Business travel transaction abnormity identification method and system based on multi-modal data fusion

Abstract

The invention discloses a business travel transaction abnormity identification method and system based on multi-modal data fusion, and relates to the technical field of business travel transactions, and the method comprises the following steps: obtaining multi-modal target data in a business travel transaction scene; counting a target transmission path of the multi-modal target data in the transaction link; if the target transmission path is a single-node direct connection path, extracting historical single-node path data matched with the target transmission path from a historical business travel transaction database; calculating an abnormal association weight of each modal data in the historical single-node path data; performing exception primary judgment on the current multi-modal target data based on the exception association weight to obtain a primary judgment exception result; if the target transmission path is a multi-node transfer path, inputting the multi-modal target data into a multi-modal data fusion model to obtain an abnormal probability value of each modal data; the effect is of great significance for guaranteeing safe and standard operation of business travel transactions.

Description

Technical Field

[0001] This invention relates to the field of business travel transaction technology, and more specifically, to a method and system for identifying anomalies in business travel transactions using multimodal data fusion. Background Technology

[0002] In the business travel transaction sector, as business scale expands and transaction scenarios become increasingly complex, the importance of transaction anomaly identification is becoming increasingly prominent. Traditional methods do not provide targeted processing for transactions along different transmission paths. For transactions with complex paths involving multiple transit nodes, there is a lack of effective analytical tools, making it difficult to accurately identify potential anomalies. Traditional methods also lack the refinement to generate risk management strategies, failing to formulate differentiated control measures based on different anomaly identification results. This could lead to overly strict controls that disrupt normal transactions, or insufficient controls that fail to effectively prevent risks. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for identifying anomalies in business travel transactions through multimodal data fusion.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for identifying anomalies in business travel transactions using multimodal data fusion, comprising the following steps:

[0006] Acquire multimodal target data in business travel transaction scenarios;

[0007] Statistical analysis of the target transmission path of multimodal target data in the transaction chain;

[0008] If the target transmission path is a single-node direct connection path, extract historical single-node path data that matches the target transmission path from the historical business travel transaction database; calculate the abnormal association weight of each modality data in the historical single-node path data; and perform an initial anomaly judgment on the current multimodal target data based on the abnormal association weight to obtain the initial anomaly judgment result.

[0009] If the target transmission path is a multi-node relay path, the multimodal target data is input into the multimodal data fusion model to obtain the anomaly probability value of each modality; the feature matching degree between the multimodal target data and the feature database of illegal transactions in the business travel industry is calculated; a comprehensive anomaly score is calculated based on the anomaly probability value and the feature matching degree; when the comprehensive anomaly score is greater than the preset anomaly threshold, the anomaly identification result is output, otherwise the normal identification result is output.

[0010] Based on the initial assessment of abnormal results, abnormal identification results, or normal identification results, corresponding business travel transaction risk management strategies are generated.

[0011] Preferably, the multimodal target data includes transaction text data, transaction amount data, transaction time data, and user behavior profile data.

[0012] Preferably, the statistical analysis of the target transmission path of multimodal target data in the transaction link specifically includes the following steps:

[0013] Construct a transaction path tracing model;

[0014] Multimodal target data is input into the transaction link path tracing model to obtain the target transmission path; wherein, the target transmission path includes a single-node direct connection path or a multi-node relay path.

[0015] Preferably, calculating the anomaly association weights of each modality in the historical single-node path data specifically includes the following steps:

[0016] The Pearson correlation coefficient algorithm was used to calculate the correlation coefficient between historical data of each modality and historical transaction anomaly results in historical single-node path data;

[0017] The correlation coefficient is used as the weight for abnormal associations in the corresponding modality data.

[0018] Preferably, the preliminary anomaly judgment result is obtained by performing an anomaly assessment on the current multimodal target data based on the anomaly correlation weight, specifically including the following steps:

[0019] Each modality of the multimodal target data is compared with the corresponding historical normal data threshold range;

[0020] If a certain modality data exceeds the corresponding historical normal data threshold range, then the abnormal value of that modality is calculated based on the abnormal correlation weight of that modality data;

[0021] The total outlier value is obtained by summing the outliers of all modal data;

[0022] If the total number of outliers is greater than the preset initial judgment threshold, the initial judgment result of the outlier is output.

[0023] If the total abnormal value is less than or equal to the preset initial judgment threshold, then the initial judgment result is output as normal.

[0024] Preferably, the multimodal target data is input into a multimodal data fusion model to obtain the anomaly probability value of each modality, specifically including the following steps:

[0025] A multimodal data fusion model based on an attention mechanism is constructed; wherein, the multimodal data fusion model includes a text modality processing sub-model, a numerical modality processing sub-model, and a profile modality processing sub-model;

[0026] The transaction text data is input into the text modality processing sub-model to obtain the text modality anomaly probability value;

[0027] The transaction amount data and transaction time data are input into the numerical mode processing sub-model to obtain the numerical mode anomaly probability value;

[0028] User behavior profile data is input into the profile modality processing sub-model to obtain profile modality anomaly probability values;

[0029] The text modality anomaly probability value, the numerical modality anomaly probability value, and the portrait modality anomaly probability value together constitute the anomaly probability value of each modality data.

[0030] Preferably, calculating the feature matching degree between the multimodal target data and the feature database of illegal transactions in the business travel industry specifically includes the following steps:

[0031] The database of illegal transactions in the business travel industry is obtained from the business travel industry supervision platform. The database includes textual features of illegal transactions, fluctuations in illegal amounts, distribution of illegal time periods, and characteristics of illegal user behavior.

[0032] The cosine similarity algorithm is used to calculate the text similarity between the transaction text data and the illegal transaction text features, the amount similarity between the transaction amount data and the illegal amount fluctuation features, the time similarity between the transaction time data and the illegal time distribution features, and the behavior similarity between the user behavior profile data and the illegal user behavior features.

[0033] Calculate the average of text similarity, monetary similarity, time similarity, and behavioral similarity, and use this average as the feature matching degree.

[0034] Preferably, a corresponding business travel transaction risk management strategy is generated based on the initial judgment of anomalies, anomaly identification results, or normal identification results, specifically including the following steps:

[0035] If the initial judgment of an abnormal situation is an abnormal situation, a first risk control strategy is generated. The first risk control strategy includes suspending the current transaction, sending an abnormal verification notification to the user, and manually reviewing the transaction information.

[0036] If the result is an anomaly identification, a second risk control strategy is generated. The second risk control strategy includes freezing the transaction account, recording the illegal transaction record, and synchronizing the abnormal data to the business travel industry supervision platform.

[0037] If the identification result is normal, a third risk control strategy is generated. The third risk control strategy includes continuing to complete the transaction, retaining the transaction data in the historical database, and updating the user behavior profile.

[0038] A multimodal data fusion-based business travel transaction anomaly identification system includes:

[0039] Acquisition module: Acquires multimodal target data in business travel transaction scenarios;

[0040] Statistics module: Analyzes the target transmission path of multimodal target data within the transaction chain;

[0041] First processing module: If the target transmission path is a single-node direct connection path, extract historical single-node path data that matches the target transmission path from the historical business travel transaction database; calculate the abnormal association weight of each modality data in the historical single-node path data; and perform an initial anomaly judgment on the current multimodal target data based on the abnormal association weight to obtain the initial anomaly judgment result.

[0042] The second processing module: If the target transmission path is a multi-node relay path, the multimodal target data is input into the multimodal data fusion model to obtain the abnormal probability value of each modality; the feature matching degree between the multimodal target data and the feature database of illegal transactions in the business travel industry is calculated; a comprehensive abnormal score is calculated based on the abnormal probability value and the feature matching degree; when the comprehensive abnormal score is greater than the preset abnormal threshold, the abnormal identification result is output, otherwise the normal identification result is output.

[0043] Control module: Generates corresponding business travel transaction risk control strategies based on the initial judgment of abnormal results, abnormal identification results, or normal identification results.

[0044] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements a multimodal data fusion method for identifying anomalies in business travel transactions.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] This invention fully acquires multimodal target data in business travel transaction scenarios, covering transaction text, amount, time, and user behavior profiles. It characterizes transactions from multiple dimensions, enabling a more comprehensive capture of transaction features and providing a rich and accurate information foundation for subsequent anomaly identification. Regarding transaction path differentiation, it employs differentiated identification strategies for two different target transmission paths: single-node direct connection paths and multi-node transit paths. For single-node direct connection paths, it extracts matching historical single-node path data from the historical business travel transaction database and calculates the anomaly correlation weights of each modality for initial anomaly judgment. This method fully utilizes the empirical value of historical data, enabling rapid preliminary anomaly screening of relatively simple single-node direct connection transaction paths. For multi-node transit paths, it uses a multimodal data fusion model to obtain the anomaly probability values ​​of each modality and combines this with the feature matching degree of the business travel industry's violation transaction feature database to calculate a comprehensive anomaly score. This fusion model and industry feature database approach improves the accuracy and professionalism of anomaly identification for complex multi-node transit transaction paths, enabling more precise identification of deeply hidden abnormal transactions. Based on different identification results, corresponding risk management strategies for business travel transactions are generated, achieving precise and differentiated risk management. This method comprehensively improves the accuracy of anomaly identification in business travel transactions and the targeting of risk management, which is of great significance for ensuring the safe and standardized operation of business travel transactions. Attached Figure Description

[0047] Figure 1 This invention provides a schematic diagram illustrating the steps of a multimodal data fusion method for identifying anomalies in business travel transactions.

[0048] Figure 2 This invention presents a schematic diagram of a multimodal data fusion-based business travel transaction anomaly identification system.

[0049] Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention.

[0050] 610. Processor; 620. Communication interface; 630. Memory; 640. Communication bus. Detailed Implementation

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0052] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0054] Reference Figures 1-3 .

[0055] The embodiments further illustrate the multimodal data fusion-based business travel transaction anomaly identification method and system proposed in this invention.

[0056] A method for identifying anomalies in business travel transactions using multimodal data fusion, comprising the following steps:

[0057] Acquire multimodal target data in business travel transaction scenarios;

[0058] Statistical analysis of the target transmission path of multimodal target data in the transaction chain;

[0059] If the target transmission path is a single-node direct connection path, extract historical single-node path data that matches the target transmission path from the historical business travel transaction database; calculate the abnormal correlation weight of each modality data in the historical single-node path data; and perform an initial anomaly judgment on the current multimodal target data based on the abnormal correlation weight to obtain the initial anomaly judgment result.

[0060] If the target transmission path is a multi-node relay path, the multimodal target data is input into the multimodal data fusion model to obtain the anomaly probability value of each modality; the feature matching degree between the multimodal target data and the feature database of illegal transactions in the business travel industry is calculated; a comprehensive anomaly score is calculated based on the anomaly probability value and the feature matching degree; when the comprehensive anomaly score is greater than the preset anomaly threshold, the anomaly identification result is output, otherwise the normal identification result is output.

[0061] Based on the initial assessment of abnormal results, abnormal identification results, or normal identification results, corresponding business travel transaction risk management strategies are generated.

[0062] Multimodal target data includes transaction text data, transaction amount data, transaction time data, and user behavior profile data.

[0063] Transaction text data reflects the specific content and descriptive information of a transaction, such as the transacting party and the services involved. This textual information may contain clues to abnormal transactions, such as unusual descriptive vocabulary. Transaction amount data is a crucial quantitative indicator for determining whether a transaction is abnormal; abnormal transactions are often accompanied by significant fluctuations in amount or values ​​exceeding normal ranges. Transaction time data reveals the temporal patterns of transactions. Normal transactions typically occur within a reasonable timeframe, while abnormal transactions may occur outside of working hours or during uncommon trading periods. User behavior profile data provides a comprehensive representation of a user's past transaction behaviors and habits. By comparing the current transaction with the user's historical behavior profile, it's possible to effectively determine whether the transaction conforms to the user's regular behavioral patterns; deviations may indicate anomalies.

[0064] In the subsequent anomaly identification process, the multimodal target data will be combined with the target transmission path factors of the transaction link and processed in different ways. For example, for single-node direct connection paths and multi-node transit paths, the method of matching historical data to calculate the anomaly association weight and inputting multimodal data fusion model to calculate the anomaly probability value and feature matching degree will be adopted respectively. In the end, the accurate identification of whether business travel transactions are abnormal will be achieved, and corresponding risk control strategies will be generated accordingly.

[0065] The statistical analysis of the target transmission path of multimodal target data in the transaction chain includes the following steps:

[0066] Construct a transaction path tracing model;

[0067] Multimodal target data is input into the transaction link path tracing model to obtain the target transmission path; wherein, the target transmission path includes a single-node direct connection path or a multi-node relay path.

[0068] The transaction link path tracing model aims to accurately capture the flow trajectory and correlation logic of multimodal target data in the transaction link within a business travel transaction scenario. It integrates the connection relationships and data interaction rules of each node in the transaction system to form an algorithm and structural system capable of simulating the data transmission process.

[0069] Multimodal target data is input into the transaction path tracing model. The model performs in-depth analysis of this multi-dimensional information, including transaction text, amount, time, and user behavior profiles, and tracks the data's transmission at each step in the transaction path along the pre-defined algorithm logic.

[0070] The transaction link path tracing model outputs the target transmission path. If data is transmitted directly from one node to another without being forwarded by other intermediate nodes, then it is a single-node direct path. If the data passes through two or more intermediate nodes during transmission, then it is a multi-node relay path. This method can clearly and accurately determine the specific transmission path of multimodal target data in the transaction link, laying the foundation for subsequent transaction anomaly identification work for different path types.

[0071] Calculating the anomaly association weights of each modality in historical single-node path data includes the following steps:

[0072] The Pearson correlation coefficient algorithm was used to calculate the correlation coefficient between historical data of each modality and historical transaction anomaly results in historical single-node path data;

[0073] The correlation coefficient is used as the weight for abnormal associations in the corresponding modality data.

[0074] The Pearson correlation coefficient measures the strength of the linear correlation between two variables. Here, for each modality of historical data in the historical single-node path data, the correlation coefficient between them and historical transaction anomaly results is calculated separately. This calculation clarifies the degree of association between historical data of different modalities and the result of whether a transaction is abnormal. The obtained correlation coefficients are determined as the anomaly association weights of the corresponding modal data. In other words, the higher the correlation coefficient between a certain modality of data and historical transaction anomaly results, the higher the importance of that modality of data in judging whether a transaction is abnormal, and the greater its anomaly association weight. This provides a key weighting basis for subsequent preliminary judgment of transaction anomalies based on each modality of data.

[0075] The initial anomaly assessment of the current multimodal target data is performed based on anomaly correlation weights to obtain an initial anomaly result, specifically including the following steps:

[0076] Each modal data of the multimodal target data is compared with the threshold range of the historical normal data of the corresponding modality;

[0077] If a certain modality data exceeds the corresponding historical normal data threshold range, then the abnormal value of that modality is calculated based on the abnormal correlation weight of that modality data;

[0078] The total outlier value is obtained by summing the outliers of all modal data;

[0079] If the total number of outliers exceeds the preset initial threshold, the initial outlier result will be output.

[0080] If the total outlier is less than or equal to the preset initial threshold, the initial judgment result is output as normal.

[0081] First, each modality in the multimodal target data is compared with the corresponding historical normal data threshold range. The purpose of this step is to initially screen out modality data that may be abnormal, because the historical normal data threshold range is based on a summary of a large number of normal transaction situations. If the current modality data exceeds this range, it is suspected of being abnormal.

[0082] If a modality of data is found to exceed the corresponding historical normal data threshold range, the outlier value for that modality is calculated based on its anomaly correlation weight. This anomaly correlation weight reflects the strength of the correlation between the modality's historical data and historical abnormal transaction results. A higher weight means a greater impact of the modality's data on whether a transaction is abnormal. Therefore, when calculating outliers, modal data with high weights exceeding the normal range will produce outliers that more accurately reflect the degree of abnormality.

[0083] The total outlier is obtained by summing up the outliers of all modal data. This step comprehensively considers each modal data that may contain anomalies and summarizes their degree of anomaly to form an anomaly measurement index.

[0084] The total outlier value is compared with the preset initial judgment threshold. If the total outlier value is greater than the preset initial judgment threshold, an initial judgment of anomaly is output, indicating that the current transaction is likely to be abnormal; if the total outlier value is less than or equal to the preset initial judgment threshold, an initial judgment of normal is output, indicating that the current transaction is normal under the initial judgment.

[0085] The multimodal target data is input into a multimodal data fusion model to obtain the anomaly probability value of each modality. The specific steps include:

[0086] Construct a multimodal data fusion model based on an attention mechanism; the multimodal data fusion model includes a text modality processing sub-model, a numerical modality processing sub-model, and a profile modality processing sub-model;

[0087] The transaction text data is input into the text modality processing sub-model to obtain the text modality anomaly probability value;

[0088] The transaction amount data and transaction time data are input into the numerical mode processing sub-model to obtain the numerical mode anomaly probability value;

[0089] User behavior profile data is input into the profile modality processing sub-model to obtain profile modality anomaly probability values;

[0090] Among them, the text modality anomaly probability value, the numerical modality anomaly probability value, and the portrait modality anomaly probability value together constitute the anomaly probability value of each modality data.

[0091] Attention mechanisms enable models to automatically focus on information more critical to identifying transaction anomalies when processing multimodal data, enhancing their ability to capture important features. This multimodal data fusion model comprises three sub-models: a text modality processing sub-model, a numerical modality processing sub-model, and a profile modality processing sub-model, each specifically designed for processing different types of modal data.

[0092] The transaction text data is input into the text modality processing sub-model. This sub-model conducts in-depth analysis of the content of the transaction text, such as identifying whether there are any illegal or abnormal expressions in the text, like non-compliant transaction project descriptions or abnormal transaction object names. By extracting and analyzing these text features, a text modality anomaly probability value is obtained, which reflects the likelihood of anomalies in the transaction text.

[0093] The transaction amount and transaction time data are input into the numerical modality processing sub-model. This sub-model considers abnormal fluctuations in the amount, such as whether the amount exceeds the normal transaction range or whether there is an abnormal trend in the amount change. It also takes into account whether the time falls within a period when illegal transactions often occur, such as large transactions outside of working hours. After comprehensive calculation, a numerical modality anomaly probability value is obtained, which reflects the possibility of anomalies in the transaction amount and time dimensions.

[0094] User behavior profile data is input into the profile modality processing sub-model. This sub-model meticulously compares the user's current behavior with historical normal behavior profiles to determine if there are any abnormal behavior patterns. For example, if a user's past transaction amounts are usually within a certain range, but the current transaction amount deviates significantly; or if a user's past transaction times are mostly on weekdays, but the current transaction time is frequently on non-weekdays, etc. Through such comparative analysis, the profile modality anomaly probability value is obtained, reflecting the possibility of abnormal user behavior.

[0095] The anomaly probability values ​​of text modality, numerical modality, and profile modality together constitute the anomaly probability values ​​of each modality of data. These values ​​comprehensively reflect the possibility of anomalies in various aspects of the transaction from different dimensions such as transaction text, amount and time, and user behavior profile, providing an important basis for subsequent comprehensive judgment on whether the transaction is abnormal.

[0096] Calculating the feature matching degree between multimodal target data and the feature database of illegal transactions in the business travel industry includes the following steps:

[0097] The database of features of illegal transactions in the business travel industry was obtained from the business travel industry supervision platform. The database includes features of illegal transaction text, fluctuations in illegal amounts, distribution of illegal time, and characteristics of illegal user behavior.

[0098] The cosine similarity algorithm is used to calculate the text similarity between transaction text data and illegal transaction text features, the amount similarity between transaction amount data and illegal amount fluctuation features, the time similarity between transaction time data and illegal time distribution features, and the behavior similarity between user behavior profile data and illegal user behavior features.

[0099] Calculate the average of text similarity, monetary similarity, time similarity, and behavioral similarity, and use this average as the feature matching degree.

[0100] This application obtains a database of features for illegal business travel transactions from a business travel industry regulatory platform. This database includes features of illegal transaction text, fluctuations in illegal transaction amounts, distribution of illegal transaction time, and characteristics of illegal user behavior. These features are derived from in-depth analysis and summarization of numerous cases of illegal business travel transactions, accurately depicting the typical manifestations of illegal transactions across different dimensions such as text, amount, time, and user behavior, providing a standard basis for subsequent feature matching.

[0101] Cosine similarity algorithm is used to calculate the similarity between different types of data and corresponding violation features. For transaction text data, text similarity is calculated between it and the features of violation transaction text. This calculation determines the degree of semantic and expressive fit between the transaction text content and the features of violation text; a higher fit indicates a greater likelihood of violation. For transaction amount data, amount similarity is calculated between it and the fluctuation features of violation amounts. This measures the volatility of transaction amounts, such as whether the size and frequency of changes in amount match the fluctuation patterns of violation amounts. A high amount similarity increases the probability of violation in terms of transaction amount. For transaction time data, time similarity is calculated between it and the time distribution features of violation times. This checks whether the time of the transaction matches the time distribution of frequent violation transactions, such as whether transactions were conducted outside of working hours or during peak periods of violation transactions, such as special holidays. A high time similarity increases the probability of violation risk in the time dimension. For user behavior profile data, behavioral similarity is calculated between it and the behavioral features of violation users. This determines whether the user's current transaction behavior, such as transaction frequency and choice of transaction objects, is consistent with the behavioral features of violation users. A high behavioral similarity indicates that the user's behavior may be in violation.

[0102] The average of four similarities—text similarity, amount similarity, time similarity, and behavior similarity—is calculated and used as the feature matching score. This feature matching score comprehensively reflects the matching degree between the transaction and the violation characteristics across the four dimensions of text, amount, time, and user behavior. It can comprehensively reflect the overall matching degree between the current multimodal target data and the characteristics of illegal transactions in the business travel industry, providing an important basis for subsequent calculation of a comprehensive anomaly score based on the anomaly probability value, thereby determining whether the transaction is abnormal.

[0103] Based on the initial assessment of abnormal results, or the results of abnormal identification or normal identification, a corresponding business travel transaction risk management strategy is generated, which specifically includes the following steps:

[0104] If the initial judgment of abnormality is a preliminary abnormal situation, a first risk control strategy is generated. The first risk control strategy includes suspending the current transaction, sending an abnormality verification notification to the user, and manually reviewing the transaction information.

[0105] If the result is an anomaly identification, a second risk control strategy is generated. The second risk control strategy includes freezing the transaction account, recording the illegal transaction record, and synchronizing the abnormal data to the business travel industry supervision platform.

[0106] If the identification result is normal, a third risk control strategy will be generated. The third risk control strategy includes continuing to complete the transaction, retaining the transaction data in the historical database, and updating the user behavior profile.

[0107] This application generates corresponding risk management strategies based on different transaction identification results. When the identification result indicates an initial judgment of an anomaly, a first risk management strategy is generated. This strategy includes suspending the current transaction to prevent potentially abnormal transactions from continuing and causing further losses; simultaneously sending an anomaly verification notification to the user, allowing the user to confirm and explain the transaction situation; and conducting manual review of the transaction information to accurately determine whether the transaction is truly abnormal through professional human review.

[0108] If the identification result is an anomaly, a second risk control strategy is generated. This strategy includes freezing the transaction account to prevent the funds in the account from being misused due to illegal transactions; recording illegal transaction records to provide a basis for subsequent supervision and inquiries; and synchronizing the abnormal data to the business travel industry supervision platform so that industry regulators can promptly grasp the situation of illegal transactions and conduct broader industry supervision.

[0109] If the identification result is normal, a third risk management strategy is generated. This strategy covers: continuing the transaction to ensure its smooth progress; retaining transaction data in a historical database for future analysis and querying; and updating user behavior profiles to reflect their latest transaction behavior, providing a more accurate basis for subsequent transaction risk identification. Through this differentiated risk management strategy, transaction risks can be effectively prevented while ensuring the smooth operation of transactions and the dynamic updating of user behavior profiles.

[0110] A multimodal data fusion-based business travel transaction anomaly identification system includes:

[0111] Acquisition module: Acquires multimodal target data in business travel transaction scenarios;

[0112] Statistics module: Analyzes the target transmission path of multimodal target data within the transaction chain;

[0113] First processing module: If the target transmission path is a single-node direct connection path, extract historical single-node path data that matches the target transmission path from the historical business travel transaction database; calculate the abnormal correlation weight of each modality data in the historical single-node path data; and perform an initial anomaly judgment on the current multimodal target data based on the abnormal correlation weight to obtain the initial anomaly judgment result.

[0114] The second processing module: If the target transmission path is a multi-node relay path, the multimodal target data is input into the multimodal data fusion model to obtain the anomaly probability value of each modality; the feature matching degree between the multimodal target data and the feature database of illegal transactions in the business travel industry is calculated; a comprehensive anomaly score is calculated based on the anomaly probability value and the feature matching degree; when the comprehensive anomaly score is greater than the preset anomaly threshold, the anomaly identification result is output, otherwise the normal identification result is output.

[0115] Control module: Generates corresponding business travel transaction risk control strategies based on the initial judgment of abnormal results, abnormal identification results, or normal identification results.

[0116] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a multimodal data fusion method for identifying anomalies in business travel transactions.

[0117] like Figure 3 As shown, the electronic device may include a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a multimodal data fusion method for identifying anomalies in business travel transactions.

[0118] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0119] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute a multimodal data fusion method for identifying anomalies in business travel transactions.

[0120] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform a multimodal data fusion method for identifying anomalies in business travel transactions.

[0121] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. 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 this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0122] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying anomalies in business travel transactions through multimodal data fusion, characterized in that, The method includes the following steps: Acquire multimodal target data in business travel transaction scenarios; Statistical analysis of the target transmission path of multimodal target data in the transaction chain; If the target transmission path is a single-node direct connection path, extract historical single-node path data that matches the target transmission path from the historical business travel transaction database; calculate the abnormal association weight of each modality data in the historical single-node path data; and perform an initial anomaly judgment on the current multimodal target data based on the abnormal association weight to obtain the initial anomaly judgment result. If the target transmission path is a multi-node relay path, the multimodal target data is input into the multimodal data fusion model to obtain the anomaly probability value of each modality; the feature matching degree between the multimodal target data and the feature database of illegal transactions in the business travel industry is calculated; a comprehensive anomaly score is calculated based on the anomaly probability value and the feature matching degree; when the comprehensive anomaly score is greater than the preset anomaly threshold, the anomaly identification result is output, otherwise the normal identification result is output. Based on the initial assessment of abnormal results, abnormal identification results, or normal identification results, corresponding business travel transaction risk management strategies are generated.

2. The method for identifying anomalies in business travel transactions based on multimodal data fusion according to claim 1, characterized in that, The multimodal target data includes transaction text data, transaction amount data, transaction time data, and user behavior profile data.

3. The method for identifying anomalies in business travel transactions based on multimodal data fusion according to claim 2, characterized in that, The statistical analysis of the target transmission path of multimodal target data in the transaction chain includes the following steps: Construct a transaction path tracing model; Multimodal target data is input into the transaction link path tracing model to obtain the target transmission path; wherein, the target transmission path includes a single-node direct connection path or a multi-node relay path.

4. The method for identifying anomalies in business travel transactions based on multimodal data fusion according to claim 3, characterized in that, Calculating the anomaly association weights of each modality in historical single-node path data includes the following steps: The Pearson correlation coefficient algorithm was used to calculate the correlation coefficient between historical data of each modality and historical transaction anomaly results in historical single-node path data; The correlation coefficient is used as the weight for abnormal associations in the corresponding modality data.

5. The method for identifying anomalies in business travel transactions based on multimodal data fusion according to claim 4, characterized in that, Based on the aforementioned anomaly correlation weights, an initial anomaly judgment is performed on the current multimodal target data to obtain an initial anomaly result, specifically including the following steps: Each modality of the multimodal target data is compared with the corresponding historical normal data threshold range; If a certain modality data exceeds the corresponding historical normal data threshold range, then the abnormal value of that modality is calculated based on the abnormal correlation weight of that modality data; The total outlier value is obtained by summing the outliers of all modal data; If the total number of outliers is greater than the preset initial judgment threshold, the initial judgment result of the outlier is output. If the total abnormal value is less than or equal to the preset initial judgment threshold, then the initial judgment result is output as normal.

6. The method for identifying anomalies in business travel transactions based on multimodal data fusion according to claim 5, characterized in that, The multimodal target data is input into a multimodal data fusion model to obtain the anomaly probability value of each modality. The specific steps include: A multimodal data fusion model based on an attention mechanism is constructed; wherein, the multimodal data fusion model includes a text modality processing sub-model, a numerical modality processing sub-model, and a profile modality processing sub-model; The transaction text data is input into the text modality processing sub-model to obtain the text modality anomaly probability value; The transaction amount data and transaction time data are input into the numerical mode processing sub-model to obtain the numerical mode anomaly probability value; User behavior profile data is input into the profile modality processing sub-model to obtain profile modality anomaly probability values; The text modality anomaly probability value, the numerical modality anomaly probability value, and the portrait modality anomaly probability value together constitute the anomaly probability value of each modality data.

7. The method for identifying anomalies in business travel transactions based on multimodal data fusion according to claim 6, characterized in that, Calculating the feature matching degree between multimodal target data and the feature database of illegal transactions in the business travel industry includes the following steps: The database of illegal transactions in the business travel industry is obtained from the business travel industry supervision platform. The database includes textual features of illegal transactions, fluctuations in illegal amounts, distribution of illegal time periods, and characteristics of illegal user behavior. The cosine similarity algorithm is used to calculate the text similarity between the transaction text data and the illegal transaction text features, the amount similarity between the transaction amount data and the illegal amount fluctuation features, the time similarity between the transaction time data and the illegal time distribution features, and the behavior similarity between the user behavior profile data and the illegal user behavior features. Calculate the average of text similarity, monetary similarity, time similarity, and behavioral similarity, and use this average as the feature matching degree.

8. The method for identifying anomalies in business travel transactions based on multimodal data fusion according to claim 7, characterized in that, Based on the initial assessment of abnormal results, or the results of abnormal identification or normal identification, a corresponding business travel transaction risk management strategy is generated, which specifically includes the following steps: If the initial judgment of an abnormal situation is an abnormal situation, a first risk control strategy is generated. The first risk control strategy includes suspending the current transaction, sending an abnormal verification notification to the user, and manually reviewing the transaction information. If the result is an anomaly identification, a second risk control strategy is generated. The second risk control strategy includes freezing the transaction account, recording the illegal transaction record, and synchronizing the abnormal data to the business travel industry supervision platform. If the identification result is normal, a third risk control strategy is generated. The third risk control strategy includes continuing to complete the transaction, retaining the transaction data in the historical database, and updating the user behavior profile.

9. A multimodal data fusion-based business travel transaction anomaly identification system, applied to the multimodal data fusion-based business travel transaction anomaly identification method according to any one of claims 1-8, characterized in that, include: Acquisition module: Acquires multimodal target data in business travel transaction scenarios; Statistics module: Analyzes the target transmission path of multimodal target data within the transaction chain; First processing module: If the target transmission path is a single-node direct connection path, extract historical single-node path data that matches the target transmission path from the historical business travel transaction database; calculate the abnormal correlation weight of each modal data in the historical single-node path data. Based on the aforementioned anomaly correlation weights, an initial anomaly judgment is performed on the current multimodal target data to obtain an initial anomaly judgment result; The second processing module: If the target transmission path is a multi-node relay path, the multimodal target data is input into the multimodal data fusion model to obtain the anomaly probability value of each modality; the feature matching degree between the multimodal target data and the feature database of illegal transactions in the business travel industry is calculated; and a comprehensive anomaly score is calculated based on the anomaly probability value and the feature matching degree. When the overall anomaly score is greater than the preset anomaly threshold, the anomaly identification result is output; otherwise, the normal identification result is output. Control module: Generates corresponding business travel transaction risk control strategies based on the initial judgment of abnormal results, abnormal identification results, or normal identification results.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the multimodal data fusion method for identifying anomalies in business travel transactions as described in any one of claims 1 to 8.

Images