] It should be noted that the above-mentioned user action trajectory prediction device based on social network may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that Figure 6 The structural diagram is only an example of the above-mentioned user action trajectory prediction device based on social network, and does not constitute a limitation of the above-mentioned user action trajectory prediction device based on social network, and may include more or less components than shown in the figure, or combine certain components, or different components.

[0109] An embodiment of the present invention further provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the user action trajectory prediction method based on social network described in any of the above embodiments.

[0110] The embodiment of the present invention provides a user action trajectory prediction method, device and storage medium based on social network, which obtains dynamic information posted by users in social network and extracts text features and picture features from the dynamic information; analyzes the correlation between the text features and the picture features, and obtains a condition set according to the correlation; trains a user action trajectory prediction model according to the condition set to obtain a trained user action trajectory prediction model; wherein the user action trajectory prediction model includes a generator and a discriminator; and inputs text features and picture features extracted from dynamic information posted by users in social network into the trained user action trajectory prediction model to obtain a user action trajectory prediction result. The embodiment of the present invention can effectively improve the prediction accuracy of user action trajectory by using a generative adversarial network to train features extracted from dynamic information posted by users in social network.

[0111] It should be noted that the system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the system embodiment provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0112] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A user action trajectory prediction method based on social network, characterized in that: include: Obtain dynamic information posted by users in social networks, and extract text features and image features from the dynamic information; Analyzing the correlation between the text feature and the image feature, and obtaining a condition set according to the correlation; The user action trajectory prediction model is trained according to the condition set to obtain a trained user action trajectory prediction model; wherein the user action trajectory prediction model includes a generator and a discriminator; Inputting text features and image features extracted from dynamic information posted by users on social networks into the trained user action trajectory prediction model to obtain a user action trajectory prediction result; The analyzing the correlation between the text feature and the image feature and obtaining a condition set according to the correlation specifically includes: Analyze the correlation between the text features and the image features through the Attention mechanism; The text features and picture features corresponding to the correlation degree being greater than a preset value are used as the condition set; The method further comprises: Generate the user location when the user posts the dynamic information according to the correlation between the text feature and the image feature through the Encoder-Decoder model; Determine whether the generated user position deviates from the actual position, and correct the user position with the deviate.

2. The user action trajectory prediction method based on social network as claimed in claim 1, characterized in that: The extracting text features and picture features from the dynamic information specifically includes: Extracting text features from the dynamic information through an Encoder-Decoder model; Image features are extracted from the dynamic information through a convolutional neural network.

3. The user action trajectory prediction method based on social network as claimed in claim 1, characterized in that: The training of the user action trajectory prediction model according to the condition set to obtain a trained user action trajectory prediction model specifically includes: Inputting the condition set and the noise vector into a generator in the user action trajectory prediction model, and having the generator predict the user's future action trajectory; The discriminator analyzes the error between the motion trajectory predicted by the generator in a noise interference environment and the actual motion trajectory; The generator and the discriminator are repeatedly trained according to the error until the error meets a preset condition, thereby obtaining a trained user action trajectory prediction model.

4. The user action trajectory prediction method based on social network as claimed in claim 3, characterized in that: The network of the generator includes two layers: the first layer is a network layer that only uses the condition set for prediction; the second layer is a network layer that uses both the condition set and the noise vector for prediction.

5. A user action trajectory prediction device based on social network, characterized in that: include: An acquisition module is used to acquire dynamic information posted by users in social networks and extract text features and image features from the dynamic information; An analysis module, used for analyzing the correlation between the text features and the image features, and obtaining a condition set according to the correlation; A training module, used to train the user action trajectory prediction model according to the condition set to obtain a trained user action trajectory prediction model; wherein the user action trajectory prediction model includes a generator and a discriminator; A prediction module, used to input text features and image features extracted from dynamic information posted by users in social networks into the trained user action trajectory prediction model to obtain a user action trajectory prediction result; Wherein, the analysis module is specifically used for: Analyze the correlation between the text features and the image features through the Attention mechanism; The text features and picture features corresponding to the correlation degree being greater than a preset value are used as the condition set; The device further comprises a correction module, wherein the correction module is used for: Generate the user location when the user posts the dynamic information according to the correlation between the text feature and the image feature through the Encoder-Decoder model; Determine whether the generated user position deviates from the actual position, and correct the user position with the deviate.

6. The user action trajectory prediction device based on social network as claimed in claim 5, characterized in that: The acquisition module extracts text features and image features from the dynamic information, specifically including: Extracting text features from the dynamic information through an Encoder-Decoder model; Image features are extracted from the dynamic information through a convolutional neural network.

7. A user action trajectory prediction device based on social network, characterized in that: The invention comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for predicting user action trajectory based on social network as described in any one of claims 1 to 4 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the user action trajectory prediction method based on a social network according to any one of claims 1 to 4.

Images