Information processing methods, apparatuses, electronic devices and computer-readable storage media

Abstract

This invention discloses an information processing method, apparatus, electronic device, and computer-readable storage medium. These embodiments can be applied to various scenarios such as cloud technology, artificial intelligence, intelligent transportation, and assisted driving. The invention involves acquiring interactive information to be processed, extracting interactive objects from this information to obtain a set of interactive objects, filtering target interactive information for each interactive object in a preset set of interactive information, extracting interactive features from the target interactive information, converting these features into time-series features, and then fusing these time-series features to obtain target time-series features corresponding to the interactive information to be processed. Finally, based on these target time-series features, the target information of the interactive information to be processed is determined. This approach can improve the accuracy of information processing.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and more specifically to an information processing method, apparatus, electronic device, and computer-readable storage medium. Background Technology

[0002] In recent years, with the rapid development of internet technology, various platforms have become increasingly diverse in their services, especially interactive services on various interactive platforms. To ensure the security of interactive services, it is often necessary to detect the interactive information corresponding to these services. Existing information processing methods often employ specific expert rules to detect interactive information in order to complete the information processing.

[0003] In the process of researching and practicing existing technologies, the inventors of this invention discovered that since specific expert rules are formulated based on previously collected interaction data, there is often a certain lag, which reduces the accuracy of interaction information detection and thus reduces the accuracy of information processing. Summary of the Invention

[0004] The present invention provides an information processing method, apparatus, electronic device, and computer-readable storage medium, which can improve the accuracy of information processing.

[0005] An information processing method, comprising:

[0006] Obtain the interaction information to be processed, and extract the interaction objects participating in the interaction from the interaction information to obtain a set of interaction objects;

[0007] In the preset set of interactive information, the target interactive information of each interactive object in the set of interactive objects is selected, and the interactive features of the interactive object are extracted from the target interactive information.

[0008] The interaction features are converted into time series features, which are used to indicate the interaction changes of the interaction object in the target interaction information.

[0009] The time series features are fused to obtain the target time series features corresponding to the interaction information to be processed;

[0010] Based on the target time series characteristics, the target information of the interaction information to be processed is determined.

[0011] Accordingly, embodiments of the present invention provide an information processing apparatus, including:

[0012] The acquisition unit is used to acquire the interaction information to be processed and extract the interaction objects participating in the interaction from the interaction information to obtain a set of interaction objects;

[0013] The filtering unit is used to filter out the target interaction information of each interaction object in the set of interaction objects from the preset set of interaction information, and extract the interaction features of the interaction object from the target interaction information.

[0014] A conversion unit is used to convert the interaction features into time series features, wherein the time series features are used to indicate the interaction change information of the interaction object in the target interaction information;

[0015] A fusion unit is used to fuse the time series features to obtain the target time series features corresponding to the interaction information to be processed;

[0016] The determining unit is used to determine the target information of the interaction information to be processed based on the target time series characteristics.

[0017] Optionally, in some embodiments, the conversion unit may be specifically used to obtain the expansion parameters of the interaction feature, and expand the interaction feature according to the expansion parameters to obtain the expanded interaction feature; and convert the expanded interaction feature into the time series feature of the interaction object.

[0018] Optionally, in some embodiments, the conversion unit may be specifically used to identify a feature region to be expanded in the interaction features according to the expansion parameters; expand the region to be expanded to obtain an initial expanded interaction feature; and add preset feature information to the initial expanded interaction feature to obtain an expanded interaction feature.

[0019] Optionally, in some embodiments, the transformation unit may be specifically used to perform convolution processing on the expanded interaction features to obtain initial time series features of the interaction object in multiple dimensions; calculate the standard deviation of the initial time series features based on the feature values ​​of the initial time series features; and normalize the initial time series features based on the standard deviation to obtain the time series features of the interaction object.

[0020] Optionally, in some embodiments, the conversion unit may be specifically used to filter the initial time series feature with the smallest feature value from the initial time series features to obtain the target initial time series feature; calculate the feature difference between the initial time series feature and the target initial time series feature respectively; and determine the time series feature of the interactive object based on the feature difference and the standard deviation.

[0021] Optionally, in some embodiments, the conversion unit may be specifically used to calculate the ratio between the feature difference and the standard deviation to obtain the target feature value corresponding to each initial time series feature; and to fuse the initial time series features according to the target feature value to obtain the time series feature of the interactive object.

[0022] Optionally, in some embodiments, the determining unit may be specifically used to detect the target score of the interaction information to be processed in the target time series features using a trained information detection model; when the target score exceeds a preset score threshold, extract the current interaction features in the interaction information to be processed; determine the target level of the interaction information to be processed based on the current interaction features and the target score, and use the target level as the target information of the interaction information to be processed.

[0023] Optionally, in some embodiments, the processing unit may be specifically used to: intercept the interactive service corresponding to the interactive information to be processed when the target level exceeds a first target level threshold; generate interactive prompt information and send the interactive prompt information to each interactive object in the interactive information to be processed when the target level exceeds a second target level threshold, wherein the second target level threshold is less than the first target level threshold; and allow the interactive service corresponding to the interactive information to be processed when the target level does not exceed the second target level threshold.

[0024] Optionally, in some embodiments, the information processing device may further include a training unit, which may be used to acquire an interactive information sample set, the interactive information sample set including interactive information samples labeled with target ratings; predict the target rating of the interactive information samples using a preset information detection model to obtain a predicted target rating; and converge the preset information detection model based on the labeled target rating and the predicted target rating to obtain the trained information detection model.

[0025] Optionally, in some embodiments, the fusion unit may be used to concatenate the time series features to obtain concatenated time series features; and to perform dimensionality reduction processing on the concatenated time series features to obtain the target time series features corresponding to the interaction information to be processed.

[0026] Optionally, in some embodiments, the filtering unit may be specifically used to filter out a preset number of target interaction information from the target interaction information to obtain a set of target interaction information; extract basic interaction features of each target interaction information from the set of target interaction information; and fuse the basic interaction features to obtain the interaction features of the interaction object.

[0027] Furthermore, embodiments of the present invention also provide an electronic device, including a processor and a memory, wherein the memory stores an application program, and the processor is used to run the application program in the memory to implement the information processing method provided in embodiments of the present invention.

[0028] Furthermore, embodiments of the present invention also provide a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute steps in any of the information processing methods provided in embodiments of the present invention.

[0029] This invention, in its embodiments, acquires interactive information to be processed, extracts interactive objects participating in the interaction from the interactive information to obtain a set of interactive objects, and then filters out the target interactive information of each interactive object in the set of interactive objects from a preset set of interactive information. Interactive features of the interactive objects are extracted from the target interactive information, and then the interactive features are converted into time series features. These time series features are then fused to obtain the target time series features corresponding to the interactive information to be processed. Finally, the target information of the interactive information to be processed is determined based on the target time series features. Because this scheme extracts interactive features after acquiring the target interactive information of each interactive object in the interactive information to be processed and converts the interactive features into time series features, it accurately determines the characteristics and variability of interactive objects over a long lifespan, thereby enabling accurate detection of interactive information and improving the accuracy of information processing. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of a scenario for the information processing method provided in an embodiment of the present invention;

[0032] Figure 2 This is a flowchart illustrating the information processing method provided in an embodiment of the present invention;

[0033] Figure 3 This is a schematic diagram of dilated convolution for extracting initial time series features according to an embodiment of the present invention;

[0034] Figure 4 This is a schematic diagram of the training process of a preset risk detection model in a transaction order scenario provided by an embodiment of the present invention;

[0035] Figure 5This is a schematic diagram of the process for processing transaction orders provided in an embodiment of the present invention;

[0036] Figure 6 This is another flowchart illustrating the information processing method provided in this embodiment of the invention;

[0037] Figure 7 This is a schematic diagram of the structure of the information processing device provided in an embodiment of the present invention;

[0038] Figure 8 This is another structural schematic diagram of the information processing device provided in the embodiment of the present invention;

[0039] Figure 9 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] This invention provides an information processing method, apparatus, electronic device, and computer-readable storage medium. The business processing apparatus can be integrated into an electronic device, which may be a server or a terminal, etc.

[0042] The embodiments of the present invention can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, and assisted driving.

[0043] The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery network (CDN) acceleration services, and big data and artificial intelligence platforms. The terminal includes, but is not limited to, mobile phones, computers, smart voice interaction devices, smart home appliances, and in-vehicle terminals. The terminal and server can be directly or indirectly connected via wired or wireless communication, and this application does not impose any restrictions on this connection.

[0044] For example, see Figure 1Taking the integration of information processing devices into electronic devices as an example, the electronic device acquires interactive information to be processed, extracts the interactive objects participating in the interaction from the interactive information to be processed, and obtains a set of interactive objects. Then, it filters out the target interactive information of each interactive object in the set of interactive objects from the preset set of interactive information, extracts the interactive features of the interactive objects from the target interactive information, converts the interactive features into time series features, and then fuses the time series features to obtain the target time series features corresponding to the interactive information to be processed. Finally, based on the target time series features, the target information of the interactive information to be processed is determined, thereby improving the accuracy of information processing.

[0045] The information processing method provided in this application relates to machine learning within the field of artificial intelligence. This application embodiment can convert interaction features into time-series features, fuse the time-series features to obtain target time-series features of the interaction information to be processed, and determine the target information of the transaction to be processed based on the target time-series features.

[0046] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making capabilities. AI technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. AI software technology primarily includes computer vision technology and machine learning / deep learning.

[0047] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0048] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the preferred order of the embodiments.

[0049] This embodiment will be described from the perspective of an information processing device, which can be integrated into an electronic device, such as a server or a terminal. The terminal can include tablet computers, laptops, personal computers (PCs), wearable devices, virtual reality devices, or other smart devices capable of information processing.

[0050] An information processing method, comprising:

[0051] The system acquires the interaction information to be processed and extracts the interaction objects involved in the interaction from the interaction information to obtain a set of interaction objects. It then filters the target interaction information of each interaction object in the set of interaction objects from the preset set of interaction information and extracts the interaction features of the interaction objects from the target interaction information. The interaction features are converted into time series features, which are used to indicate the interaction changes of the interaction objects in the target interaction information. The time series features are then fused to obtain the target time series features corresponding to the interaction information to be processed. Based on the target time series features, the target information of the interaction information to be processed is determined.

[0052] like Figure 2 As shown, the specific process of this information processing method is as follows:

[0053] 101. Obtain the interaction information to be processed, and extract the interaction objects involved in the interaction from the interaction information to obtain a set of interaction objects.

[0054] Among them, the interaction information to be processed can be interaction information that needs to be reviewed. The so-called interaction information can be understood as the business information generated when the interaction objects interact. The interaction can include transactions, gifts, and other business that creates a relationship between at least two objects. For example, in the e-commerce scenario, the interaction information can be the shopping order submitted by the user when purchasing items from the merchant, or the payment order or payment information generated when the user pays for items from a certain merchant, or the business information generated when the user returns or complains about items from a certain merchant, etc. In the personal transaction scenario, it can also be the business information generated by the user transferring money, making payments, sending red envelopes, or giving tips to other users.

[0055] The interaction object can be any object participating in the interaction in the interaction information to be processed. For example, if the interaction information to be processed is a payment order, the interaction object can be the seller and the buyer in the interaction order.

[0056] There are several ways to obtain the interaction information to be processed, as follows:

[0057] For example, it can directly receive information processing requests sent by users through terminals, and identify the interaction information to be processed in the information processing requests. Alternatively, it can receive business information sent in real time by the interaction platform, and identify the interaction information to be processed that needs to be processed in the business information. Or, when there is a large number of interaction information to be processed or the memory is large, it can also receive information processing requests sent by the interaction platform, which carry the storage address of the business information. Based on the address, the business information can be obtained, and the interaction information to be processed that needs to be processed can be identified in the business information.

[0058] After obtaining the interaction information to be processed, the interaction objects involved in the interaction can be extracted from the interaction information. There are multiple ways to extract interaction objects. For example, at least two objects with an interaction relationship can be identified in the interaction information to be processed, and these objects can be used as interaction objects to obtain a set of interaction objects. For example, if the interaction information to be processed is a payment order, the payment object and the receiving object can be identified in the payment order, and the payment object and the receiving object can be used as interaction objects to obtain a set of interaction objects.

[0059] It should be noted that the set of interactive objects must contain at least two interactive objects. When multi-person interaction is involved, the number of interactive objects can be more than two. Multi-person interaction scenarios can take many forms. For example, on an e-commerce platform, when a buyer places an order for an item from a seller, they can invite a friend to pay on their behalf during the payment process. In this case, there are three parties involved in the interaction: the seller, the buyer, and the friend who is paying on behalf of the buyer. Alternatively, different branches of the same company may purchase the same item from the same seller. In this case, the interactive objects involved in the product interaction could be the seller and the buyers from the different branches, and so on.

[0060] 102. Select the target interaction information for each interaction object in the set of interaction objects from the preset set of interaction information, and extract the interaction features of the interaction objects from the target interaction information.

[0061] Interaction features are used to indicate the characteristic information generated by the interaction object in the interaction business. For example, taking the interaction information to be processed as a payment order, the interaction features can be characteristic information such as order quantity, order time, order amount and order object.

[0062] The target transaction information can be understood as the historical interaction information of interactive objects within a preset time period. There are multiple ways to filter out the target interaction information of each interactive object in the preset set of interaction information, as follows:

[0063] For example, you can directly filter the interaction information of each interaction object in the set of interaction objects within a preset time period from the preset set of interaction information to obtain the target interaction information. Alternatively, you can obtain the interaction information index in the preset set of interaction information and quickly filter the interaction information of each interaction object in the set of interaction objects within a preset time period up to the current time based on the interaction information index to obtain the target interaction information.

[0064] After filtering out the target interaction information of the interactive object, the interaction features of the interactive object can be extracted from the target interaction information. There are multiple ways to extract interaction features. For example, a preset number of target interaction information can be filtered out from the target interaction information to obtain a set of target interaction information. The basic interaction features of each target interaction information can be extracted from the set of target interaction information, and the basic interaction features can be merged to obtain the interaction features of the interactive object.

[0065] There are several ways to filter out a preset number of target interaction information from the target interaction information. For example, a preset number of target interaction information can be filtered out from historical interaction information. When the current number of target interaction information is less than the preset number, the difference between the current number and the preset number is calculated, and the preset interaction information with the preset number difference is added to the filtered target interaction information to obtain the target interaction information. The preset interaction information can be blank interaction information (0). For example, taking a payment order as the interaction information to be processed, for the interaction object contained in the payment order, N payment orders of the interaction object are extracted. If there are no N orders, 0 is added to form N orders, thus obtaining the set of target interaction information.

[0066] Among them, basic interaction features are used to characterize the interaction and variation information of interactive objects. Each target interaction information can have multiple dimensions of basic interaction features. There are various ways to fuse basic interaction features. For example, taking the case that there are N target interaction information in the set of target interaction information, and each target interaction information has M basic features, an N*M interaction sequence feature matrix can be constructed. The interaction sequence feature matrix is ​​used as the interaction feature of the interaction object. Specifically, formula (1) can be used as follows:

[0067]

[0068] in, Let N be the interaction features of the interactive object, N be the number of target interaction information in the target interaction information set of the interactive object, and M be the number of basic interaction features. Based on basic interaction features.

[0069] 103. Convert interaction features into time series features.

[0070] Time-series features are used to indicate changes in the interaction of an interactive object within the target interaction information. Taking a payment order as an example, the time-series features could include merchant / user behavior information, variation information, and payment information, etc. Variation information can be understood as information about sudden changes in the merchant's or user's interactive behavior within a certain period of time. For example, a user might pay 10 yuan once a day, but one day suddenly pay multiple times 1000 yuan. This sudden change can be understood as the user's variation information during that period. In simpler terms, variation information can be understood as information indicating a certain change in a preset number of historical interaction messages.

[0071] There are several ways to convert interaction features into time series data, including the following:

[0072] For example, the expansion parameters of the interaction features can be obtained, and the interaction features can be expanded according to the expansion parameters to obtain the expanded interaction features. The expanded interaction features can then be converted into time series features of the interaction objects.

[0073] The dilation parameter, also known as the dilation rate, is a hyperparameter in dilated convolution and indicates the location of holes in the interactive features. These holes are feature regions used to fill predefined feature information. Based on this dilation parameter, there are various ways to dilate interactive features. For example, the region to be dilated can be identified within the interactive features, dilated to obtain an initial dilated interactive feature, and then predefined feature information can be added to this initial dilated interactive feature to obtain the final dilated interactive feature.

[0074] There are several ways to identify the region to be expanded in the interaction features based on the expansion parameter. For example, with an expansion parameter of 2, the position of every row and column in the interaction sequence feature matrix corresponding to the interaction feature can be identified as the feature region to be expanded.

[0075] After identifying the feature region to be expanded, the region can be expanded. There are various expansion methods. For example, taking an expansion parameter of 2 as an example, the expansion can be performed on the interaction sequence feature matrix. Every other row and column is set to a hole to obtain the interactive features after the initial expansion.

[0076] After obtaining the initial expanded interactive features, preset feature information can be added to them. There are several ways to add preset feature information; for example, zeros can be filled into each hole to obtain the expanded interactive features. The methods for expanding the interactive features are as follows:

[0077] ➡

[0078] After expanding the interaction features, the expanded interaction features can be converted into time series features of the interaction object. There are several ways to do this. For example, convolution can be performed on the expanded interaction features to obtain initial time series features of the interaction object in multiple dimensions. Based on the feature values ​​of the initial time series, the standard deviation of the initial time series features can be calculated. Based on the standard deviation, the initial time series features can be normalized to obtain the time series features of the interaction object.

[0079] There are several ways to perform convolution processing on the expanded interaction features. For example, a dilated convolutional network of the trained information detection model can be used to convolve the expanded interaction features to obtain local features in multiple dimensions, which can then be used as the initial time series features.

[0080] It's worth noting that this scheme uses a dilated convolutional network to scan the entire two-dimensional interaction sequence feature matrix. This is primarily because dilated convolutions offer advantages over standard convolutions. Under the same computational conditions and without pooling loss, dilated convolutions increase the receptive field, allowing each convolution output to contain a wider range of information. A filter (convolutional kernel) is used to filter the various small regions of the interaction sequence feature matrix, thereby obtaining the feature values ​​of these small regions—that is, acquiring local features. These local features can then be used as initial time-series features of different dimensions. The process of dilated convolution extracting initial time-series features can be described as follows: Figure 3 As shown.

[0081] Among them, the initial time series features are normalized based on the standard deviation. There are various normalization methods. For example, the initial time series features with canceling feature values ​​are selected from the initial time series features to obtain the target initial time series features. The feature difference between the initial time series features and the target initial time series features is calculated respectively. Based on the feature difference and the standard deviation, the time series features of the interactive object are determined.

[0082] There are several ways to determine the time series features of the interactive object based on the feature difference and standard deviation. For example, the ratio between the feature difference and the standard deviation can be calculated to obtain the target feature value corresponding to each initial time series feature, as shown in formula (2).

[0083]

[0084] in, For the target feature value, These are the feature values ​​of the initial time series features. The feature values ​​of the initial time series features of the target are... denoted as the standard deviation of the eigenvalues ​​of the initial time series features.

[0085] After calculating the target feature value corresponding to each initial time series feature, the initial time series features can be fused based on the target feature value to obtain the time series features of the interactive object. There are several ways to fuse these features; for example, the target feature values ​​can be constructed as a time series matrix. This allows us to obtain the time-series characteristics of the interactive objects.

[0086] Alternatively, interaction features can be converted into time series features by using LSTM (Long Short-Term Memory) and RNN (Recurrent Neural Network) models to extract time series features from interaction features.

[0087] 104. The target time series features are fused to obtain the target time series features corresponding to the interaction information to be processed.

[0088] For example, time series features can be concatenated to obtain concatenated time series features, and then dimensionality reduction processing can be performed on the concatenated time series features to obtain the target time series features corresponding to the interaction information to be processed.

[0089] There are several ways to concatenate time-series features. For example, taking the interaction objects in the interaction information to be processed as merchants and users as an example, the user's time-series feature matrix can be used to... Time series feature matrix of merchants The data are concatenated to form an N*(M+G) fused time series feature matrix of merchants and users. The time series feature matrix will be fused. As a feature of the spliced ​​time series.

[0090] After concatenating the time series features, dimensionality reduction can be performed on the concatenated time series features. There are various ways to perform dimensionality reduction, such as fusing the time series feature matrix. A vector multiplied by (M+G)*1 This yields an N*1 fused feature vector. fused feature vectors of N*1 As the target time series features corresponding to the interaction information to be processed.

[0091] 105. Based on the characteristics of the target time series, determine the target information of the interactive information to be processed.

[0092] Among them, the target information can be used to evaluate the degree of risk of the interaction information to be processed. For example, it can be the target level of risk of the interaction information to be processed or other information to evaluate the degree of risk.

[0093] There are several ways to determine the target information of the interaction information to be processed, as follows:

[0094] For example, a post-trained information detection model can be used to detect the target score of the interaction information to be processed in the target time series features. When the target score exceeds the preset score threshold, the current interaction features are extracted from the interaction information to be processed. Based on the interaction features and the target score, the target level of the interaction information to be processed is determined, and the target level is used as the target information of the interaction information to be processed.

[0095] The target score can be a rating of whether the interaction information to be processed poses a risk. For example, taking a payment order as the interaction information to be processed, the target score could be a rating of whether the payment order is malicious. There are several ways to use a post-trained information detection model to detect the target score of the interaction information to be processed in the target time series. For instance, a fully connected layer network of the post-trained information detection model can be used to map the target time series features to a preset target score set, thereby obtaining the classification probability of each target score in the preset target score set. Based on this classification probability, the target score corresponding to the target time series features is selected from the preset target score set and used as the target score of the interaction information to be processed. After the fully connected layer network outputs the target score, the target score and the target time series features can also be stored as a key-value pair.

[0096] The post-trained information detection model can be configured according to the time application. Furthermore, it should be noted that the post-trained information detection model can be pre-configured by maintenance personnel or trained automatically by the information processing device. Specifically, before the step "detecting the target score of the interaction information to be processed in the target time series features using the post-trained information detection model," the information processing method may further include:

[0097] Obtain an interactive information sample set, which includes interactive information samples labeled with target scores. Use a pre-defined information detection model to predict the target score of the interactive information samples to obtain the predicted target score. Converge the pre-defined information detection model based on the labeled target score and the predicted target score to obtain the trained information detection model. Specifically, it can be done as follows:

[0098] (1) Obtain a sample set of interactive information.

[0099] The interactive information sample set includes interactive information samples that label the target rating.

[0100] There are several ways to obtain a sample set of interaction information, as follows:

[0101] For example, historical interaction information can be obtained from the interactive business platform, and target scores can be marked in the historical interaction information based on the risk review results of the historical interaction information, thereby obtaining interaction information samples. The interaction information samples can be combined to obtain an interaction information sample set.

[0102] There are several ways to label target scores. For example, taking payment orders as an example of historical interaction information, the target scores of the orders that have been complained about in the historical payment orders are labeled as positive samples, and the target scores of the orders of merchants that have been reviewed as normal, successfully appealed, or are obviously credible by human judgment are labeled as negative samples.

[0103] (2) The target score of the interactive information sample is predicted by using the preset information detection model to obtain the predicted target score.

[0104] For example, a dilated convolutional network of a pre-defined information detection model can be used to extract the time-series features of the interactive object for each interactive information sample in the interactive information sample set. The time-series features of the samples can be fused to obtain the time-series features of the target sample corresponding to the interactive information sample. A fully connected network of a pre-defined target detection model can be used to map the time-series features of the target sample to a pre-defined target rating set to obtain the classification probability corresponding to each target rating in the pre-defined target rating set. Based on the target probability, the target rating corresponding to the time-series features of the target sample can be selected from the pre-defined target rating set, and the target rating can be used as the predicted target rating of the interactive information sample.

[0105] (3) The preset information detection model is converged based on the labeled target score and the predicted target score to obtain the trained information detection model.

[0106] For example, the labeled target score and the predicted target score can be compared to obtain the loss information of the interactive information sample. Based on this loss information, the preset information detection model can be converged to obtain the trained information detection model.

[0107] There are several ways to converge the preset information detection model based on the loss information. For example, the gradient descent algorithm can be used to update the network parameters of the preset information detection model based on the loss information, thereby obtaining the trained information detection model.

[0108] When using interactive information samples as transaction order samples, the training process for the preset information detection model can involve acquiring historical transaction orders, using orders that have been complained about as positive samples, and using orders from merchants whose reviews are normal, whose appeals are successful, or whose orders are deemed credible by human judgment as negative samples, and utilizing the merchant user fusion feature vector. Using a fully connected neural network to predict the probability that an order is suspicious and fraudulent, such as... Figure 4 As shown, the specific process can be as follows:

[0109] 1) Label the relevant samples according to the logic of defining positive and negative samples.

[0110] 2) Process the relevant data of the labeled samples to obtain merchant-user fusion feature vectors that can be input into the model.

[0111] 3) Use a fully connected neural network to train the samples.

[0112] 4) Deploy the trained information detection model to a server with Python and related data dependencies installed. When a new transaction order occurs, predict the order's maliciousness score and provide the score results for risk control processing in the next step.

[0113] After detecting the target score of the interaction information to be processed, the target score can be checked. When the target score exceeds the preset score threshold, it can be determined that the interaction information to be processed has a high probability of having a target. At this time, the real-time strategy engine needs to be driven to extract the current interaction features from the interaction information to be processed. After extracting the current interaction features, the target level of the interaction information to be processed can be determined based on the current interaction features and the target score. There are several ways to determine the target level. For example, a first initial target level of the interaction information to be processed can be determined based on the target score, and a second initial target level can be determined based on the current interaction features. The first initial target level and the second initial target level are weighted to obtain the target level of the interaction information to be processed. Alternatively, the target level range of the interaction information to be processed can be determined based on the target score, and the target level of the interaction information to be processed can be determined based on the target level range and the current interaction features. Alternatively, the current interaction features can be converted into target features, and the target levels corresponding to the target features can be selected from the preset target level set to obtain the initial target level. The target score is then fused with the initial target level to obtain the target level of the interaction information to be processed. Once the target level of the interactive information to be processed is determined, the target level can be used as the target information of the interactive information to be processed.

[0114] Optionally, after obtaining the target information of the interaction information to be processed, the interaction information to be processed can be processed based on the target information. There are multiple ways to process the interaction information to be processed. For example, when the target level exceeds the first target level threshold, the interaction service corresponding to the interaction information to be processed is intercepted. When the target level exceeds the second target level threshold, an interaction prompt message is generated and sent to each interaction object in the interaction information to be processed. The second target level threshold is less than the first target level threshold. When the target level does not exceed the second target level threshold, the interaction service corresponding to the interaction information to be processed is allowed to proceed.

[0115] There are several ways to intercept the interactive business corresponding to the interactive information to be processed. For example, the business process of the interactive business corresponding to the interactive information to be processed can be directly intercepted, or the target level of the interactive information to be processed can be sent to the business server where the interactive business corresponding to the interactive information to be processed is located, so that the business server can directly intercept the interactive business, or the interactive information to be processed can be directly intercepted.

[0116] When the target level does not exceed the second target level threshold, it means that the interaction information to be processed does not pose a risk or the probability of a risk is low. In this case, the interaction business corresponding to the interaction information to be processed can be directly allowed. This allowance process can be understood as not intercepting the interaction information to be processed and the interaction business corresponding to the interaction information to be processed, and not generating any prompt information, so that the interaction object in the interaction information to be processed can complete the interaction business corresponding to the interaction information to be processed on the business platform.

[0117] The process of processing the interaction information to be processed mainly includes three parts. Taking a new transaction order as an example, where the interaction objects are a user and a merchant, the first part is to use the dilated convolutional layer of the trained information detection model to extract the user's time series features and the merchant's time series features respectively, and then fuse the user's time series features and the merchant's time series features. The second part is to use the fully connected layer of the trained information detection model to output the order's malicious score corresponding to the fused time series features. The third part is the strategy processing layer. When the order's malicious score exceeds a preset score threshold, the real-time strategy engine is driven, and the interaction features of the current order are extracted through the real-time strategy engine. Combined with the current interaction features, the maliciousness of the new transaction order is judged. If the maliciousness is low or there is no maliciousness, the order is allowed to proceed. If the maliciousness is high or there is maliciousness, the order is blocked or a warning is issued. Figure 5As shown, the target scores detected by the trained information detection model, combined with the final service and risk control processing, improved the online interception accuracy of fraud by 20% and the alert accuracy by 15%. Furthermore, the model's high iteration efficiency effectively solved the problem of rapid mutation in black market activities.

[0118] As can be seen from the above, the embodiments of this application, after acquiring the interaction information to be processed and extracting the interaction objects participating in the interaction from the interaction information to obtain a set of interaction objects, select the target interaction information of each interaction object in the set of interaction objects from the preset set of interaction information, extract the interaction features of the interaction objects from the target interaction information, convert the interaction features into time series features, and then fuse the time series features to obtain the target time series features corresponding to the interaction information to be processed. Then, based on the target time series features, the target information of the interaction information to be processed is determined, and the interaction information to be processed is processed based on the target information. Since this scheme extracts the interaction features after acquiring the historical interaction information of each interaction object in the interaction information to be processed and converts the interaction features into time series features, it can accurately determine the characteristics and variability of the interaction object over a long life cycle, and thus accurately detect the interaction information. Therefore, it can improve the accuracy of information processing.

[0119] Based on the method described in the above embodiments, the following examples will provide further detailed explanations.

[0120] In this embodiment, the information processing device is specifically integrated into an electronic device, the electronic device is a server, the interaction information to be processed is the transaction order to be processed, the transaction objects are merchants and users, the target rating is a malicious rating, the target level is a risk level, the target information is risk information, the target interaction information is historical transaction information, and the information detection model is a risk detection model, as an example.

[0121] (a) The server trains the risk detection model to obtain the trained risk detection model.

[0122] (1) The server obtains a sample set of transaction orders.

[0123] For example, the server can obtain historical transaction orders from the transaction business platform, and mark malicious scores in the historical transaction orders based on the risk assessment results of the historical transaction orders, thereby obtaining transaction order samples. The transaction order samples are then combined to obtain a transaction order sample set.

[0124] (2) The server uses a preset risk detection model to predict the malicious score of the transaction order sample and obtains the predicted malicious score.

[0125] For example, the server can use a dilated convolutional layer network with a pre-defined risk detection model to extract the sample time series features of the transaction object for each transaction order sample in the transaction order sample set. The sample time series features are then fused to obtain the target sample time series features corresponding to the transaction order sample. A fully connected layer network with a pre-defined risk detection model is then used to map the target sample time series features to a pre-defined malicious scoring set to obtain the classification probability corresponding to each malicious score in the pre-defined malicious scoring set. Based on this risk probability, the malicious score corresponding to the target sample time series features is selected from the pre-defined malicious scoring set, and this malicious score is used as the predicted malicious score for the transaction order sample.

[0126] (3) The server converges the preset risk detection model based on the labeled malicious score and the predicted malicious score to obtain the trained risk detection model.

[0127] For example, the labeled malicious scores and predicted malicious scores can be compared to obtain the loss information of the transaction order samples. Based on the loss information, the gradient descent algorithm is used to update the network parameters of the preset risk detection model, thereby obtaining the trained risk detection model.

[0128] (ii) Risk detection using a post-trained risk detection model.

[0129] The post-trained risk detection model can include dilated convolutional layer networks and fully connected layer networks.

[0130] like Figure 6 As shown, an information processing method has the following specific flow:

[0131] 201. The server obtains pending transaction orders and extracts the transaction objects involved in the transaction from the pending transaction orders to obtain a set of transaction objects.

[0132] For example, the server can directly receive information processing requests sent by users through terminals, and identify pending transaction orders in the information processing requests. Alternatively, it can receive business information sent by the trading platform in real time, and identify pending transaction orders that need to be processed in the business information. Or, when there are many pending transaction orders or a large amount of memory, it can also receive information processing requests sent by the trading platform, which carry the storage address of the business information. Based on the address, the server can obtain the business information and identify pending transaction orders that need to be processed in the business information.

[0133] After obtaining the pending transaction orders, the server can identify the merchants and users involved in the transaction from the transaction orders, and use the merchants and users as transaction objects to obtain a set of transaction objects.

[0134] 202. The server filters out the historical transaction orders of each transaction object in the transaction object set from the preset transaction order set.

[0135] For example, for a user in a transaction order, the server can extract N orders containing that user from a preset transaction order set. If there are no N orders, it can be padded with zeros to form N orders, thus obtaining the user's historical transaction orders. For a merchant in a transaction order, the server can extract N orders containing that merchant from a preset transaction order set. If there are no N orders, it can be padded with zeros to form N orders, thus obtaining the merchant's historical transaction orders.

[0136] 203. The server extracts the transaction characteristics of the transaction object from historical transaction orders.

[0137] For example, the server can filter out a preset number of historical transaction orders from the historical transaction orders to obtain a set of historical transaction orders. It can then extract the basic transaction characteristics of each historical transaction order from the set of historical transaction orders and merge the basic transaction characteristics to obtain the transaction characteristics of the transaction object.

[0138] Specifically, for each user in a transaction order, M features can be constructed. These M features are used to characterize the user's transaction and variation information, thus constructing an N*M user transaction sequence feature matrix. ,Will As a user's transaction characteristics Specifically as follows:

[0139]

[0140] Specifically, for each merchant in a transaction order, G features can be constructed. These G features are commonly used to characterize merchant transaction and variation information, thus constructing an N*G merchant transaction sequence feature matrix. ,Will As a characteristic of merchants' transactions, Specifically as follows:

[0141]

[0142] 204. The server converts transaction features into time series features.

[0143] For example, the server can obtain the expansion parameters of transaction features, identify the region to be expanded within the transaction features, expand the region to obtain the initial expanded transaction features, and add preset feature information to the initial expanded transaction features to obtain the expanded transaction features. A dilated convolutional network of the trained risk detection model is then used to convolve the expanded transaction features, thereby obtaining multi-dimensional local features, which are then used as the initial time series features.

[0144] Specifically, for users in a transaction order, taking an expansion parameter of 2 as an example, in the transaction sequence feature matrix... Every other row and column is set with holes to obtain the initial expanded transaction features. Each hole is then filled with zeros to obtain the expanded transaction features. Direct convolution is then performed to finally obtain a convolutional matrix of the user's initial time features. For merchants in transaction orders, the initial time feature matrix of the merchant can be transformed using the same process, which will not be elaborated here.

[0145] The server calculates the standard deviation of the initial time series features based on their feature values, selects those features whose feature values ​​cancel each other out, and obtains the target initial time series features. It then calculates the feature difference between the initial and target initial time series features, and the ratio between the feature difference and the standard deviation, thus obtaining the target feature value for each initial time series feature, as shown in formula (2). The target feature values ​​of the user are then used to construct the user time series matrix. The target feature values ​​of merchants are used to construct a merchant time series matrix. ,Will As a user's time series feature, As a time-series feature of merchants.

[0146] 205. The server fuses the target time series features to obtain the target time series features corresponding to the transaction orders to be processed.

[0147] For example, the server will use the user's time series feature matrix Time series feature matrix of merchants The data are concatenated to form an N*(M+G) fused time series feature matrix of merchants and users. The time series feature matrix will be fused. As features of the concatenated time series, the fused time series feature matrix will be used. A vector multiplied by (M+G)*1 This yields an N*1 fused feature vector. fused feature vectors of N*1 As the target time series feature corresponding to the pending transaction orders.

[0148] 206. The server determines the risk information of the pending transaction orders based on the characteristics of the target time series.

[0149] For example, the server can use a fully connected layer network of a trained risk detection model to map the target time series features to a preset malicious score set, thereby obtaining the classification probability of each malicious score in the preset malicious score set. Based on the classification probability, the malicious score corresponding to the target time series features is selected from the preset malicious score set, and the malicious score is used as the malicious score of the transaction order to be processed.

[0150] When a malicious score exceeds a preset scoring threshold, the server needs to drive a real-time strategy engine to extract the current transaction characteristics from the pending transaction orders. Based on the malicious score, a first initial risk level for the pending transaction order is determined. Based on the current transaction characteristics, a second initial risk level for the pending transaction order is determined. The first and second initial risk levels are weighted to obtain the risk level of the pending transaction order. Alternatively, the malicious score can be used to determine a risk level range for the pending transaction order, and the risk level of the pending transaction order can be determined based on this risk level range and the current transaction characteristics. Alternatively, the current transaction characteristics can be converted into risk characteristics, and the risk levels corresponding to the risk characteristics can be selected from a preset risk level set to obtain an initial risk level. The malicious score is then fused with the initial risk level to obtain the risk level of the pending transaction order.

[0151] 207. The server processes pending transaction orders based on risk information.

[0152] For example, when the risk level exceeds the first risk level threshold, the server can directly intercept the business process of the transaction corresponding to the pending transaction order, or it can send the risk level of the pending transaction order to the business server where the transaction corresponding to the pending transaction order is located, so that the business server can directly intercept the transaction, or it can directly intercept the pending transaction order.

[0153] When the risk level exceeds the second risk level threshold, the server generates a transaction alert and sends the alert to the merchants and users in the pending transaction orders. This second risk level threshold is less than the first risk level threshold.

[0154] When the risk level does not exceed the second risk level threshold, it means that the pending transaction order does not have any risk or the probability of having a risk is low. In this case, the transaction business corresponding to the pending transaction order can be directly released. This release process can be understood as not intercepting the pending transaction order and the transaction business corresponding to the pending transaction order, and not generating any prompt information, so that the merchants and users in the pending transaction order can complete the transaction business corresponding to the pending transaction order on the business platform.

[0155] As can be seen from the above, in this embodiment, the server acquires the transaction orders to be processed, extracts the transaction objects participating in the transaction from the transaction orders to obtain a set of transaction objects, and then filters out the historical transaction orders of each transaction object in the preset transaction order set. The transaction characteristics of the transaction objects are then extracted from the historical transaction orders, converted into time series characteristics, and then fused to obtain the target time series characteristics corresponding to the transaction order to be processed. Based on the target time series characteristics, the risk information of the transaction order to be processed is determined, and the transaction order to be processed is processed based on the risk information. Because this scheme extracts transaction characteristics after acquiring the historical transaction orders of each transaction object in the transaction order to be processed and converts the transaction characteristics into time series characteristics, it accurately determines the characteristics and variability of the transaction object over a long lifespan. This allows for timely responses to black market mutations and accurate risk detection of transaction orders, thus improving the accuracy of information processing.

[0156] To better implement the above methods, embodiments of the present invention also provide an information processing device, which can be integrated into an electronic device, such as a server or terminal, and the terminal may include a tablet computer, a laptop computer, and / or a personal computer.

[0157] For example, such as Figure 7 As shown, the information processing device may include an acquisition unit 301, a filtering unit 302, a conversion unit 303, a fusion unit 304, and a determination unit 305, as follows:

[0158] (1) Obtain unit 301;

[0159] The acquisition unit 301 is used to acquire the interaction information to be processed, and extract the interaction objects participating in the interaction from the interaction information to obtain a set of interaction objects.

[0160] For example, the acquisition unit 301 can specifically be used to receive information processing requests sent by users through terminals, identify the interaction information to be processed in the information processing requests, or receive business information sent in real time by the interaction platform, and identify the interaction information to be processed that needs to be processed in the business information, or, when the number of interaction information to be processed is large or the memory is large, it can also receive information processing requests sent by the interaction platform, which carry the storage address of the business information, acquire the business information based on the address, and identify the interaction information to be processed that needs to be processed in the business information. The interaction objects participating in the interaction are extracted from the interaction information to obtain a set of interaction objects.

[0161] (2) Screening unit 302;

[0162] The filtering unit 302 is used to filter out the target interaction information of each interaction object in the set of interaction objects from the preset set of interaction information, and extract the interaction features of the interaction object from the target interaction information.

[0163] For example, the filtering unit 302 can be used to filter the target interaction information of each interaction object in the set of interaction objects from a preset set of interaction information. Alternatively, it can obtain the interaction information index in the preset set of interaction information and quickly filter the target interaction information of each interaction object in the set of interaction objects up to the current time based on the interaction information index. A preset number of target interaction information is filtered from the target interaction information to obtain a set of target interaction information. The basic interaction features of each target interaction information are extracted from the target interaction information set, and the basic interaction features are fused to obtain the interaction features of the interaction objects.

[0164] (3) Conversion unit 303;

[0165] The conversion unit 303 is used to convert interactive features into time series features, which are used to indicate the interactive changes of interactive objects in the target interactive information.

[0166] For example, the conversion unit 303 can be used to obtain the expansion parameters of the interaction features, identify the region to be expanded in the interaction features, expand the region to obtain the initial expanded interaction features, add preset feature information to the initial expanded interaction features, and obtain the expanded interaction features. The expanded interaction features are then converted into time series features of the interaction object.

[0167] (4) Fusion unit 304;

[0168] The fusion unit 304 is used to fuse time series features to obtain the target time series features corresponding to the interaction information to be processed.

[0169] For example, the fusion unit 304 can be used to concatenate time series features to obtain concatenated time series features, and then perform dimensionality reduction processing on the concatenated time series features to obtain the target time series features corresponding to the interaction information to be processed.

[0170] (5) Determine unit 305;

[0171] The determining unit 305 is used to determine the target information of the interactive information to be processed based on the target time series characteristics.

[0172] For example, the determination unit 305 can be used to detect the target score of the interaction information to be processed in the target time series features using the trained target detection model. When the target score exceeds the preset score threshold, the current interaction features are extracted from the interaction information to be processed. Based on the interaction features and the target score, the target level of the interaction information to be processed is determined, and the target level is used as the target information of the interaction information to be processed.

[0173] Optionally, the information processing device may also include a training unit 306, such as Figure 8 As shown, the specific details are as follows:

[0174] Training unit 306 is used to train the information detection model to obtain the trained information detection model.

[0175] For example, training unit 306 can be used to obtain a set of interactive order samples, use a preset information detection model to predict the malicious rating of the interactive order samples, obtain the predicted malicious rating, and converge the preset information detection model based on the labeled malicious rating and the predicted malicious rating to obtain the trained information detection model.

[0176] In practice, each of the above units can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units, please refer to the previous method embodiments, which will not be repeated here.

[0177] As can be seen from the above, in this embodiment, after the acquisition unit 301 acquires the interaction information to be processed and extracts the interaction objects participating in the interaction from the interaction information to obtain a set of interaction objects, the filtering unit 302 filters the target interaction information of each interaction object in the set of interaction objects from the preset interaction information set, and extracts the interaction features of the interaction objects from the target interaction information. Then, the conversion unit 303 converts the interaction features into time series features. Then, the fusion unit 304 fuses the time series features to obtain the target time series features corresponding to the interaction information to be processed. Then, the determination unit 305 determines the target information of the interaction information to be processed based on the target time series features. Since this scheme extracts the interaction features after acquiring the historical interaction information of each interaction object in the interaction information to be processed and converts the interaction features into time series features, it can accurately determine the characteristics and variability of the interaction object over a long life cycle, and thus accurately detect the interaction information. Therefore, it can improve the accuracy of information processing.

[0178] This invention also provides an electronic device, such as... Figure 9 As shown, it illustrates a structural schematic diagram of the electronic device involved in an embodiment of the present invention, specifically:

[0179] The electronic device may include components such as a processor 401 with one or more processing cores, a memory 402 with one or more computer-readable storage media, a power supply 403, and an input unit 404. Those skilled in the art will understand that... Figure 9 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0180] The processor 401 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 402, and by calling data stored in the memory 402, it performs various functions and processes data, thereby performing overall detection of the electronic device. Optionally, the processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 401.

[0181] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0182] The electronic device also includes a power supply 403 that supplies power to the various components. Preferably, the power supply 403 can be logically connected to the processor 401 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 403 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0183] The electronic device may also include an input unit 404, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0184] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 401 in the electronic device loads the executable files corresponding to the processes of one or more applications into the memory 402 according to the following instructions, and the processor 401 runs the applications stored in the memory 402 to realize various functions, as follows:

[0185] The system acquires the interaction information to be processed and extracts the interaction objects involved in the interaction from the interaction information to obtain a set of interaction objects. It then filters the target interaction information of each interaction object in the set of interaction objects from the preset set of interaction information and extracts the interaction features of the interaction objects from the target interaction information. The interaction features are converted into time series features, which are used to indicate the interaction changes of the interaction objects in the target interaction information. The time series features are then fused to obtain the target time series features corresponding to the interaction information to be processed. Based on the target time series features, the target information of the interaction information to be processed is determined.

[0186] For example, an electronic device receives an information processing request sent by a user through a terminal. Within this request, it identifies the interaction information to be processed. Alternatively, it can receive real-time business information sent by an interactive platform and identify the interaction information currently requiring processing. Or, when the number of interaction information items is large or the memory is large, it can also receive information processing requests sent by the interactive platform, which carry the storage address of the business information. Based on the address, the business information is retrieved, and the interaction information currently requiring processing is identified within it. The interaction objects participating in the interaction are extracted from the interaction information to obtain a set of interaction objects. The target interaction information for each interaction object in the set of interaction objects is then selected from a preset set of interaction information. Alternatively, the interaction information index in the preset set of interaction information can be obtained, and the target interaction information for each interaction object up to the current time can be quickly selected based on this index. A preset number of target interaction information items are then selected from the target interaction information to obtain a set of target interaction information. The basic interaction features of each target interaction information item are extracted from this set of target interaction information, and these basic interaction features are fused to obtain the interaction features of the interaction objects. The process involves obtaining expansion parameters for the interaction features, identifying the region to be expanded within these features, expanding the region to obtain initial expanded interaction features, adding preset feature information to these initial expanded interaction features, and converting the expanded interaction features into time-series features of the interaction object. These time-series features are then concatenated to obtain concatenated time-series features, which are then subjected to dimensionality reduction to obtain the target time-series features corresponding to the interaction information to be processed. A trained information detection model is used to detect the target score of the interaction information to be processed within the target time-series features. When the target score exceeds a preset score threshold, the current interaction feature is extracted from the interaction information to be processed. Based on the interaction features and the target score, the target level of the interaction information to be processed is determined and used as the target information. Finally, the interaction information to be processed is processed based on the target information.

[0187] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0188] As can be seen from the above, in this embodiment of the invention, after acquiring the interaction information to be processed and extracting the interaction objects participating in the interaction from the interaction information to obtain a set of interaction objects, the target interaction information of each interaction object in the set of interaction objects is selected from the preset set of interaction information, and the interaction features of the interaction objects are extracted from the target interaction information. Then, the interaction features are converted into time series features, and then the time series features are fused to obtain the target time series features corresponding to the interaction information to be processed. Then, the target information of the interaction information to be processed is determined based on the target time series features. Since this scheme extracts the interaction features after acquiring the historical interaction information of each interaction object in the interaction information to be processed and converts the interaction features into time series features, it can accurately determine the characteristics and variability of the interaction object over a long life cycle, and thus accurately detect the target of the interaction information. Therefore, it can improve the accuracy of information processing.

[0189] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0190] To this end, embodiments of the present invention provide a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute steps in any of the information processing methods provided in the embodiments of the present invention. For example, the instructions can execute the following steps:

[0191] The system acquires the interaction information to be processed and extracts the interaction objects involved in the interaction from the interaction information to obtain a set of interaction objects. It then filters the target interaction information of each interaction object in the set of interaction objects from the preset set of interaction information and extracts the interaction features of the interaction objects from the target interaction information. The interaction features are converted into time series features, which are used to indicate the interaction changes of the interaction objects in the target interaction information. The time series features are then fused to obtain the target time series features corresponding to the interaction information to be processed. Based on the target time series features, the target information of the interaction information to be processed is determined.

[0192] For example, the system can receive information processing requests sent by users through their terminals, identify the interaction information to be processed within these requests, or receive real-time business information sent by the interaction platform and identify the interaction information currently requiring processing within that information. Alternatively, when the number of interaction information items is large or the memory is large, the system can also receive information processing requests sent by the interaction platform, which carry the storage address of the business information. Based on this address, the system retrieves the business information and identifies the interaction information currently requiring processing within it. The system then extracts the interaction objects participating in the interaction from the interaction information to obtain a set of interaction objects. It filters the target interaction information for each interaction object in the set of interaction objects from a preset set of interaction information. Alternatively, it can obtain the interaction information index from the preset set of interaction information and quickly filter the target interaction information for each interaction object up to the current time. Finally, it filters a preset number of target interaction information items from the target interaction information to obtain a set of target interaction information. The system then extracts the basic interaction features of each target interaction information item from this set and fuses these basic interaction features to obtain the interaction features of the interaction objects. The process involves obtaining expansion parameters for the interaction features, identifying the region to be expanded within these features, expanding the region to obtain initial expanded interaction features, adding preset feature information to these initial expanded interaction features, and converting the expanded interaction features into time-series features of the interaction object. These time-series features are then concatenated to obtain concatenated time-series features, which are then subjected to dimensionality reduction to obtain the target time-series features corresponding to the interaction information to be processed. A trained information detection model is used to detect the target score of the interaction information to be processed within the target time-series features. When the target score exceeds a preset score threshold, the current interaction feature is extracted from the interaction information to be processed. Based on the interaction features and the target score, the target level of the interaction information to be processed is determined and used as the target information. Finally, the interaction information to be processed is processed based on the target information.

[0193] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0194] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0195] Since the instructions stored in the computer-readable storage medium can execute the steps of any information processing method provided in the embodiments of the present invention, the beneficial effects that any information processing method provided in the embodiments of the present invention can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0196] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations of the information processing or transaction order processing aspects described above.

[0197] The foregoing has provided a detailed description of an information processing method, apparatus, electronic device, and computer-readable storage medium provided by embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. An information processing method, characterized in that, include: Obtain the interaction information to be processed, and extract the interaction objects participating in the interaction from the interaction information to obtain a set of interaction objects; In the preset set of interactive information, the target interactive information of each interactive object in the set of interactive objects is selected, and the interactive features of the interactive object are extracted from the target interactive information. Obtain the expansion parameters of the interaction feature, and expand the interaction feature according to the expansion parameters to obtain the expanded interaction feature; The expanded interaction features are convolved to obtain the initial time series features of the interaction object in multiple dimensions. Calculate the standard deviation of the initial time series features based on the feature values ​​of the initial time series features; The target initial time series feature is obtained by selecting the initial time series feature with the smallest feature value from the initial time series features. Calculate the feature difference between the initial time series features and the target initial time series features respectively; Based on the feature difference and standard deviation, the time series features of the interactive object are determined, and the time series features are used to indicate the interaction change information of the interactive object in the target interaction information; The time series features are fused to obtain the target time series features corresponding to the interaction information to be processed; Based on the target time series characteristics, the target information of the interaction information to be processed is determined.

2. The information processing method according to claim 1, characterized in that, The step of expanding the interaction feature according to the expansion parameter to obtain the expanded interaction feature includes: Based on the expansion parameters, the feature region to be expanded is identified in the interactive features; The feature region to be expanded is expanded to obtain the initial expanded interactive features; Preset feature information is added to the initial expanded interactive features to obtain the expanded interactive features.

3. The information processing method according to claim 1, characterized in that, Determining the time-series features of the interactive object based on the feature difference and standard deviation includes: Calculate the ratio between the feature difference and the standard deviation to obtain the target feature value corresponding to each initial time series feature; Based on the target feature value, the initial time series features are fused to obtain the time series features of the interactive object.

4. The information processing method according to any one of claims 1 to 3, characterized in that, The step of determining the target information of the interaction information to be processed based on the target time series characteristics includes: The target score of the interaction information to be processed is detected in the target time series features using a post-trained information detection model. When the target score exceeds a preset score threshold, the current interaction features are extracted from the interaction information to be processed. Based on the current interaction features and the target score, the target level of the interaction information to be processed is determined, and the target level is used as the target information of the interaction information to be processed.

5. The information processing method according to claim 4, characterized in that, After setting the target level as the target information of the interaction information to be processed, the method further includes: When the target level exceeds the first target level threshold, the interactive service corresponding to the interactive information to be processed is intercepted. When the target level exceeds the second target level threshold, an interactive prompt message is generated and sent to each interactive object in the interactive information to be processed, wherein the second target level threshold is less than the first target level threshold; When the target level does not exceed the second target level threshold, the interactive service corresponding to the interactive information to be processed is allowed to proceed.

6. The information processing method according to claim 4, characterized in that, Before the post-trained information detection model extracts the target features from the target time series features, the method further includes: Obtain an interaction information sample set, the interaction information sample set including interaction information samples labeled with target ratings; A preset information detection model is used to predict the target score of the interactive information sample, and the predicted target score is obtained. The preset information detection model is converged based on the labeled target score and the predicted target score to obtain the trained information detection model.

7. The information processing method according to any one of claims 1 to 3, characterized in that, The step of fusing the time series features to obtain the target time series features corresponding to the interaction information to be processed includes: The time series features are concatenated to obtain the concatenated time series features; The spliced ​​time series features are subjected to dimensionality reduction processing to obtain the target time series features corresponding to the interaction information to be processed.

8. The information processing method according to any one of claims 1 to 3, characterized in that, Extracting the interaction features of the interaction object from the target interaction information includes: A preset number of target interaction information are selected from the target interaction information to obtain a target interaction information set. Extract the basic interaction features of each target interaction information from the target interaction information set; The basic interaction features are fused together to obtain the interaction features of the interaction object.

9. An information processing device, characterized in that, include: The acquisition unit is used to acquire the interaction information to be processed and extract the interaction objects participating in the interaction from the interaction information to obtain a set of interaction objects; The filtering unit is used to filter out the target interaction information of each interaction object in the set of interaction objects from the preset set of interaction information, and extract the interaction features of the interaction object from the target interaction information. A conversion unit is used to obtain the expansion parameters of the interaction feature, and expand the interaction feature according to the expansion parameters to obtain the expanded interaction feature; The expanded interaction features are convolved to obtain initial time-series features of the interaction object in multiple dimensions; the standard deviation of the initial time-series features is calculated based on their feature values; the initial time-series features with the smallest feature values ​​are selected from the initial time-series features to obtain the target initial time-series features; the feature differences between the initial time-series features and the target initial time-series features are calculated respectively; the time-series features of the interaction object are determined based on the feature differences and the standard deviation, and the time-series features are used to indicate the interaction changes of the interaction object in the target interaction information; A fusion unit is used to fuse the time series features to obtain the target time series features corresponding to the interaction information to be processed; The determining unit is used to determine the target information of the interaction information to be processed based on the target time series characteristics.

10. An electronic device, characterized in that, It includes a processor and a memory, the memory storing an application program, and the processor running the application program within the memory to perform the steps of the information processing method according to any one of claims 1 to 8.

11. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the information processing method according to any one of claims 1 to 8.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the steps of the information processing method according to any one of claims 1 to 8.

Images