Event prediction method and related device

Abstract

The invention discloses an event prediction method and a related device, and the method comprises the steps: firstly obtaining task description information and task exception information of a plurality of feature value transfer tasks, the plurality of feature value transfer tasks comprise at least one transfer-out party and the same target transfer-in party, and the task exception information comprises an image certificate provided by the transfer-out party; the image voucher is an image identifying an abnormal link of the corresponding feature value transfer task. Then, according to the task description information and the task exception information, generating description features of a target transfer party; and finally, determining the possibility that the target transfer party relates to the transfer abnormal event based on the information identified by the image certificate in the description feature. According to the invention, the characteristics of the target transfer-in party related to the transfer abnormal event can be expressed more comprehensively through the image certificate, and the possibility that the target transfer-in party relates to the abnormal event can be determined more accurately according to the image certificate, so that the pre-warning of the transfer-out party can be realized based on the possibility.

Description

Technical Field

[0001] This application relates to the field of event prediction, and in particular to an event prediction method and related apparatus. Background Technology

[0002] With the development of the internet, it is common to obtain virtual or physical products through feature transfer. Feature transfer tasks include a transferor of the feature value and a transferee of the feature value. The transferee can be a provider of the product corresponding to the feature value, or in some feature transfer scenarios, it can also include an intermediary that transfers the feature value during the transfer process.

[0003] For the receiving party, there is a possibility that it may not provide the transferring party with products or transfer feature values, especially for receiving parties involved in a large number of feature value transfer tasks, which could result in abnormal transfer events that harm the interests of a large number of transferring parties.

[0004] To reduce the risks posed by abnormal transfer events to eigenvalue transfer scenarios, there is an urgent need for a way to accurately predict such events. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides an event prediction method and related apparatus that can more comprehensively express the characteristics of the target transferee related to abnormal transfer events through image credentials, more accurately determine the possibility that the target transferee is involved in abnormal events based on the image credentials, and thus provide advance warning to the transferor based on this possibility.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] On one hand, embodiments of this application provide an event prediction method, the method comprising:

[0008] Obtain task description information and task exception information for multiple feature value transfer tasks. The multiple feature value transfer tasks include at least one transferor and the same target transferee. The task exception information includes image credentials provided by the transferor. The image credentials are images that identify the abnormal links of the corresponding feature value transfer tasks.

[0009] Based on the task description information and the task anomaly information, generate the descriptive features of the target transferee;

[0010] Based on the information identified by the image credentials in the descriptive features, the likelihood that the target recipient is involved in an abnormal transfer event is determined.

[0011] On the other hand, embodiments of this application provide an event prediction device, the device comprising: an acquisition module, a generation module, and a determination module;

[0012] The acquisition module is used to acquire task description information and task exception information of multiple feature value transfer tasks. The multiple feature value transfer tasks include at least one transferor and the same target transferee. The task exception information includes an image certificate provided by the transferor. The image certificate is an image that identifies the abnormal link of the corresponding feature value transfer task.

[0013] The generation module is used to generate descriptive features of the target recipient based on the task description information and the task exception information.

[0014] The determining module is used to determine the likelihood that the target transferee is involved in an abnormal transfer event based on the information identified by the image credential in the descriptive features.

[0015] In another aspect, embodiments of this application provide a computer device, which includes a processor and a memory:

[0016] Memory is used to store computer programs;

[0017] The processor is used to execute the methods described above according to a computer program.

[0018] On another front, embodiments of this application provide a computer-readable storage medium for storing a computer program that performs the methods described above.

[0019] In another aspect, embodiments of this application provide a computer program product including a computer program, which, when run on a computer device, causes the computer device to perform the methods described above.

[0020] As can be seen from the above technical solution, in order to effectively predict possible transfer anomalies in feature value transfer scenarios, for the target transferee, task description information and task anomaly information of multiple associated feature value transfer tasks are obtained. The task anomaly information includes image credentials provided by the transferor, which identify the abnormal steps involved in the feature value transfer task in image form. Since feature value transfer tasks involve numerous steps and complex numerical and temporal information, it is difficult to directly and clearly describe the abnormal steps using only text. However, by taking screenshots of key interfaces of the abnormal steps, the complex information of the abnormal steps can be clearly demonstrated. Through the task anomaly information of multiple feature value transfer tasks related to the target transferee, the characteristics of the target transferee related to transfer anomalies can be more comprehensively expressed. The task description information identifies the basic information of these feature value transfer tasks. Combining the task description information and the task anomaly information can yield a comprehensive description of the target transferee's characteristics, thereby determining the likelihood that the target transferee may be involved in transfer anomalies in the future, and thus enabling early warning based on this likelihood. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A schematic diagram of a scenario for an event prediction method provided in an embodiment of this application;

[0023] Figure 2 A flowchart illustrating an event prediction method provided in this application embodiment;

[0024] Figure 3 This is a schematic diagram of comment information provided in an embodiment of this application;

[0025] Figure 4 This is a schematic diagram of an event prediction method provided in an embodiment of this application;

[0026] Figure 5(a) illustrates a source of comment information provided in an embodiment of this application;

[0027] Figure 5(b) shows another source of comment information provided in the embodiments of this application;

[0028] Figure 6 This is a schematic diagram illustrating the extraction of text information from an image, as provided in an embodiment of this application.

[0029] Figure 7 A schematic diagram illustrating a training-obtained prediction model provided in an embodiment of this application;

[0030] Figure 8 A flowchart illustrating an event prediction method provided in this application embodiment;

[0031] Figure 9 A schematic diagram of an event prediction device provided in an embodiment of this application;

[0032] Figure 10 A structural diagram of a terminal device provided in an embodiment of this application;

[0033] Figure 11 This is a structural diagram of a server provided in an embodiment of this application. Detailed Implementation

[0034] The embodiments of this application will now be described with reference to the accompanying drawings.

[0035] In related technologies, virtual or physical products can be acquired through feature value transfer. Feature value transfer tasks generally involve two roles: the transferor (transferring the feature value) and the transferee (receiving the feature value). Multiple feature value transfer tasks may include one transferee and one or more transferors, with the transferee considered the target transferee. In feature value transfer tasks, there is a possibility that the target transferee may acquire the transferee's feature value without providing the product or transferring the feature value. This would cause significant losses to the transferors, and such feature value transfer tasks would be considered abnormal events. To ensure the security of feature value transfers for the relatively weaker transferors, it is desirable to accurately predict target transferees potentially involved in abnormal transfer events, thus providing advance warning to the transferor and avoiding feature value transfer tasks with such target transferees. However, current technologies cannot accurately predict target transferees potentially involved in abnormal transfer events.

[0036] Therefore, this application provides an event prediction method and related apparatus, which can more comprehensively express the characteristics of the target transferee and the abnormal transfer event through image credentials, making it easier to more accurately determine the possibility that the target transferee is involved in the abnormal event based on the image credentials, thereby enabling early warning to the transferor based on the possibility.

[0037] The event prediction method provided in this application can be implemented using a computer device, which can be a terminal device or a server. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. Terminal devices include, but are not limited to, mobile phones, computers, smart voice interaction devices, smart home appliances, vehicle terminals, and aircraft. The terminal device and the server can be directly or indirectly connected via wired or wireless communication, and this application does not impose any limitations on this connection.

[0038] First, we will explain several terms that may be involved in the embodiments of this application below.

[0039] Second Clearing: This refers to the second step in the feature value clearing process, which involves verifying and adjusting the feature values ​​between the transferor and the target transferee.

[0040] Text pretraining refers to preprocessing the input image and text information before training the initial prediction model in order to improve the performance of the resulting prediction model.

[0041] Abnormal behavior characteristics of feature value transfer: refers to behavioral characteristics that do not conform to normal feature value transfer tasks during the feature value transfer process, such as large feature value transfer tasks, suspicious feature value transfer tasks, etc.

[0042] 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 possess the functions of perception, reasoning, and decision-making.

[0043] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, pre-trained model technology, operating / interactive systems, and mechatronics. Among these, pre-trained models, also known as large-scale models or foundational models, can be widely applied to downstream tasks across various AI fields after fine-tuning. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0044] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP deals with natural language, the language people use in daily life, and is closely related to linguistics research. It also involves crucial techniques for model training in artificial intelligence; pre-trained models are developed from large language models in NLP. After fine-tuning, large language models can be widely applied to downstream tasks. NLP techniques typically include text processing, semantic understanding, machine translation, question answering, and knowledge graphs.

[0045] Computer vision (CV) is the science that studies how to enable machines to "see." More specifically, it refers to machine vision, which uses cameras and computers to replace human eyes in tasks such as target recognition, tracking, and measurement, and further performs image processing to create images more suitable for human observation or transmission to instruments. As a scientific discipline, computer vision studies related theories and technologies, attempting to build artificial intelligence systems capable of extracting information from images or multidimensional data. Computer vision technologies typically include image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, 3D object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping (SLAM), and common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0046] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. 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 pre-trained learning. Pre-trained models represent the latest development in deep learning, integrating all of these techniques.

[0047] With the research and advancement of artificial intelligence (AI) technology, AI is being studied and applied in various fields, such as smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, autonomous driving, drones, digital twins, virtual humans, robots, AI-generated content (AIGC), conversational interaction, smart healthcare, smart customer service, and game AI. It is believed that with the development of technology, AI will be applied in more fields and play an increasingly important role.

[0048] The solutions provided in this application involve artificial intelligence technologies such as machine learning, natural language processing, and computer vision. For example, before using a predictive model to determine the likelihood of a transferee being involved in an abnormal transfer event, it is necessary to train an initial predictive model to obtain a predictive model. Through the training process, the initial predictive model learns to have the ability to identify abnormal transfer events. Furthermore, this application involves using computer vision technology to identify and extract descriptive features of the target transferee from the image text information in the image voucher, and enabling the predictive model to determine the likelihood of the target transferee being involved in an abnormal transfer event based on the descriptive features of the target transferee. Additionally, this application involves natural language processing, requiring preprocessing of the acquired image text information to remove non-objective information. Specific examples are provided below:

[0049] Figure 1 This is a schematic diagram of a scenario for an event prediction method provided in an embodiment of this application, wherein the aforementioned computer device is a server.

[0050] When it is necessary to determine the possibility of a transfer exception event involving the target transferee, the server first needs to obtain the task description information and task exception information of multiple feature value transfer tasks. These multiple feature value transfer tasks may include one or more transferors, each corresponding to the same transferee (i.e., the target transferee). For example... Figure 1 As shown in the diagram, there are n feature value transfer tasks involved. Feature value transfer task 1 involves transferor B, feature value transfer task 2 involves transferor C, and feature value transfer task n involves transferor N. All n feature value transfer tasks involve the same target transferee A. Different task description information and task exception information correspond to different feature value transfer tasks. Based on all task description information and task exception information, descriptive features of target transferee A can be generated.

[0051] The task anomaly information includes image credentials provided by each transferor, which are images used to identify abnormal stages in the feature value transfer task. Figure 1The image shown is a schematic diagram of an image credential, which may include, but is not limited to, the following: task stage, sender, target recipient, number of feature value transfers, and occurrence time. Based on the descriptive characteristics of target recipient A, the likelihood of target recipient A being involved in an abnormal transfer event can be determined. This image credential is beneficial for determining the details of the abnormal stage and the target recipient's feature value transfer behavior within that stage. Compared to describing the abnormal stage of a feature value transfer task solely through text, using an image credential makes the occurrence process of the abnormal stage more visual and intuitive, facilitating a more accurate determination of the likelihood of target recipient being involved in an abnormal event.

[0052] Figure 2 This is a flowchart of an event prediction method provided in an embodiment of this application. The method can be executed by a computer device. In this embodiment, the computer device is a server as an example for illustration.

[0053] S201: Obtain task description information and task exception information for multiple feature value transfer tasks.

[0054] Multiple feature value transfer tasks can involve one sender and one receiver, or multiple senders and one receiver, and these tasks can occur on different interaction platforms. More complex feature value transfer tasks often involve multiple receivers and multiple senders; the number of senders and receivers involved in a feature value transfer task is not limited here. In this embodiment, the receivers involved in multiple feature value transfer tasks all include at least one same receiver (i.e., the target receiver). This is because, in this embodiment, it is desirable to determine the probability of transfer anomalies involving the target receiver. Therefore, the obtained feature value transfer tasks need to be for the same target receiver. Thus, the corresponding task description information and task anomaly information are both for this target receiver, and the generated descriptive features are a more comprehensive descriptive feature for the target receiver obtained by mixing multiple task description information and task anomaly information.

[0055] The task description information identifies the basic information of the feature value transfer task, which may include, but is not limited to, the following: additional information about the feature value transfer task, and the industry characteristics involved in the feature value transfer task. The task anomaly information identifies abnormal stages in the feature value transfer task. This anomaly information includes image credentials provided by the transferor, which are images identifying the abnormal stages of the corresponding feature value transfer task. Abnormal stages refer to special stages in the feature value transfer task that do not conform to the behavioral characteristics of a normal feature value transfer process. For example, large feature value transfers to the target recipient, suspicious feature value transfers, etc. Generally speaking, when abnormal stages occur, it means that there is a certain probability that the transferor's interests in the feature value transfer task will be harmed.

[0056] Feature transfer tasks often involve numerous stages, each with different feature values ​​and timing information. These stages can include outgoing stages and intermediary stages. The outgoing stage can be understood as the stage where the feature value is transferred from the originating party to the target receiving party. The intermediary stage can be understood as the stage where the originating party transfers the feature value to an intermediary third party for safekeeping, and subsequently, the intermediary third party transfers the feature value to the target receiving party (this can be considered the intermediary outgoing stage).

[0057] Because complex numerical and time information is involved in each stage of the feature value transfer task, it is difficult to clearly and accurately reflect the various complex information involved in the abnormal stage if only a textual description of the abnormal stage is used. Therefore, in this embodiment of the application, by including image credentials provided by the transferor in the task abnormality information, the information involved in the abnormal stage can be visually and intuitively displayed, which is more accurate than textual description while ensuring the clarity of the information. Image credentials can refer to the image representation of the abnormal stage by the transferor in the feature value transfer task. Image credentials can include: screenshots of the interface corresponding to the abnormal stage, screenshots of the communication interface between the transferor and the target transferee regarding the abnormal stage, and screenshots of the interface at different stages of the abnormal stage handling process, etc.

[0058] S202: Generate descriptive features of the target transferee based on the task description information and the task exception information.

[0059] Descriptive features refer to the description of the behavioral characteristics of the target transferee in the feature value transfer task. By fusing task description information and task exception information, descriptive features for the target transferee can be generated. These descriptive features can comprehensively reflect the performance of the target transferee in multiple feature value transfer tasks, as well as the behavioral characteristics corresponding to different feature value transfer tasks.

[0060] One method for generating descriptive features is to fuse the task description information and task exception information of a single feature value transfer task to obtain the descriptive features of the target recipient in that feature value transfer task. Then, the descriptive features generated by all feature value transfer tasks are fused to obtain the descriptive features of the target recipient. For example, suppose there are three feature value transfer tasks for target recipient A: Feature Value Transfer Task A, Feature Value Transfer Task B, and Feature Value Transfer Task C. In this case, the task description information A and task exception information A in Feature Value Transfer Task A correspond to generate sub-descriptive feature A of the target recipient; the task description information B and task exception information B in Feature Value Transfer Task B correspond to generate sub-descriptive feature B of the target recipient; and the task description information C and task exception information C in Feature Value Transfer Task C correspond to generate sub-descriptive feature C of the target recipient. The final descriptive feature of the target recipient is a feature that fuses sub-descriptive features A, B, and C. The generation of descriptive features using the first generation method described above allows for the determination of corresponding sub-descriptive features for different feature value transfer tasks during the generation process. These sub-descriptive features can specifically reflect the behavioral characteristics of the target transferee in different feature value transfer tasks, establishing a correlation between the feature value transfer task and the behavioral characteristics of the target transferee. Finally, the descriptive features of the target transferee obtained by fusing the sub-descriptive features will better reflect the behavioral characteristics of the target transferee for different feature value transfer tasks.

[0061] A second method for generating descriptive features involves fusing task description information and task exception information from all feature value transfer tasks to obtain the descriptive features of the target recipient in all feature value transfer tasks. Continuing with the example above, suppose there are three feature value transfer tasks for target recipient A: Feature Value Transfer Task A, Feature Value Transfer Task B, and Feature Value Transfer Task C. In this case, the descriptive features are directly generated by fusing task description information A and task exception information A from Feature Value Transfer Task A, task description information B and task exception information B from Feature Value Transfer Task B, and task description information C and task exception information C from Feature Value Transfer Task C. Compared to the first method, this second method of generating descriptive features is simpler to implement, easier to carry out, and improves the efficiency of descriptive feature generation.

[0062] S203: Based on the information identified by the image credentials in the descriptive features, determine the likelihood that the target transferee is involved in an abnormal transfer event.

[0063] The aforementioned transfer anomaly event refers to an abnormal situation where, after receiving the feature value transferred by the transferor, the target recipient fails to continue the subsequent steps according to the original feature value transfer task process. For example, suppose the target recipient is a product provider, and the transferor expects to transfer feature values ​​to the target recipient to obtain the target recipient's product. If, after the transferor completes the feature value transfer, the target recipient fails to deliver a product equivalent to the feature value to the transferor according to the original feature value transfer task process, or if the target recipient transfers the feature value transferred by the transferor through improper means, this is considered a transfer anomaly event.

[0064] Descriptive features refer to a comprehensive description of the target recipient's behavioral characteristics within the feature value transfer tasks, obtained from multiple feature value transfer tasks. These descriptive features include information identified by image credentials, which use images to display information related to abnormal stages in the feature value transfer task. The information identified in the image credentials identifies the target recipient's behavioral characteristics during abnormal stages of the feature value transfer task. This identified information may include, but is not limited to, time information, numerical information, and other information related to transfer anomalies. Task anomaly information for a single feature value transfer task only reflects the target recipient's behavior in that specific task and is insufficient to provide a comprehensive macro-level understanding of the target sender's overall behavioral characteristics. For example, suppose that for a certain feature value transfer task, the anomaly information provided by the sender indicates that the target recipient has experienced product delays. Determining the likelihood of the target recipient being involved in a transfer anomaly solely based on the descriptive information corresponding to that specific feature value transfer task would likely result in a biased conclusion.

[0065] In this embodiment, the aim is to determine the likelihood of the target transferee being involved in transfer anomalies by considering a more comprehensive set of overall characteristics of the target transferee, thereby improving the accuracy of the determined likelihood. Therefore, the descriptive features integrate task description information and task anomaly information corresponding to multiple feature value transfer tasks. By expanding the amount of data related to the target transferee's behavioral characteristics in relation to feature value transfer tasks, a more comprehensive overall information on the target transferee's behavior in these tasks can be obtained. Determining the likelihood of the target transferee being involved in transfer anomalies based on these descriptive features will yield results that are more closely aligned with actual circumstances and more instructive.

[0066] In addition to task anomaly information, task description information is also considered as a factor. Task description information reflects the basic information of the feature value transfer task. When task description information and task anomaly information are combined to generate descriptive features, this descriptive information includes not only the characteristics of the target transferee related to the transfer anomaly event, but also the scenario corresponding to the occurrence of the transfer anomaly event (i.e., the task description information). That is, the descriptive feature will simultaneously include the scenario characteristics of the transfer anomaly event and the characteristics of the target transferee involved in the transfer anomaly event. The descriptive feature obtained by combining these two factors can be used to determine the likelihood of the target transferee being involved in future transfer anomaly events. When the likelihood of the target transferee being involved in a transfer anomaly event is high, it means that the target transferee is likely to experience a transfer anomaly event in the near future, requiring an early warning for the transferring party to mitigate losses to some extent. When the likelihood of the target transferee being involved in a transfer anomaly event is low, it means that the target transferee is unlikely to experience a transfer anomaly event in the near future, and therefore no early warning is needed.

[0067] As can be seen from the above technical solution, in order to effectively predict possible transfer anomalies in feature value transfer scenarios, for the target transferee, task description information and task anomaly information of multiple associated feature value transfer tasks are obtained. The task anomaly information includes image credentials provided by the transferor, which identify the abnormal steps involved in the feature value transfer task in image form. Since feature value transfer tasks involve numerous steps and complex numerical and temporal information, it is difficult to directly and clearly describe the abnormal steps using only text. However, by taking screenshots of key interfaces of the abnormal steps, the complex information of the abnormal steps can be clearly demonstrated. Through the task anomaly information of multiple feature value transfer tasks related to the target transferee, the characteristics of the target transferee related to transfer anomalies can be more comprehensively expressed. The task description information identifies the basic information of these feature value transfer tasks. Combining the task description information and the task anomaly information can yield a comprehensive description of the target transferee's characteristics, thereby determining the likelihood that the target transferee may be involved in transfer anomalies in the future, and thus enabling early warning based on this likelihood.

[0068] The foregoing provided a detailed description of task anomaly information, which is used to identify abnormal stages in a feature value transfer task. Generally, task anomaly information is provided by the sender through the backend of the interactive platform conducting the feature value transfer task. By providing task anomaly information specific to the target recipient to the backend, the interactive platform assists in reducing or recovering feature values ​​lost due to abnormal stages in the feature value transfer task. To enrich the sources of task anomaly information and increase the accuracy of determining the likelihood of the target recipient being involved in transfer anomaly events, one possible implementation is to obtain task anomaly information as follows: obtain the sender's comments on the target stage of the feature value transfer task, and determine the task anomaly information based on the comments.

[0069] The transferor's eigenvalue transfer task with the target transferee involves multiple stages. For each stage, the transferor can post comments related to that stage or the eigenvalue transfer task on the corresponding interface (or other interfaces involved in the task). These comments include image evidence describing the problem. This image evidence typically includes some or all stages involved in the eigenvalue transfer task, the eigenvalue transfer details at each stage, and the time when each stage occurred. The image evidence for the target stage typically includes the eigenvalue transfer details for the problematic stage, the time when the problematic stage occurred, and information about the target transferee.

[0070] The sender's comments on the target stage of the feature value transfer task directly reflect their perception of that stage. These comments may describe the target stage as unusual compared to standard feature value transfer tasks, or express the sender's opinion that there are problems. While such comments may not be sufficient to directly identify abnormal behavior from the target recipient, they can serve as indirect evidence to enrich the information on task anomalies, thereby increasing the accuracy of determining the likelihood of the target recipient being involved in anomaly events.

[0071] Since the sender's comments can appear at various stages involved in the feature value transfer task, and these comments are generally targeted descriptions based on each stage, the stages are divided into target stages and other stages based on whether problems occur within them. Target stages are those that encounter problems during the feature value transfer task. Considering the correlation between comments and stages, this embodiment obtains the sender's comments on the target stages of the feature value transfer task. Comment information can be obtained by identifying keywords related to the target stage within the comments, or by searching the interface where the target stage occurs; the specific method of acquisition is not limited here.

[0072] Based on the comments from the sender regarding the target stage (i.e. the stage where the problem occurred), and combined with the image evidence of the problem in the comments, the task anomaly information for that target stage can be identified.

[0073] By obtaining the task anomaly information through the methods described above, we can obtain the transferor's comments on the target stage of the feature value transfer task. This comments directly reflect the transferor's feelings about the target stage and can serve as a source of corroborating evidence to enrich the task anomaly information. Since the information in the task anomaly information is more comprehensive, the information in the generated descriptive features for the target transferee is also more comprehensive, which helps to improve the accuracy of determining the possibility that the target transferee is involved in the transfer anomaly event.

[0074] As mentioned earlier, the comment information from the transferring party regarding the target stage of the feature value transfer task includes image credentials describing problems in the target stage. To further enrich the sources of task anomaly information, in addition to image credentials, the comment information can also include textual information describing the problems. Textual information refers to information generated through textual descriptions. Figure 3 This is a schematic diagram of comment information provided in an embodiment of this application, such as... Figure 3As shown in the figure, interface A displays relevant information about the target recipient A, and the comment section below displays the comment information. As can be seen from the figure, the comment information can include both image evidence and text information.

[0075] In this embodiment, when determining task anomaly information for a problematic stage in the feature value transfer process, the image credentials provided by the transferor are used as image-dimensional information to describe the problem. These image credentials may include information such as the target transferor involved in the target stage, the feature value transferred, and the time of occurrence. Based on these image credentials, the occurrence process of the target stage can be visualized, and the interface where the stage occurred and the subject of the feature value transfer task can be clearly and intuitively identified. Since the information in the image credentials is relatively independent, the transferor can describe the information in the image credentials through text, thus organizing and combining the relatively independent information in the image credentials to form a detailed textual description or targeted description (i.e., textual information) of the problem in the target stage.

[0076] When identifying task anomalies, both image evidence and text information can be used as two different dimensions of commentary. Adding text information as another dimension to describe problems arising during the feature value transfer task ensures that the task anomalies identified based on commentary include multi-dimensional descriptions. Different dimensions of information emphasize different aspects of the problem description. Image evidence clearly and intuitively demonstrates the specific process of the target stage, focusing on a holistic and complete presentation of the actual situation. Text information can express the occurrence of the target stage from the perspective of the sender, emphasizing the sender's feelings. By combining these two dimensions to describe problems arising during the feature value transfer task, the sources of description are enriched, making the description of the problems more comprehensive and resulting in more complete information in the identified task anomalies.

[0077] When the comment information includes both image evidence and text information describing the problem, the corresponding task anomaly information includes information from both the image and text dimensions. When generating descriptive features of the target transferee based on the task description information and task anomaly information, the task description information and task anomaly information need to be fused to obtain the descriptive features of the target transferee. Then, based on these descriptive features, the likelihood of the target transferee being involved in a transfer anomaly event is determined.

[0078] Figure 4 This is a schematic diagram of an event prediction method provided in an embodiment of this application, such as... Figure 4As shown, this method involves the following modules: an image acquisition module, a text acquisition module, a feature fusion module, and an anomaly recognition module. Specifically, the image credential in the task anomaly information can be acquired by the image acquisition module, while the image text information in the image credential, the text information in the task anomaly information, and the task description information can be acquired by the text acquisition module. The fusion of the task description information and the task anomaly information can be performed by the feature fusion module, and finally, the anomaly recognition module determines the likelihood that the target transferor is involved in a transfer anomaly event.

[0079] The image acquisition module can employ a Convolutional Recurrent Neural Network (CRNN) model. The detailed model structure will be explained in detail later during the processing of image text information in image credentials when determining task anomalies based on comment information.

[0080] The text acquisition module can employ a Bidirectional Encoder Representations from Transformers (BERT) model. A detailed text acquisition module may include two parts: word embeddings and a bidirectional Long Short-Term Memory (LSTM) network. In this embodiment, a 300-dimensional Global Vectors for Word Representation (GloVe) pre-trained word vectors are used to convert the image text information in the acquired image credentials, the text information in the task anomaly information, and the task description information into vector representations. Then, a bidirectional LSTM is used to extract features from the aforementioned three types of text, with the LSTM's hidden layer size set to 128.

[0081] The feature fusion module can use a fully connected layer to connect the information extracted by the image acquisition module and the text acquisition module, and then fuse these features through a fully connected layer or other fusion techniques (such as attention mechanisms).

[0082] The anomaly identification module can be implemented using a prediction model. Detailed information about prediction models will be introduced later in the section on methods for determining transfer anomaly events involving the target transferor based on descriptive features.

[0083] By adding textual information to the comment information provided above, and using both image credentials and textual information—two descriptive dimensions for the stages where problems arise during the feature value transfer task—the information about the target stage can be fully expressed. Since the comment information includes descriptive information of multiple dimensions, the corresponding task anomaly information will also have richer information about the anomaly. This will make the descriptive features generated based on the task anomaly information more comprehensive in terms of the target transferor, which will help improve the accuracy of determining the possibility that the target transferor is involved in anomaly events.

[0084] The foregoing has provided a detailed description of the content of the comment information. The comment information may include image evidence describing problems encountered during the feature value transfer task, as well as text information describing problems encountered during the feature value transfer task. The source of the comment information is explained below. The foregoing mentioned "obtaining the comment information from the transferor regarding the target stage of the feature value transfer task." In one possible implementation, the comment information is obtained as follows: (1) Obtaining information from the interactive platform used to process the feature value transfer task, describing the target stage of the feature value transfer task from the transferor, as the comment information. Or, (2) Obtaining the conversation interface between the transferor and the target transferee regarding the target stage of the feature value transfer task as the comment information.

[0085] The obtained comment information originates from either the interactive platform handling the feature value transfer task (for the first acquisition method) or the conversation interface between the sender and the recipient (for the second acquisition method). When the comment information originates from the interactive platform handling the transfer task, it can be information posted or displayed by the sender on the platform's interface describing the target stage of the feature value transfer task. This comment information can be generated as text or as an image (i.e., image credential) on the interface, and its location can include, but is not limited to, the interface of the corresponding interactive platform when the target stage occurs.

[0086] When the comment information originates from the conversation interface between the sender and the recipient, the corresponding comment information can be a screenshot of the conversation interface (partial or full screen), or it can be text or image information extracted from the information in the conversation interface (i.e., images related to the target segment presented in the conversation content). The screenshots and extracted image information mentioned here can be regarded as image evidence.

[0087] The aforementioned comment information refers to a description of the target stage of the feature value transfer task. This description may include expressions indicating that the feature value transfer task is unusual compared to a conventional feature value transfer task, or that the sender believes there are problems. The sender's description of the target stage of the feature value transfer task may appear on the interface corresponding to the stage (or other interfaces involved in the feature value transfer task). For example, the interface where the stage appears may be the interactive platform where the feature value transfer task occurs, or it may be the conversation interface where the sender and the target recipient discuss the feature value transfer task during the process of the feature value transfer task. Figure 5(a) shows one source of comment information provided by an embodiment of this application. As shown in Figure 5(a), the source of comment information is the interactive platform processing the feature value transfer task. Figure 5(b) shows another source of comment information provided by an embodiment of this application. As shown in Figure 5(b), the source of comment information is the conversation interface between the sender and the target recipient.

[0088] Therefore, in this embodiment, the source of comment information may include, but is not limited to, the interactive platform for processing feature value transfer tasks, and the session interface for communication between the transferor and the target transferee. The comment information provided by the transferor on the interactive platform for processing feature value transfer tasks generally includes part or all of the feature value transfer process. The comment information provided by the transferor on the session interface for communication between the transferor and the target transferee generally includes the negotiation process between the transferor and the target transferee regarding the feature value transfer task, as well as the consensus reached. In this embodiment, the source of comment information is not limited; when the content of the descriptive information is a problem of the target stage of the feature value transfer task, it can be used as comment information to determine task anomaly information.

[0089] As can be seen from the above description of the sources of the comment information, the sources of the comment information in this embodiment are not limited, and at least originate from the interactive platform for processing feature value transfer tasks or the session interface between the sender and the target receiver. Because the sources of the comment information are relatively broad, the corresponding descriptions of the target stages of the feature value transfer task are more comprehensive, and the information contained in the task anomaly information determined based on this comment information is also richer.

[0090] The aforementioned "determining task anomaly information based on comment information" refers to the inclusion of image credentials describing problems arising during the feature value transfer task. These image credentials include image-text information, which contains both objective and subjective information. Subjective information refers to information that does not conform to objective facts and is subjective or biased. To improve the effectiveness of the task anomaly information, it is advisable to remove the subjective information, thereby further improving the accuracy of subsequent assessments of the likelihood of transfer anomalies involving the target recipient. Therefore, in one possible implementation, the method for determining task anomaly information is as follows: First, obtain the image-text information from the image credentials and keywords used to describe objective information in the feature value transfer task. Then, preprocess the image-text information based on the keywords to remove the subjective information, thus obtaining the task anomaly information.

[0091] In the eigenvalue transfer task, objective information refers to information or factors used to objectively describe the process occurring in the eigenvalue transfer task. For example, this objective information may include: the target recipient, the transferor, the eigenvalue transfer value, and the stages in which the task occurs. Keywords corresponding to this objective information may include "stage", "date", "transferor", "target recipient", and "eigenvalue transfer value", etc.

[0092] Non-objective information refers to information that has a low correlation with the feature transfer task and is highly subjective. For example, it may include the transferor's dissatisfaction with the feature transfer process, such as "He went too far" or "I was cheated".

[0093] The image text information in an image certificate refers to the text-type information extracted from the image certificate. In this embodiment, the method for obtaining the image text information in an image certificate can be implemented through the aforementioned image acquisition module, which can employ a Convolutional Recurrent Neural Network (CRNN) model. This model combines the characteristics of Convolutional Neural Network (CNN) layers, Gated Recurrent Unit (GRU) layers, and Connectionist Temporal Classification (CTC) layers. CNN is used to extract the image text information from the image certificate, GRU is used to capture the sequence information of the extracted image text information, and CTC is used to recognize the image text information whose sequence information has been determined. The three layers mentioned above are described below.

[0094] A CNN consists of convolutional layers, activation functions, and pooling layers. It can have three convolutional layers, using 3x3, 5x5, and 7x7 kernels respectively, with Rectified Linear Units (ReLU) as the activation function, and max pooling layers with 2x2 kernels.

[0095] GRU is a recurrent neural network primarily used for processing sequential data, featuring fewer parameters and faster computation speed. Using GRU as a subsequent processing layer after a CNN layer allows for the capture of the sequential information of image text within image credentials. In this embodiment, a bidirectional GRU can be used to capture the sequential information of image text.

[0096] CTC is a loss function used for sequence labeling problems, which can convert image text information in image credentials into a sequence of labels. Greedy decoding or bundle search decoding is used to recognize the image text information in image credentials.

[0097] Figure 6 This is a schematic diagram illustrating the extraction of image text information according to an embodiment of this application. See also... Figure 6 As shown, after inputting the image voucher into the CRNN model, the image text information can be obtained. This image text information includes information such as "Transferor B", "A is a shell company, don't trust him!!!", "Transfer process", and "Number of feature value transfers".

[0098] According to the above Figure 6 The image and text information obtained can be preprocessed by combining it with the keywords mentioned above used to describe objective information in the feature value transfer task. This allows for the removal of non-objective information, resulting in task anomaly information. The keywords for objective information mentioned earlier can include terms such as "stage," "date," "transferor," "target transferee," and "feature value transfer value." Therefore, by preprocessing the obtained image and text information according to the keyword types, we can identify objective information including: "feature value transfer value," "occurrence time: January 1, 20XX," "target transferee: A," and "transferor B," etc.; and non-objective information including: "A is a shell company, don't trust them!!!." This non-objective information needs to be removed, and the remaining image and text information is considered task anomaly information.

[0099] It should be noted that, in the embodiments of this application, a Transformer-based model, such as a deep learning model based on a visual attention mechanism (Vision Transformer, ViT), can also be used instead of CRNN. ​​The ViT model divides the image credential into a series of small blocks, and then processes these small blocks as sequence data to obtain image text information.

[0100] The method for determining task anomaly information described above involves using feature value transfer to extract keywords from objective task information. This preprocessing of image text information in image credentials removes non-objective information to obtain the task anomaly information. Removing non-objective information avoids including information that contradicts objective facts or reflects the subjectivity or bias of the sender, thus ensuring the validity of the information in the task anomaly information to a certain extent.

[0101] S203 mentioned above states that "based on the information identified by the image credential in the descriptive features, the probability that the target transferee is involved in a transfer anomaly event is determined." The descriptive features are features obtained by fusing task description information and task anomaly information to describe the target transferee's performance in the feature value transfer task. In determining the probability that the target transferee is involved in a transfer anomaly event through the descriptive features, a specialized model can be used to improve efficiency. Therefore, in one possible implementation, the method for determining the probability that the target transferee is involved in a transfer anomaly event is: based on the information identified by the image credential in the descriptive features, a prediction model is used to determine the probability that the target transferee is involved in a transfer anomaly event.

[0102] The training process for the prediction model is as follows:

[0103] A1: Obtain training samples for training based on historical feature value transfer tasks.

[0104] The historical feature value transfer task includes historical transfer-out and historical transfer-in components. The training samples include sample task description information and sample task anomaly information for the historical transfer-in components.

[0105] The historical feature value transfer tasks involved in the training samples are not limited to the same historical transferee. That is, the training samples can include historical feature value transfer tasks corresponding to multiple historical transferors and multiple historical transferees. For example, multiple historical transferors may be responsible for historical feature value transfer tasks for the same historical transferee, or multiple historical transferors may be responsible for historical feature value transfer tasks for different historical transferees. This can ensure the diversity of the training samples, so that the initial prediction model can learn more sample features from the training samples about the probability of determining that the historical transferee is involved in transfer anomaly events.

[0106] The training samples include sample task description information and sample task anomaly information from the historical transferor. The sample task description information refers to the basic information of the historical feature value transfer task, which may include, but is not limited to, the following: additional information about the historical feature value transfer task, industry characteristics involved in the historical feature value transfer task, etc. The historical task anomaly information identifies abnormal steps in the historical feature value transfer task. This information includes sample image vouchers provided by the historical transferor, which identify the abnormal steps in the corresponding historical feature value transfer task. By using sample image vouchers as a description method for abnormal steps in historical feature value transfer tasks, the complex information involved in the abnormal steps, such as numerical and time information, can be clearly and accurately represented. Image vouchers provide a more intuitive and visual representation of the information than textual descriptions.

[0107] In addition to the sample task description information and sample task anomaly information of the historical transferee, the training samples also include sample labels. The sample labels of the training samples are used to identify whether the corresponding historical transferee is involved in transfer anomaly events. The training samples include positive samples and negative samples. The historical transferee corresponding to the positive sample is involved in transfer anomaly events, and the historical transferee corresponding to the negative sample is not involved in transfer anomaly events.

[0108] In this embodiment, the training samples are distinguished into positive and negative samples. Positive samples represent historical transferees of the type of interest, while negative samples represent historical transferees of other types. Considering that the purpose of training the initial prediction model is to enable the completed prediction model to determine the probability that the target transferee is involved in transfer anomaly events, the training process tends to learn the sample features of historical transferees involved in transfer anomaly events. Therefore, historical transferees involved in transfer anomaly events are treated as positive samples, and historical transferees not involved in transfer anomaly events are treated as negative samples.

[0109] It should be noted that training samples can undergo data preprocessing before being input into the initial prediction model. Data preprocessing includes steps such as data cleaning, data augmentation, and data standardization. Normalization is performed on the sample image credentials, transforming the pixel value range from (0,255) to (0,1). Word segmentation and word embedding are performed on the text information in the sample task anomaly information.

[0110] A2: Input the initial prediction model based on the training samples, and train the initial prediction model based on the difference between the prediction results of the initial prediction model and the sample labels to obtain the prediction model.

[0111] The training samples, carrying sample task description information, sample task anomaly information, and sample labels, are input into the initial prediction model. The initial prediction model then obtains a prediction result, which is the result of the initial prediction model predicting that the corresponding historical transferee is involved in transfer anomaly events, including both historical transferees involved in transfer anomaly events and historical transferees not involved in transfer anomaly events. The prediction result is compared with the sample labels carried in the training samples to determine the differences between the two. Based on the differences, the initial prediction model is trained to obtain the prediction model. During the training process of the initial prediction model, based on the differences between the training samples, prediction results, and sample labels, the initial prediction model can gradually learn what types of transferors, what feature value transfer task scenarios, and what characteristics of target transferees are likely to be involved in transfer anomaly events.

[0112] During the training of the initial prediction model, a loss function can be used to determine the difference between the prediction result and the sample label. In this embodiment, the loss function can be the binary cross-entropy loss function. An optimizer is used to update the parameters of the initial prediction model based on the loss function to minimize the loss function. In this embodiment, the optimizer can be the Adam optimizer. The learning rate controls the step size of the parameter updates of the initial prediction model. In this embodiment, the initial learning rate is set to 0.001, and then a learning rate decay strategy is used, for example, decaying once every 10 epochs at a decay rate of 0.1. The batch size determines the number of training samples for each parameter update, for example, setting the batch size to 32. The number of iterations determines the number of training epochs of the model, which can be set to 500. Simultaneously, a portion of the training samples can be allocated as validation samples.

[0113] In this embodiment, determining the likelihood of a target transferor being involved in an abnormal transfer event based on descriptive features can be achieved through the aforementioned anomaly identification module. This anomaly identification module can be implemented using a prediction model, which may include a fully connected layer and an activation function. The fully connected layer linearly transforms the output of the previous layer to map and classify the descriptive features, and the number of neurons in the fully connected layer can be set to 1. The activation function is used to convert the linear output into a non-linear output. A sigmoid function is used as the activation function for the fully connected layer. The output range of the sigmoid function is (0,1), and the output value can be used to represent the probability value (i.e., the likelihood of the target transferor being involved in an abnormal transfer event). The prediction model can be an ensemble learning model, such as a Gradient Boosting Decision Tree (GBDT).

[0114] Figure 7This is a schematic diagram of a training-obtained prediction model provided in an embodiment of this application, such as... Figure 7 As shown, during the training of the initial prediction model, training samples are input into the model. These training samples include task descriptions, anomaly information, and labels for historical transferees in different historical feature value transfer tasks. The anomaly information includes sample image credentials provided by the historical transferor, which are used to identify the abnormal stages of the corresponding historical transfer task. The sample labels indicate whether the corresponding historical transferee was involved in an anomaly event. Samples where the historical transferee was involved in an anomaly event are considered positive samples, while those where the historical transferee was not involved are considered negative samples.

[0115] It should be noted that both positive and negative samples refer to the set of historical feature value transfer tasks involving the same historical transferee. The specific set is shown in the figure. Taking positive samples as an example, this set consists of n historical feature value transfer tasks involving historical transferee A. These tasks are: Historical Feature Value Transfer Task 1 (feature value transfer task between historical transferee A and historical transferor B), Historical Feature Value Transfer Task 2 (feature value transfer task between historical transferee A and historical transferor C), ..., Historical Feature Value Transfer Task n (feature value transfer task between historical transferee A and historical transferor N). The initial prediction model predicts the probability of historical transferees being involved in abnormal transfer events based on the above input data. This prediction result is then compared with the sample labels. The difference between the prediction result and the sample labels is used to train the initial prediction model to obtain the final prediction model.

[0116] The method described above for determining the probability of a target transferee being involved in anomaly events through a predictive model, along with the corresponding training process of the predictive model, improves efficiency by using the predictive model to determine the probability of a target transferee being involved in anomaly events based on descriptive features. Furthermore, during the training of the initial predictive model, the use of positive and negative samples allows the model to better distinguish between historical transferees involved in and not involved in anomaly events, thereby improving the accuracy of determining the probability of a target transferee being involved in anomaly events.

[0117] The aforementioned S201 mentions "obtaining task description information and task anomaly information for multiple feature value transfer tasks." The event prediction method proposed in this application has the following application requirement: it needs to determine the probability that the target transferee in a certain interactive platform is involved in an abnormal transfer event. This enables the interactive platform to ensure the security of feature value transfer tasks performed on the platform to a certain extent, by pre-assessing and predicting the probability of the target transferee's transfer event, thus providing advance warning to the transferor. Therefore, in one possible implementation, the method for obtaining task description information and task anomaly information can be as follows: first, obtain the task information of the feature value transfer task set of the interactive platform; then, based on the object identifier of the target transferee, match multiple feature value transfer tasks from the feature value transfer task set that include the target transferee; finally, obtain the task information of multiple feature value transfer tasks.

[0118] The task information of the feature value transfer task set of the aforementioned interactive platform refers to the set of feature value transfer tasks that occur on the interactive platform. Here, "occurrence" can mean that all or part of the steps in the feature value transfer task occur on the interactive platform. Accordingly, the feature value transfer task set can include task information corresponding to all or part of the steps in the feature value transfer task.

[0119] The interactive platform handles feature value transfer tasks. Task information includes task descriptions and exception information for the feature value transfer tasks processed by the platform. The platform includes two object types: the target transferee and the transferor. Both target transferees and transferors have corresponding object identifiers that uniquely identify different target transferees and transferors. Based on the target transferee's object identifier, multiple feature value transfer tasks involving the target transferee are matched from the feature value transfer task set; that is, feature value transfer tasks involving the target transferee are extracted. This process can be viewed as dividing the feature value transfer task set according to the target transferee. The result of this division is that feature value transfer tasks involving the same target transferee are grouped into one "category," which can include more than one transferor.

[0120] After the feature value transfer task is divided, the corresponding task information can be obtained based on the division results. This includes the task description information and task exception information processed by the interactive platform for the same target transferee. Subsequently, a descriptive feature for that target transferee is generated based on this task description information and task exception information. Based on this descriptive feature, the likelihood of that target transferee being involved in a transfer exception event is determined. This process enables the determination of the likelihood of target transferees within the interactive platform being involved in transfer exception events. This determination process is essentially a screening process for potential transfer exception events involving the interactive platform itself. By identifying target transferees with a high probability of involvement in transfer exception events, the interactive platform can provide early warnings for such target transferees, ensuring the security of feature value transfer tasks on the interactive platform.

[0121] The method described above for obtaining task description information and task anomaly information can determine the likelihood of anomalies involving the target transferor in a feature value transfer task involving an interactive platform. This method can determine the likelihood of anomalies involving the target transferors within the interactive platform itself, thereby screening for target transferors in the interactive platform that may be involved in anomalies. This is beneficial for the interactive platform to provide early warnings for such target transferors and, to a certain extent, ensure the security of the feature value transfer task performed by the interactive platform.

[0122] The aforementioned "acquiring task information from the feature value transfer task set of the interaction platform" includes task descriptions and exception information for the feature value transfer tasks processed by the interaction platform. Then, descriptive features are generated based on these task descriptions and exception information, and the likelihood of the target transferee being involved in transfer anomalies is determined based on these descriptive features. To improve the likelihood of proactively detecting potential transfer anomalies involving the target transferee and enhance the real-time nature of early warnings, it is possible to periodically determine the likelihood of transfer anomalies based on the target transferee's recent performance. Therefore, in one possible implementation, the method for acquiring task information is: periodically acquiring task information from the feature value transfer task set of the interaction platform based on the target duration.

[0123] The target duration refers to the time interval between the generation time of the task information for the feature value transfer task set of the interactive platform and the current time. For example, the target duration can be 1 day, 3 days, etc. The specific target duration can be set by those skilled in the art according to the actual situation and application scenario, and is not limited here. Each acquisition of task information is generated within the same target duration. That is to say, when it is necessary to determine the possibility of the target transferee being involved in transfer anomalies, task information generated within the target duration will be acquired. For example, assuming the target duration is 1 day, and the current time is February 22, then it is necessary to acquire the task information of the feature value transfer task set generated by the interactive platform on February 21. By setting the target duration, the acquired task information can have higher timeliness and be closer to the real-time situation of the target transferee. The possibility of the target transferee being involved in transfer anomalies determined based on this task information will also have higher reference value.

[0124] It should also be noted that, to further ensure the possibility of proactively detecting potential transfer anomalies involving the target transferee and improve the real-time nature of early warnings, in addition to setting a target duration as the time interval during task information acquisition, the acquisition method will also be determined as periodic acquisition. This periodic acquisition method allows for the periodic determination of the possibility of transfer anomalies involving the target transferee based on the target duration. Compared to random acquisition, this method is more systematic. This systematic determination of the possibility more accurately reflects the trend of the potential transfer anomalies involving the target transferee, thereby enabling the interactive platform to provide advance warnings for target transferees with the potential for transfer anomalies.

[0125] The method for obtaining task information described above employs a periodic acquisition approach based on a target duration. Setting the target duration ensures the acquired task information has higher timeliness, more closely reflecting the real-time situation of the target recipient. Simultaneously, the periodic approach allows for the identification of potential transfer anomalies involving the target recipient, reflecting trends in this likelihood. This improves the ability to proactively detect potential transfer anomalies involving the target recipient and enhances the real-time nature of the interactive platform's pre-emptive warnings to the transferor.

[0126] As mentioned in S202 above, "based on the task description information and the task anomaly information, a descriptive feature of the target transferee is generated." However, the task description information and task anomaly information contain a large amount of information. When the prediction model determines the probability that the target transferee is involved in a transfer anomaly event based on these descriptive features, it indiscriminately receives and learns from the numerous pieces of information in the descriptive features. In fact, there is data information in the descriptive features that is strongly correlated with transfer anomaly events. To improve the efficiency of the prediction model in determining the probability that the target transferee is involved in a transfer anomaly event, the weight of data information strongly correlated with transfer anomaly events in the generated descriptive information can be increased. Therefore, in one possible implementation, the method for generating descriptive features is: performing anomaly word matching on the task anomaly information based on an anomaly lexicon; responding to the inclusion of anomaly words from the anomaly lexicon in the task anomaly information, when generating the descriptive features of the target transferee based on the task description information and task anomaly information, increasing the weight of the anomaly words in the task anomaly information in the descriptive features.

[0127] The aforementioned abnormal terminology refers to a database composed of abnormal words. Abnormal words are terms used to reflect or express abnormal steps or situations in the feature value transfer task. Generally, the abnormal terminology is determined based on the abnormal transfer events. For example, the abnormal words may include: "overdue," "dishonest," "suspicious," and "abnormal feature value transfer," etc.

[0128] When generating descriptive features for the target recipient based on task description information and task anomaly information, the task anomaly information is used to identify abnormal steps in the feature value transfer task. Before generating descriptive features, the content of the task anomaly information is matched against an anomaly lexicon; this matching process is essentially a screening process for anomaly words appearing in the task anomaly information.

[0129] When anomaly words are identified in the task's abnormal information through matching with an anomaly terminology database, the weight of these abnormal words is increased in the descriptive features generated based on the task description and the anomaly information. For example, suppose the task's abnormal information includes the text "Target recipient A's delivery is seriously overdue." Matching the task's abnormal information with the anomaly terminology database reveals the presence of the abnormal word "overdue." Therefore, when generating the descriptive features of the target recipient based on the task description and the anomaly information, the weight of the abnormal word "overdue" is increased in the descriptive features. If the weight of regular words (excluding anomaly words) in the descriptive features is set to 0.1, then the weight of anomaly words is generally greater than the weight of regular words, such as 0.3. By increasing the weight of anomalous words in the descriptive features, the anomalous information related to the feature value transfer task corresponding to the anomalous words can be made significant in the descriptive features. In this way, when the prediction model determines the probability that the target transferee is involved in the transfer anomalous event based on the descriptive features, it can focus on learning the features corresponding to the anomalous information involved in the descriptive features, so as to improve the efficiency of determining the probability that the target transferee is involved in the transfer anomalous event.

[0130] The process of generating descriptive features can be implemented through the aforementioned feature fusion module. Specifically, this feature fusion module may include a fully connected layer, an activation function, and a Dropout (a regularization technique) layer. A fully connected layer is used to fuse features corresponding to image credentials and features corresponding to text information; the number of neurons is 256. ReLU is used as the activation function of the fully connected layer, and a Dropout layer with a ratio of 0.2 is added after the fully connected layer to prevent overfitting.

[0131] The method for generating descriptive features described above performs abnormal word matching on task anomaly information using an anomaly vocabulary library. When generating descriptive features, weights are added to the abnormal words in the task anomaly information. This weighting makes the feature information corresponding to the abnormal words more salient in the descriptive features. This allows the subsequent prediction model to selectively focus on learning the abnormal words when determining the probability of the target transferor being involved in transfer anomaly events based on the descriptive features, thus improving the efficiency of the prediction model in determining the probability of the target transferor being involved in transfer anomaly events.

[0132] The aforementioned anomaly lexicon includes anomalous words. The process of setting up the anomaly lexicon can be determined empirically; this empirical approach can be understood as a statistical summary of anomalous situations that occurred in historical feature value transfer tasks. To further improve the relevance and accuracy of the anomalous words in the lexicon, when the probability of the target transferee involving a transfer anomaly event is determined based on the prediction model, the anomaly lexicon can be generated by combining positive samples from the training samples used to train the prediction model in historical feature value transfer tasks. These positive samples correspond to historical transferees involving anomalous events. Therefore, in one possible implementation, the anomaly lexicon is generated as follows: keywords that meet the characteristics of anomalous events are determined based on positive samples, and the anomaly lexicon is generated based on these determined keywords.

[0133] The training process of the prediction model has been described in detail above. The training samples used for training include positive and negative samples. Positive samples correspond to historical transferees involved in anomalous transfer events, while negative samples correspond to historical transferees that do not. During training, the focus is on learning the features of samples from historical transferees involved in anomalous transfer events; therefore, these historical transferees are considered positive samples. Since the historical transferees in the positive samples involve anomalous transfer events, they include relevant descriptions of these events. Based on these positive samples, the representation of anomalous links or events in the feature value transfer task can be determined.

[0134] Anomaly representation conditions refer to the relevance conditions to the expression of anomalies in the feature value transfer task. For example, suppose the historical transferee in the positive sample is involved in an anomaly where, after receiving the feature value transferred from the transferor, the transferee failed to provide the corresponding product to the transferor. The corresponding conditions (i.e., anomaly representation conditions) related to this transfer anomaly are: feature value transfer, product provision, and overdue payment. Then, the keywords obtained from the positive sample that match the anomaly representation conditions for this historical transferee could be: overdue payment, product not provided, and large feature value transfer. An anomaly thesaurus can be generated based on these keywords.

[0135] The above-described method for generating an anomaly lexicon identifies keywords that meet the characteristics of anomaly events based on positive samples, and then generates an anomaly lexicon based on these keywords. By identifying keywords from positive samples involving anomaly events in the historical transfer process, the anomaly lexicon is constructed, ensuring a strong correlation between the anomaly words in the lexicon and the transferred anomaly events in the feature value transfer task. This improves the correlation between the anomaly words in the lexicon and the transferred anomaly events, as well as the accuracy of the acquisition.

[0136] Figure 8A flowchart of an event prediction method provided in an embodiment of this application is shown below. Figure 8 As shown, this event prediction method includes the following four steps:

[0137] S301: Periodically obtain multiple task description information and task exception information for a target duration.

[0138] Task description information generally identifies the basic information of the feature value transfer task, exemplified in the figure as additional information and industry characteristics. Task anomaly information is used to identify abnormal aspects in the feature value transfer task, exemplified in the figure as image credentials and text information, which may come from the sender's comments. There can be a correspondence between task description information and task anomaly information; both correspond to a single feature value transfer task. In this embodiment, task description information and task anomaly information for multiple feature value transfer tasks are typically acquired periodically for a target duration. The specific target duration can be determined by those skilled in the art based on the actual situation and application scenario. For example, a target duration of one day could be used. When it is necessary to determine the possibility of anomalies involving the target recipient, the task description information and task anomaly information from the previous day need to be retrieved.

[0139] S302: Obtain the task description information and task exception information of the target recipient.

[0140] To obtain the task description information and task exception information of the target transferee, it is necessary to match multiple feature value transfer tasks, including the target transferee, from the feature value transfer task set based on the object identifier of the target transferee in S301.

[0141] S303: Determine whether the image certificate has been formatted.

[0142] Once the task description information and task exception information from the target recipient are obtained, for the image credential in the task exception information, it is necessary to determine whether the image credential has been formatted. The purpose of formatting is to facilitate subsequent image OCR processing of the image credential. If the image feature has been formatted, proceed to S304; otherwise, proceed to S306 (i.e., directly use the text information to perform subsequent steps).

[0143] S304: Image OCR uses a CRNN model to recognize text information.

[0144] That is, for the obtained image voucher that has been formatted, the CRNN model is used to recognize the text information in the image voucher to obtain the text information corresponding to the image voucher.

[0145] S305: Format text information.

[0146] This involves formatting the text information in the task description and task exception information. Formatting facilitates subsequent analysis of the text information.

[0147] S306: Text feature preprocessing: Convert text information into vector representation.

[0148] The obtained task description information and task exception information are subjected to text feature preprocessing, that is, the text information is converted into vector representation. The text information in the task exception information includes the text information in the image certificate obtained according to S304 and the original text information in the task exception information.

[0149] S307: Generate descriptive features for the target recipient.

[0150] Text feature extraction for task anomaly information and task description information in vector form can be performed using a Bidirectional Long Short-Term Memory (BILSTM) network. By extracting text features from the task anomaly information, text features of the image text information and the text information itself can be obtained. Similarly, by extracting text features from the task description information, text features of the task description information can be obtained. By fusing the extracted text features, the descriptive features of the target recipient can be obtained. It should be noted that before preprocessing the text features of the task anomaly information and task description information, formatting processing can be performed on the image credentials and text information.

[0151] S308: Determine the likelihood that the target recipient is involved in an abnormal transfer event.

[0152] By inputting descriptive features into the XGBoost model (i.e., the prediction model), the model can generate predictions that determine the likelihood of the target being involved in anomaly events. The XGBoost model is derived using the supervised learning algorithm XGBoost.

[0153] In the foregoing Figure 1-8 Based on the corresponding embodiments, Figure 9 This is a schematic diagram of an event prediction device provided in an embodiment of the present application. The event prediction device 900 includes: an acquisition module 901, a generation module 902, and a determination module 903.

[0154] The acquisition module 901 is used to acquire task description information and task exception information of multiple feature value transfer tasks. The multiple feature value transfer tasks include at least one transferor and the same target transferee. The task exception information includes an image certificate provided by the transferor. The image certificate is an image that identifies the abnormal link of the corresponding feature value transfer task.

[0155] The generation module 902 is used to generate descriptive features of the target transferee based on the task description information and the task exception information;

[0156] The determining module 903 is used to determine the possibility that the target transferee is involved in an abnormal transfer event based on the information identified by the image certificate in the descriptive features.

[0157] In one possible implementation, the acquisition module 901 is used for:

[0158] Obtain comment information from the transferor regarding the target stage of the feature value transfer task, wherein the target stage is the abnormal stage in which a problem occurs during the completion of the feature value transfer task, and the comment information includes image credentials describing the problem.

[0159] The task anomaly information is determined based on the comment information.

[0160] In one possible implementation, the comment information acquired by the acquisition module 901 also includes text information describing the problem.

[0161] In one possible implementation, the acquisition module 901 is used for:

[0162] The comment information is obtained from the interactive platform used to process the feature value transfer task, where the sending party describes the target stage of the feature value transfer task.

[0163] or,

[0164] The target stage of the feature value transfer task for the transferor is obtained, and the conversation interface between the transferor and the target transferee is used as the comment information.

[0165] In one possible implementation, the acquisition module 901 is used for:

[0166] Obtain the image text information from the image certificate, and obtain the keywords used to describe the objective information in the feature value transfer task;

[0167] The image text information is preprocessed based on the keywords to remove non-objective information and obtain the task anomaly information.

[0168] In one possible implementation, the determining module 903 is used to:

[0169] Based on the information identified by the image credentials in the descriptive features, the possibility that the target transferee is involved in an abnormal transfer event is determined by a prediction model.

[0170] The prediction model is trained in the following manner:

[0171] Training samples for training are obtained based on historical feature value transfer tasks. The historical feature value transfer tasks include historical transferors and historical transferees. The training samples include sample task description information and sample task anomaly information of the historical transferee. The sample task anomaly information includes sample image credentials provided by the historical transferor. The sample image credentials are images that identify the abnormal links in the corresponding historical feature value transfer tasks.

[0172] The training samples are labeled to identify whether the corresponding historical transfer recipients are involved in transfer anomaly events. The training samples include positive samples and negative samples. The historical transfer recipients corresponding to the positive samples are involved in transfer anomaly events, while the historical transfer recipients corresponding to the negative samples are not involved in transfer anomaly events.

[0173] The initial prediction model is input based on the training samples, and the initial prediction model is trained based on the difference between the prediction results of the initial prediction model and the sample labels to obtain the prediction model.

[0174] In one possible implementation, the generation module 902 is used for:

[0175] The task anomaly information is matched with anomaly words based on an anomaly word library, which is determined based on the transfer anomaly event.

[0176] In response to the inclusion of abnormal words from the abnormal word library in the task abnormal information, when generating the descriptive features of the target transferee based on the task description information and the task abnormal information, the weight of the abnormal words in the task abnormal information in the descriptive features is increased.

[0177] In one possible implementation, the apparatus further includes a lexicon generation module, which, when the probability of the target transferee being involved in a transfer anomaly event is determined based on a prediction model, is used to:

[0178] Based on the positive samples, keywords that meet the criteria for representing abnormal events are identified;

[0179] The abnormal word library is generated based on the determined keywords.

[0180] In one possible implementation, the acquisition module 901 is used for:

[0181] Obtain task information of the feature value transfer task set of the interactive platform, wherein the interactive platform is used to process feature value transfer tasks, and the task information includes task description information and task exception information of the feature value transfer tasks processed by the interactive platform.

[0182] Based on the object identifier of the target transferee, multiple feature value transfer tasks, including the target transferee, are matched from the feature value transfer task set;

[0183] Obtain task information for multiple feature value transfer tasks.

[0184] In one possible implementation, the acquisition module 901 is used for:

[0185] The task information of the feature value transfer task set of the interactive platform is periodically acquired for a target duration, and the task information acquired each time is generated within the same target duration.

[0186] The aforementioned event prediction device, in order to effectively predict potential transfer anomalies in feature value transfer scenarios, acquires task description information and task anomaly information for multiple associated feature value transfer tasks for the target transferee. The task anomaly information includes image credentials provided by the transferor, which visually identify the abnormal stages involved in the feature value transfer task. Since feature value transfer tasks involve numerous stages and complex numerical and temporal information, it is difficult to directly and clearly describe the abnormal stages using only text. However, by taking screenshots of key interfaces of the abnormal stages, the complex information of the abnormal stages can be clearly demonstrated. By using the task anomaly information from multiple feature value transfer tasks related to the target transferee, the characteristics of the target transferee related to transfer anomalies can be more comprehensively expressed. The task description information identifies the basic information of these feature value transfer tasks. Combining the task description information and the task anomaly information, a comprehensive descriptive feature of the target transferee can be obtained, thereby determining the likelihood that the target transferee may be involved in transfer anomalies in the future, and thus enabling early warning based on this likelihood.

[0187] This application also provides a computer device, including a terminal device or a server, in which the aforementioned event prediction device can be configured. The computer device will now be described in conjunction with the accompanying drawings.

[0188] If the computer device is a terminal device, please refer to Figure 10 As shown, this application provides a terminal device, taking a mobile phone as an example:

[0189] Figure 10 The diagram shown is a block diagram of a portion of the structure of a mobile phone provided in an embodiment of this application. (Reference) Figure 10 The mobile phone includes components such as a radio frequency (RF) circuit 1410, a memory 1420, an input unit 1430, a display unit 1440, a sensor 1450, an audio circuit 1460, a Wi-Fi module 1470, a processor 1480, and a power supply 1490. Those skilled in the art will understand that... Figure 10 The mobile phone structure shown does not constitute a limitation on the mobile phone and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0190] The following combination Figure 10 A detailed introduction to each component of a mobile phone:

[0191] The RF circuit 1410 can be used to receive and transmit signals during information transmission or calls. In particular, it receives downlink information from the base station and processes it with the processor 1480; in addition, it transmits uplink data to the base station.

[0192] The memory 1420 can be used to store software programs and modules. The processor 1480 executes various mobile phone functions and data processing by running the software programs and modules stored in the memory 1420. The memory 1420 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a 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 mobile phone (such as audio data, phonebook, etc.). In addition, the memory 1420 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.

[0193] The input unit 1430 can be used to receive input numeric or character information, and to generate key signal inputs related to user settings and function control of the mobile phone. Specifically, the input unit 1430 may include a touch panel 1431 and other input devices 1432.

[0194] The display unit 1440 can be used to display information input by the user or information provided to the user, as well as various menus of the mobile phone. The display unit 1440 may include a display panel 1441.

[0195] The mobile phone may also include at least one sensor 1450, such as a light sensor, a motion sensor, and other sensors.

[0196] Audio circuitry 1460, speaker 1461, and microphone 1462 provide an audio interface between the user and the mobile phone.

[0197] WiFi is a short-range wireless transmission technology. Through the WiFi module 1470, mobile phones can help users send and receive emails, browse web pages, and access streaming media, providing users with wireless broadband internet access.

[0198] The processor 1480 is the control center of the mobile phone. It connects to various parts of the mobile phone through various interfaces and lines. It performs various functions of the mobile phone and processes data by running or executing software programs and / or modules stored in the memory 1420 and calling data stored in the memory 1420.

[0199] The mobile phone also includes a power supply 1490 (such as a battery) that powers the various components.

[0200] In this embodiment, the processor 1480 included in the terminal device is also used to execute the steps in the methods of the various embodiments of this application.

[0201] If the computer device is a server, this application embodiment also provides a server; please refer to [link to relevant documentation]. Figure 11 As shown, Figure 11 This is a structural diagram of a server 1500 provided in an embodiment of this application. The server 1500 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 1522 (e.g., one or more processors) and a memory 1532, and one or more storage media 1530 (e.g., one or more mass storage devices) for storing application programs 1542 or data 1544. The memory 1532 and storage media 1530 can be temporary or persistent storage. The program stored in the storage media 1530 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the server. Furthermore, the CPU 1522 may be configured to communicate with the storage media 1530 and execute the series of instruction operations in the storage media 1530 on the server 1500.

[0202] Server 1500 may also include one or more power supplies 1526, one or more wired or wireless network interfaces 1550, one or more input / output interfaces 1558, and / or one or more operating systems 1541, such as Windows Server. TM Mac OS X TM Unix TM Linux TMFreeBSD TM etc.

[0203] The steps performed by the server in the above embodiments can be based on Figure 11 The server structure shown.

[0204] In addition, this application embodiment also provides a storage medium for storing a computer program for executing the method provided in the above embodiment.

[0205] This application also provides a computer program product including a computer program, which, when run on a computer device, causes the computer device to perform the method provided in the above embodiments.

[0206] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium can be at least one of the following media: read-only memory (ROM), RAM, magnetic disk or optical disk, and other media that can store computer programs.

[0207] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0208] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0209] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Moreover, based on the implementation methods provided in the above aspects, this application can be further combined to provide more implementation methods. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An event prediction method, characterized in that, The method includes: Obtain task description information and task exception information for multiple feature value transfer tasks. The multiple feature value transfer tasks include at least one transferor and the same target transferee. The task exception information includes image credentials provided by the transferor. The image credentials are images that identify the abnormal links of the corresponding feature value transfer tasks. Based on the task description information and the task anomaly information, generate the descriptive features of the target transferee; Based on the information identified by the image credentials in the descriptive features, the likelihood that the target recipient is involved in an abnormal transfer event is determined.

2. The method according to claim 1, characterized in that, The task error information is obtained in the following way: Obtain comment information from the transferor regarding the target stage of the feature value transfer task, wherein the target stage is the abnormal stage in which a problem occurs during the completion of the feature value transfer task, and the comment information includes image credentials describing the problem. The task anomaly information is determined based on the comment information.

3. The method according to claim 2, characterized in that, The step of obtaining the comment information from the transferor regarding the target stage of the feature value transfer task includes: The comment information is obtained from the interactive platform used to process the feature value transfer task, where the sending party describes the target stage of the feature value transfer task. or, The target stage of the feature value transfer task for the transferor is obtained, and the conversation interface between the transferor and the target transferee is used as the comment information.

4. The method according to claim 2, characterized in that, The step of determining the task anomaly information based on the comment information includes: Obtain the image text information from the image certificate, and obtain the keywords used to describe the objective information in the feature value transfer task; The image text information is preprocessed based on the keywords to remove non-objective information and obtain the task anomaly information.

5. The method according to claim 1, characterized in that, The step of determining the likelihood that the target recipient is involved in an abnormal transfer event based on the information identified by the image credentials in the descriptive features includes: Based on the information identified by the image credentials in the descriptive features, the possibility that the target transferee is involved in an abnormal transfer event is determined by a prediction model. The prediction model is trained in the following manner: Training samples for training are obtained based on historical feature value transfer tasks. The historical feature value transfer tasks include historical transferors and historical transferees. The training samples include sample task description information and sample task anomaly information of the historical transferee. The sample task anomaly information includes sample image credentials provided by the historical transferor. The sample image credentials are images that identify the abnormal links in the corresponding historical feature value transfer tasks. The training samples are labeled to identify whether the corresponding historical transfer recipients are involved in transfer anomaly events. The training samples include positive samples and negative samples. The historical transfer recipients corresponding to the positive samples are involved in transfer anomaly events, while the historical transfer recipients corresponding to the negative samples are not involved in transfer anomaly events. The initial prediction model is input based on the training samples, and the initial prediction model is trained based on the difference between the prediction results of the initial prediction model and the sample labels to obtain the prediction model.

6. The method according to any one of claims 1-5, characterized in that, The step of generating descriptive features of the target recipient based on the task description information and the task anomaly information includes: The task anomaly information is matched with anomaly words based on an anomaly word library, which is determined based on the transfer anomaly event. In response to the inclusion of abnormal words from the abnormal word library in the task abnormal information, when generating the descriptive features of the target transferee based on the task description information and the task abnormal information, the weight of the abnormal words in the task abnormal information in the descriptive features is increased.

7. The method according to claim 6, characterized in that, When the probability of the target receiving party being involved in a transfer anomaly event is determined based on a prediction model, the method further includes: Based on the positive samples, keywords that meet the criteria for representing abnormal events are identified; The abnormal word library is generated based on the determined keywords.

8. The method according to claim 1, characterized in that, The task description information and task exception information for obtaining multiple feature value transfer tasks include: Obtain task information of the feature value transfer task set of the interactive platform, wherein the interactive platform is used to process feature value transfer tasks, and the task information includes task description information and task exception information of the feature value transfer tasks processed by the interactive platform. Based on the object identifier of the target transferee, multiple feature value transfer tasks, including the target transferee, are matched from the feature value transfer task set; Obtain task information for multiple feature value transfer tasks.

9. The method according to claim 8, characterized in that, The task information for obtaining the feature value transfer task set of the interactive platform includes: The task information of the feature value transfer task set of the interactive platform is periodically acquired for a target duration, and the task information acquired each time is generated within the same target duration.

10. The method according to claim 2, characterized in that, The comment information also includes text information used to describe the problem.

11. An event prediction device, characterized in that, The device includes: an acquisition module, a generation module, and a determination module; The acquisition module is used to acquire task description information and task exception information of multiple feature value transfer tasks. The multiple feature value transfer tasks include at least one transferor and the same target transferee. The task exception information includes an image certificate provided by the transferor. The image certificate is an image that identifies the abnormal link of the corresponding feature value transfer task. The generation module is used to generate descriptive features of the target recipient based on the task description information and the task exception information. The determining module is used to determine the likelihood that the target transferee is involved in an abnormal transfer event based on the information identified by the image credential in the descriptive features.

12. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs; The processor is configured to perform the method according to any one of claims 1-10 according to the computer program.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, which, when executed by a computer device, performs the method described in any one of claims 1-10.

14. A computer program product comprising a computer program, which, when run on a computer device, causes the computer device to perform the method of any one of claims 1-10.

Images