Method and device for automatically identifying suspicious bank transaction characteristics

Abstract

The present invention discloses a method and device for automatically identifying suspicious transaction features of a bank. Bank transaction records are collected and the text in the records is standardized. The processing content includes common transaction types such as transfers, consumption, and ATM cash withdrawals. Then, the standardized text is segmented, key information words are extracted, and a vocabulary is constructed. The vocabulary is used for subsequent Word2Vec model training. The Word2Vec model adopts the Skip‑Gram architecture, and generates word vectors by predicting context words to capture the semantic relationship between transaction feature words, and uses negative sampling methods to reduce the amount of model calculation. The maximum log-likelihood function is used as the loss function during the training process to optimize the model parameters. After the training is completed, the recognition accuracy of the model for suspicious transaction features is evaluated by verifying the generated word vectors, and the model parameters are further adjusted according to the verification results to obtain a more accurate suspicious transaction identification model.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and device for automatically identifying suspicious transaction characteristics of a bank. Background Art

[0002] In existing technologies, banks detect suspicious transactions by monitoring and analyzing transaction records. This typically involves analyzing factors such as transaction amount, frequency, time, and location. Natural language processing technology, particularly that based on word embedding models, has been gradually applied in the financial sector to identify and extract features related to text descriptions in transaction records. The Word2Vec model, a commonly used word embedding method, can capture the semantic relationships between words by training on large-scale text data, demonstrating high accuracy in feature extraction and pattern recognition.

[0003] However, existing natural language processing models and methods still have several shortcomings. First, the standardization and word segmentation processes lack specific treatment for financial-specific vocabulary, resulting in inaccurate extraction of transaction features. Second, the model's training efficiency is low when processing massive amounts of transaction data, especially when the vocabulary size and context window settings are affected, which significantly increases the model's computational complexity. Furthermore, the model lacks a dynamic adjustment mechanism, and parameters cannot be optimized promptly based on model performance during the verification phase, limiting its adaptability and flexibility in different transaction scenarios.

[0004] Therefore, there is an urgent need to provide an improved automatic identification method for suspicious transaction features in banks to improve the accuracy of identifying suspicious transaction features. Summary of the Invention

[0005] The present application provides a method and device for automatically identifying suspicious transaction characteristics of a bank, so as to improve the accuracy of identifying suspicious transaction characteristics.

[0006] This application provides a method for automatically identifying suspicious bank transaction characteristics, including:

[0007] Collect bank transaction records and standardize the text in the transaction records; the transaction records include transfers, purchases, ATM withdrawals, over-the-counter withdrawals, financial products, precious metals, and loan records; the standardization includes standardizing date formats, unit amounts, and removing non-text symbols;

[0008] Performing word segmentation on the text in the standardized transaction records to extract key information words; constructing a vocabulary based on the key information words; wherein the vocabulary includes words that reflect transaction characteristics and is used for subsequent model training;

[0009] The Word2Vec model is trained using the word segmentation results of the text in the transaction record and the constructed vocabulary. The Word2Vec model is implemented using the Skip-Gram model architecture. For a given central word in the transaction record text, the Skip-Gram model is trained to generate word vectors by predicting its context words to capture the semantic relationship between words. The probability of each context word occurring is calculated according to the following formula (1):

[0010]

[0011] in, Indicates the central word, represents the context word, where Indicates the position of the central word in the sentence. Represents context-dependent indexes; Representing context words word vectors; Indicates the central word word vectors; A word vector representing each word in the vocabulary; is the size of the vocabulary;

[0012] The Word2Vec model is trained using a negative sampling method, wherein the negative sampling method randomly selects words that are not related to the central word as negative samples to reduce the amount of computation required for the model to predict all words. The loss function provided by the following formula (2) is used during the training process:

[0013]

[0014] in, Represents the loss function, which is used to measure the prediction error of the model during training; Represents the total number of samples in the training data; Indicates the window size; Representing context words word vectors; Indicates the central word word vectors; Indicates the total number of negative samples; Represents the distribution of negative samples Randomly sample negative samples The expected value of probability; Represents negative samples word vectors; is the sigmoid function;

[0015] After training is completed, the generated word vectors are verified to evaluate the accuracy of the Word2Vec model in identifying transaction features;

[0016] According to the verification results of the Word2Vec model, the model parameters of the Word2Vec model are adjusted to obtain an adjusted Word2Vec model that can identify the characteristics of suspicious bank transactions.

[0017] Furthermore, the model parameters include vocabulary size, word vector dimension, context window size, number of negative samples, and learning rate.

[0018] Furthermore, after generating word vectors through the Word2Vec model and verifying the model recognition accuracy, the Spearman coefficient is used to evaluate the effect of adjusting the model parameters. The Spearman coefficient is calculated according to the following formula (3):

[0019]

[0020] in, represents the Spearman correlation coefficient, which is used to measure the correlation of model performance before and after model parameter adjustment; represents the square of the difference between the rankings of two parameters; Indicates the training round.

[0021] Furthermore, the text in the standardized transaction record is segmented to extract key information words, including:

[0022] Use natural language processing tools to perform grammatical analysis on transaction record text, identify the part of speech of each word, and use context window technology to capture the central word and its related context words to ensure that each word has accurate semantic association in its context.

[0023] Dynamically filter out irrelevant or noisy words based on the contextual vocabulary in the transaction scenario;

[0024] Combined with the characteristic vocabulary of financial transactions, after vocabulary filtering, the words related to capital flow are preferentially extracted;

[0025] Based on the high-risk characteristic words in historical suspicious transaction records, the extracted capital flow-related words are weighted to enhance their importance in subsequent model training.

[0026] Furthermore, the weighted processing of the extracted capital flow-related words based on the high-risk characteristic words in the historical suspicious transaction records includes:

[0027] According to the following formula (4), the historical data weight of capital flow related words is calculated :

[0028]

[0029] in, Words related to capital flow The number of times the suspicious transaction appears in historical records; Indicates the total number of words in historical suspicious transaction records;

[0030] Based on the risk level defined internally by the bank, the modified vocabulary weight is calculated using the following formula (5): :

[0031]

[0032] in, is the risk correction factor; Vocabulary related to fund flow within the bank The assigned risk level, which ranges from 0 to 1, indicating a degree of risk from low risk to high risk;

[0033] Combined with the time characteristics in the transaction records, the vocabulary weight is further adjusted according to the following formula (6) to obtain the time decay weight :

[0034]

[0035] in, Indicates the timestamp of the current transaction; The reference time point includes high-risk periods when suspicious transactions are concentrated; is the time decay coefficient, which is used to measure the effect of time on the importance of vocabulary;

[0036] After considering historical weight, risk correction and time decay, the final vocabulary weight is calculated using the following formula (7): :

[0037]

[0038] in: is the adjustment factor for controlling risk and time; It is a correction term used to control the smoothing effect in specific cases to prevent the weight from being too high; is a nonlinear adjustment index used to adjust the weighted effects in extreme cases.

[0039] Furthermore, after the training is completed, the generated word vectors are verified to evaluate the accuracy of the Word2Vec model in identifying transaction features, including:

[0040] By comparing the word vectors generated by the model with the word vectors of known suspicious transaction feature words, the cosine similarity between the two is calculated to evaluate the recognition accuracy of the model.

[0041] Furthermore, after the training is completed, the generated word vectors are verified to evaluate the accuracy of the Word2Vec model in identifying transaction features, which also includes:

[0042] If the cosine similarity between the generated word vector and the word vector of the known suspicious transaction feature vocabulary exceeds the first threshold, the model is judged to have a high recognition accuracy; if the cosine similarity is lower than the second threshold, the model parameter adjustment mechanism is triggered.

[0043] Furthermore, the constructed vocabulary includes at least 5,000 transaction-related words, wherein the transaction-related words include banking business terms, transaction pattern words, and descriptive words of potentially suspicious transaction behaviors.

[0044] Furthermore, the vocabulary is dynamically updated according to the frequency of historical transaction data, and high-frequency transaction feature words are given priority during each training to ensure that the model has good adaptability to the latest transaction patterns.

[0045] The present application also provides a device for automatically identifying suspicious bank transaction characteristics, comprising:

[0046] A collection unit is configured to collect bank transaction records and standardize the text in the transaction records; wherein the transaction records include transfers, purchases, ATM withdrawals, over-the-counter withdrawals, financial products, precious metals, and loan records; the standardization includes standardizing date formats, normalizing amount units, and removing non-text symbols;

[0047] a processing unit configured to perform word segmentation on the text in the standardized transaction records to extract key information words; and construct a vocabulary based on the key information words; wherein the vocabulary includes words reflecting transaction characteristics for use in subsequent model training;

[0048] A construction unit is configured to train a Word2Vec model using the word segmentation results of the text in the transaction record and the constructed vocabulary. The Word2Vec model is implemented using the Skip-Gram model architecture. For a central word in a given transaction record text, the Skip-Gram model is trained to generate word vectors by predicting its context words to capture the semantic relationship between the words.

[0049] A training unit is used to train the Word2Vec model using a negative sampling method, wherein the negative sampling method randomly selects words that are unrelated to the central word as negative samples to reduce the computational effort of the model in predicting all words;

[0050] The verification unit is used to verify the generated word vectors after training is completed and evaluate the accuracy of the Word2Vec model in identifying transaction features;

[0051] The adjustment unit is used to adjust the model parameters of the Word2Vec model according to the verification result of the Word2Vec model to obtain an adjusted Word2Vec model that can identify the characteristics of suspicious bank transactions.

[0052] The technical solution proposed in this application has the following beneficial technical effects:

[0053] (1) By standardizing the text in bank transaction records, including unifying the date format, normalizing the amount unit, and removing non-text symbols, the consistency of transaction data is ensured, the interference of noise data is reduced, and the accuracy of subsequent model recognition is improved.

[0054] (2) By performing word segmentation on the standardized transaction record text and combining it with the constructed vocabulary, the model can accurately extract key information words related to bank transactions, thereby capturing the characteristics reflecting suspicious transaction behavior and improving the recognition efficiency of the model.

[0055] (3) The Skip-Gram model architecture is combined with the negative sampling method to avoid the model's comprehensive prediction of all words, greatly reducing the amount of calculation, improving the model's training speed and resource utilization, and adapting to the processing needs of large-scale bank transaction data. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flowchart of a method for automatically identifying suspicious transaction characteristics in a bank provided in the first embodiment of the present application.

[0057] Figure 2 This is a schematic diagram of a device for automatically identifying suspicious transaction characteristics in a bank provided in the second embodiment of the present application. DETAILED DESCRIPTION

[0058] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of the present application. Therefore, the present application is not limited to the specific implementations disclosed below.

[0059] The first embodiment of this application provides a method for automatically identifying suspicious bank transaction characteristics. Figure 1 , which is a schematic diagram of the first embodiment of this application. Figure 1 The first embodiment of the present application provides a method for automatically identifying suspicious transaction characteristics in a bank, which is described in detail.

[0060] Step S101: Collect bank transaction records and standardize the text in the transaction records; wherein the transaction records include transfers, consumption, ATM withdrawals, over-the-counter withdrawals, financial management purchases, precious metal purchases, and loan records; the standardization includes standardizing date formats, normalizing amount units, and removing non-text symbols.

[0061] In step S101, the bank's transaction records must first be collected. These transaction records typically include various transaction types, such as transfers, purchases, ATM withdrawals, over-the-counter withdrawals, purchases of wealth management products, purchases of precious metals, and loan records. Transaction records may include date, time, amount, transaction type, account information of the transacting parties, etc. This data can come from various banking systems, such as the customer's account management system or the bank's internal transaction record database. Data collection should be as comprehensive as possible to ensure coverage of various banking business scenarios. The specific collection method can be exporting from the database or regularly obtaining data through an API interface. The frequency and scope of data collection should be determined based on the system design requirements.

[0062] After data collection is completed, these raw transaction records must be standardized. The text of transaction records often has problems such as inconsistent formats and symbols. Directly using this data for processing will result in reduced model recognition accuracy. Therefore, the first task of standardization is to ensure the consistency of data formats. First, the date information in all transaction records should be unified into the standard "YYYY-MM-DDHH:MM:SS" format, which ensures that all time information can be handled consistently during subsequent processing. For dates that do not conform to the format, the system needs to be able to automatically detect and correct them to ensure that all data conforms to the standard. In addition, if there are different time formats in the transaction records, such as time formats that only contain dates or do not contain seconds, the system should automatically supplement them.

[0063] Standardizing amounts is another important step in standardization. Different transaction systems may use different monetary units or different currency symbols to represent the same currency. Therefore, all transaction amounts should be converted to the banking system's base currency and expressed using a consistent number of decimal places.

[0064] In addition to standardizing dates and amounts, removing non-text symbols is a key step in the standardization process. Bank transaction records often contain various non-text symbols, which are often meaningless or useless for model training. For example, some transaction notes may contain special characters, emoticons, or excessive punctuation. These symbols add noise to the data and affect subsequent word segmentation and model training. Therefore, the standardization process must filter out or replace these symbols.

[0065] The standardization process may also require handling of other special cases. For example, some transaction records may contain mixed languages ​​or abbreviations. In such cases, the system should be able to recognize these specific character combinations and perform appropriate processing, such as expanding common abbreviations into full terms or converting different language terms into a single language. The system should consider specific banking business scenarios and the language model used to ensure that this processing meets the business logic requirements.

[0066] Through the above standardized processing steps, it is possible to ensure that key information in transaction records is effectively retained and provide a stable data foundation for subsequent text processing and feature extraction.

[0067] Step S102: Segment the text in the standardized transaction record to extract key information words; construct a vocabulary based on the key information words; wherein the vocabulary includes words that reflect transaction characteristics for subsequent model training.

[0068] In step S102, the core purpose of performing word segmentation on the standardized bank transaction record text is to extract key information vocabulary to provide high-quality input data for subsequent model training. First, the word segmentation process starts with the standardized text, which has been formatted and symbolized in step S101. This ensures the standardization and consistency of the text content. This lays the foundation for the smooth execution of the word segmentation algorithm.

[0069] The specific implementation steps of word segmentation can be accomplished using existing natural language processing tools, including rule-based word segmentation methods or machine learning-based word segmentation algorithms. When processing transaction records, due to the unique characteristics of the financial sector, transaction descriptions may contain a variety of data types, including proper nouns, transaction type abbreviations, account information, time, and amount. Therefore, the primary task of word segmentation is to identify and disassemble this mixed text information, ensuring that the key information terms reflecting transaction behavior can be accurately extracted.

[0070] When implementing word segmentation, it is necessary to process it for specific scenarios of financial transactions. For example, in transaction records, words such as "transfer," "deposit," and "large cash withdrawal" are high-frequency words in banking transactions, and the system should be able to prioritize identifying these words. This can be achieved by combining a pre-built financial feature vocabulary. The feature vocabulary should include various transaction-related terms, especially key terms that may indicate suspicious transactions, such as feature descriptions such as "multiple small transfers" and "cash withdrawals from different locations." This type of word segmentation processing based on a domain-specific vocabulary can significantly improve the accuracy of transaction feature extraction.

[0071] Furthermore, the system needs to possess a certain level of contextual awareness. Because transaction record text often contains phrases or sentence structures, simple word segmentation may overlook the semantic relationships between words. Therefore, during the word segmentation process, the system should incorporate contextual windowing techniques to analyze each word and its surrounding words to capture the connections between them. For example, while "distinct" and "large amount" in "distinct large amount cash withdrawal" are independent words, in context they together describe an important transaction feature. This requires the word segmentation process to preserve this association, ensuring that the model can recognize the overall meaning of these characteristic words during subsequent processing.

[0072] After word segmentation, the system compares the generated vocabulary with a list of financial feature words to identify key information terms. These key information terms should be highly representative and directly reflect the characteristics of transaction behavior. This step aims to filter out irrelevant and noise words, such as common stop words, punctuation marks, and other meaningless characters. In financial transaction scenarios, common terms such as "customer" and "bank," while common, are not very helpful in identifying transaction characteristics and should be ignored as irrelevant.

[0073] Once the key information terms are extracted, the system constructs a vocabulary based on them. This vocabulary is more than just a simple list; it needs to include detailed information about each term, such as its frequency, distribution within transaction records, and its association with other terms. This detailed vocabulary provides higher-quality input data for subsequent model training, enabling the model to better capture transaction characteristics based on these key information terms. This is particularly true when detecting suspicious transactions, effectively identifying abnormalities that deviate from normal transaction patterns.

[0074] When building a vocabulary, a dynamic update mechanism can also be considered. Because bank transaction records constantly evolve with new descriptions or features over time, the system should be able to update vocabulary terms and their weights in real time based on new transaction data. For example, the emergence of new fraudulent methods or transaction patterns may lead to the appearance of new key words in transaction descriptions. The system should be able to identify these new terms and add them to the vocabulary, improving the model's adaptability to emerging transaction features.

[0075] Through the above-mentioned word segmentation and vocabulary construction process, the system can extract high-quality words that are closely related to transaction characteristics from standardized transaction records, ensuring that subsequent Word2Vec model training can generate accurate word vectors based on these words.

[0076] Furthermore, the text in the standardized transaction record is segmented to extract key information words, including:

[0077] Use natural language processing tools to perform grammatical analysis on transaction record text, identify the part of speech of each word, and use context window technology to capture the central word and its related context words to ensure that each word has accurate semantic association in its context.

[0078] Dynamically filter out irrelevant or noisy words based on the contextual vocabulary in the transaction scenario;

[0079] Combined with the characteristic vocabulary of financial transactions, after vocabulary filtering, the words related to capital flow are preferentially extracted;

[0080] Based on the high-risk characteristic words in historical suspicious transaction records, the extracted capital flow-related words are weighted to enhance their importance in subsequent model training.

[0081] First, the transaction record text is parsed using natural language processing tools to identify the part of speech of each word. Next, context window technology is used to capture each central word and its related contextual words, ensuring that each word has accurate semantic associations in its specific context. Based on the specific transaction scenario, the system can dynamically filter out irrelevant or noisy words by analyzing the context, thereby reducing the possibility of misjudgment. Subsequently, combined with a vocabulary of financial transaction characteristics, it prioritizes the extraction of words related to capital flows, as these words may better reflect the characteristics and risk behavior of transactions. Finally, the system weights the extracted capital flow-related words based on high-risk characteristic words in historical suspicious transaction records, giving these words higher weights in subsequent model training and enhancing the model's ability to identify suspicious transactions.

[0082] Furthermore, the weighted processing of the extracted capital flow-related words based on the high-risk characteristic words in the historical suspicious transaction records includes:

[0083] According to the following formula (4), the historical data weight of capital flow related words is calculated :

[0084]

[0085] in, Words related to capital flow The number of times the suspicious transaction appears in historical records; Indicates the total number of words in historical suspicious transaction records;

[0086] Based on the risk level defined internally by the bank, the modified vocabulary weight is calculated using the following formula (5): :

[0087]

[0088] in, is the risk correction factor; Vocabulary related to fund flow within the bank The assigned risk level, which ranges from 0 to 1, indicating a degree of risk from low risk to high risk;

[0089] Combined with the time characteristics in the transaction records, the vocabulary weight is further adjusted according to the following formula (6) to obtain the time decay weight :

[0090]

[0091] in, Indicates the timestamp of the current transaction; The reference time point includes high-risk periods when suspicious transactions are concentrated; is the time decay coefficient, which is used to measure the effect of time on the importance of vocabulary;

[0092] After considering historical weight, risk correction and time decay, the final vocabulary weight is calculated using the following formula (7): :

[0093]

[0094] in: is the adjustment factor for controlling risk and time; It is a correction term used to control the smoothing effect in specific cases to prevent the weight from being too high; is a nonlinear adjustment index used to adjust the weighted effects in extreme cases.

[0095] The automated identification method for suspicious bank transaction characteristics employs a complex weighting approach for extracted fund flow-related terms to enhance the model's ability to identify suspicious transactions. This approach incorporates historical data, the bank's internally defined risk levels, and time factors. Using multiple formulas, the weights of fund flow-related terms are dynamically adjusted to ensure that the resulting term weights accurately reflect the term's importance in the context of the suspicious transaction.

[0096] First, formula (4) calculates the historical data weights of capital flow related terms , used to measure the frequency of a certain word appearing in historical suspicious transaction records:

[0097] in, Words related to capital flow The number of times it appears in historical suspicious transaction records, is the total number of words in historical suspicious transaction records. This weight reflects the importance of the word in past transactions. By calculating the frequency distribution of words, the system can identify high-risk words that appear frequently in historical suspicious transactions and assign them a higher initial weight.

[0098] Next, formula (5) calculates the modified vocabulary weight by combining the risk level defined within the bank .

[0099] In this formula, is the risk correction factor, which controls the impact of risk level on vocabulary weight; It is a term related to fund flow within the bank. The assigned risk level, ranging from 0 to 1, represents the degree of risk associated with a term, from low to high. If a term appears frequently in historical data and is assigned a higher risk level, its revised weight will be significantly increased. This step ensures that banks can assign additional weight to certain high-risk terms based on their own risk management experience.

[0100] Then, the system combines the time features in the transaction records and uses formula (6) to perform time decay processing on the vocabulary weight to obtain the weight after time decay .

[0101] In this formula, Indicates the timestamp of the current transaction, A reference time point, usually a high-risk period with a high concentration of suspicious transactions; is the time decay coefficient, which is used to measure the impact of time on the weight of words. As time goes by, some words that appear frequently in history may gradually lose their importance, so the system reduces the weight of these words through the time decay function. If the current transaction time is closer to the reference time, , the smaller the effect of time decay, the higher the weight of the vocabulary remains; conversely, the importance of the vocabulary will decay over time.

[0102] After comprehensively considering the effects of historical weight, risk correction and time decay, formula (7) is used to calculate the final vocabulary weight .

[0103] In this formula, is the coefficient that controls risk and time adjustment, and determines the relative impact of time decay and risk correction; It is a correction term used to smooth the weights in extreme cases and prevent excessively increasing the weights of certain words; is a nonlinear adjustment exponent that controls weight changes in extreme cases. The structure of the formula ensures that when historical weight and time decay are large, the word weight will increase significantly. However, by introducing a nonlinear adjustment exponent and a correction term, it prevents the excessive increase of certain word weights and maintains the balance of the model.

[0104] Through weighted processing, the system not only identifies frequently occurring high-risk terms but also dynamically adjusts the final weight of each term based on the bank's internal risk level definitions and transaction timing. This approach ensures that the model fully utilizes historical data, temporal information, and risk management experience during training, improving the accuracy and efficiency of identifying suspicious transactions.

[0105] Step S103: Using the word segmentation results of the text in the transaction record and combining them with the constructed vocabulary, the Word2Vec model is trained. The Word2Vec model is implemented using the Skip-Gram model architecture. For a given central word in the transaction record text, the Skip-Gram model is trained to generate word vectors by predicting its context words to capture the semantic relationship between words. The probability of occurrence of each context word is calculated according to the following formula (1):

[0106]

[0107] in, Indicates the central word, represents the context word, where Indicates the position of the central word in the sentence. Represents context-dependent indexes; Representing context words word vectors; Indicates the central word word vectors; A word vector representing each word in the vocabulary; is the size of the vocabulary.

[0108] In step S103, the transaction record segmentation results are combined with the constructed vocabulary and used to train the Word2Vec model. The Word2Vec model is a neural network model that converts text words into word vectors. It can capture the semantic relationships between words in a high-dimensional space. By training the words in the transaction records into word vectors, the model can learn the correlations between different words in these records, thereby helping to identify the characteristics of suspicious transactions.

[0109] The Word2Vec model uses the Skip-Gram model architecture, which means that the core of the model is to learn word vectors by predicting the context words of the central word. In this architecture, given a central word , the system predicts the words that may be related to the word through the context of the word. Indicates that is a relative position index, indicating the distance between the vocabulary and the center word. The goal of the model is to train the word vector of the center word by predicting the context words, thereby capturing the potential semantic relationship between the words.

[0110] Specifically, the Skip-Gram model learns these relationships by maximizing the probability of the context words appearing. In the given central word The probability of occurrence under the condition is calculated by the following formula:

[0111]

[0112] In this formula, each variable has a clear meaning. First, Representing context words The word vector is a multidimensional vector that captures the semantic features of the word in the training data. Next, Indicates the central word The word vector of , which is also a multidimensional vector. The inner product between these two vectors The semantic similarity between the center word and the context words was calculated.

[0113] The summation formula in the denominator represents the normalization of the semantic similarity of all words in the vocabulary. Represents the word vector for each word in the vocabulary, is the size of the vocabulary. Therefore, the denominator is to calculate the similarity between the central word and all the words to ensure that the probability of the context words is relative, not only related to the semantics of the central word, but also affected by other words.

[0114] This probability formula reflects the model's ability to gradually adjust word vectors by observing the relationship between the central word and context words, bringing semantically similar word vectors closer together in high-dimensional space. For example, the system might discover that "withdraw cash" and "ATM withdrawal" frequently appear together in many transaction records, so their word vectors are trained to be very close. This proximity reflects their semantic similarity, helping the model identify potentially suspicious transaction characteristics.

[0115] Step S104: The Word2Vec model is trained using a negative sampling method, wherein the negative sampling method randomly selects words that are not related to the central word as negative samples to reduce the amount of computation required for the model to predict all words. The loss function provided by the following formula (2) is used during the training process:

[0116]

[0117] in, Represents the loss function, which is used to measure the prediction error of the model during training; Represents the total number of samples in the training data; Indicates the window size; Representing context words word vectors; Indicates the central word word vectors; Indicates the total number of negative samples; Represents the distribution of negative samples Randomly sample negative samples The expected value of probability; Represents negative samples word vectors; is the sigmoid function.

[0118] In step S104, negative sampling is used to train the Word2Vec model. Negative sampling is an optimization technique whose primary purpose is to reduce the computational effort required to predict all vocabulary during training, while ensuring that the model can efficiently capture the semantic relationships between vocabulary. Because vocabulary sizes are typically very large, especially in financial transaction data involving a large number of specific words, directly calculating all vocabulary is not only very time-consuming but also unnecessary. Negative sampling significantly improves the computational efficiency of the model by selecting only a subset of negative samples for training.

[0119] First, in order to understand the use of negative sampling, it is necessary to make it clear that the goal of the model is to use the center word To predict context words , and the probability of each context word is given by the following formula:

[0120]

[0121] here, The word vector representing the context word, Represents the word vector of the central word, and the denominator represents the sum of the similarities of all words. However, due to the huge size of the vocabulary, it is impractical to calculate the probability of all words, so the negative sampling method is introduced.

[0122] The core idea of ​​negative sampling is that for each true context word, the model not only learns how to correctly predict it, but also introduces randomly selected negative samples (words unrelated to the central word) to train the model to identify which words are unlikely to appear in that context. This way, the model does not need to calculate the similarity of all words, but only needs to calculate the similarity of the actual context words and a small number of negative samples, greatly reducing the computational effort.

[0123] When implementing negative sampling, the model randomly selects negative samples, Usually a small integer, such as 5 or 10. Each negative sample is from the vocabulary based on a probability distribution Randomly selected, this probability distribution is usually related to the frequency of the word, making high-frequency words more likely to be selected as negative samples. Negative sampling ensures that the model can not only identify positive samples (i.e., true context words), but also reduces the probability of incorrect predictions by training the model with negative samples.

[0124] During the training process, the loss function of the model It is used to measure the error of the model in predicting context words. This loss function consists of two parts: one is the prediction for positive samples, and the other is the prediction for negative samples. The specific loss function form is as follows:

[0125]

[0126] In this formula, Represents the total number of samples in the training data, which means that the model will iterate on all training data; Indicates the size of the context window, which controls how many context words around the center word will be considered. Used to calculate the prediction results of the real context words, where It is a sigmoid function that maps the prediction results between 0 and 1 to indicate the degree of match between the predicted context word and the central word.

[0127] The second part It is the prediction for negative samples. Negative samples are from the distribution The random sampling is obtained from the vocabulary, indicating that the words in the vocabulary should not be compared with the central word By maximizing the negative logarithm of the sigmoid of negative samples, the model is trained to minimize the matching probability of negative samples.

[0128] Each negative sample The word vector is used Negative sampling introduces irrelevant negative examples and minimizes the probability of these words matching the central word. This allows the model to better distinguish which words should appear in the context of the central word and which should not. This training mechanism significantly improves model efficiency, especially when processing large vocabularies, enabling the model to more accurately capture complex semantic relationships in transaction records.

[0129] Negative sampling not only allows the model to complete training faster, but also avoids overcalculating the similarity of irrelevant words, thereby reducing computational costs. This is particularly important for identifying suspicious bank transactions, as transaction records often involve a large vocabulary and the correlation between different words can vary significantly. Negative sampling allows the system to efficiently train the model, enabling it to accurately identify abnormal or suspicious behavior in transactions, thereby improving the accuracy of suspicious transaction detection.

[0130] In summary, negative sampling provides an efficient training mechanism for the Word2Vec model. By randomly selecting a small number of negative samples, it avoids a comprehensive evaluation of the entire vocabulary, significantly improving model training efficiency while ensuring the accuracy of the model's semantic capture. This method is particularly suitable for processing large-scale financial transaction data and can help identify potentially suspicious transactions in practical applications.

[0131] Step S105: After the training is completed, the generated word vectors are verified to evaluate the accuracy of the Word2Vec model in identifying transaction features.

[0132] In step S105, after training is complete, the generated word embeddings need to be verified to assess the accuracy of the Word2Vec model in identifying transaction features. The main purpose of this verification step is to ensure that the word embeddings learned by the model accurately reflect the semantic relationships in bank transaction records, especially in identifying suspicious transaction features with high efficiency and accuracy.

[0133] First, the basic principle of verification is to determine whether the model successfully captures the semantic relationship between words by evaluating the distance between the generated word vectors and the words in the corresponding vocabulary in the semantic space. The "distance" here refers to the similarity in the high-dimensional vector space, which is generally quantified by calculating cosine similarity. Cosine similarity is a common metric in natural language processing that measures the similarity between two word vectors. The formula is as follows:

[0134]

[0135] In this formula, and Represent the word vectors of two words respectively, is their dot product, and Represents the modulus of the two vectors. The cosine similarity result ranges from -1 to 1, where 1 indicates complete identity, 0 indicates irrelevance, and -1 indicates complete oppositeness. Therefore, by calculating the cosine similarity between each generated word vector and its corresponding word in the vocabulary, we can evaluate the model's ability to capture the semantics of these words.

[0136] In practice, the system randomly extracts words from a subset of transaction records and compares the generated word vectors with those in the vocabulary. For each generated word vector, the system calculates its similarity with other words and ranks them based on cosine similarity. Ideally, the word vectors generated by the model should be ranked near the top of the words in the vocabulary that are most semantically similar to them. This method provides a preliminary validation of the model's ability to effectively distinguish the semantics of different words and capture the characteristics of transaction descriptions.

[0137] In transaction scenarios, especially those involving feature identification of suspicious transactions, the model must not only capture the surface vocabulary of the transaction but also identify underlying semantic connections. For example, when describing "cash withdrawals from different locations" or "frequent small transfers," the model must be able to identify the underlying connections between these terms and correctly flag them as potentially suspicious transactions. Therefore, model accuracy verification requires more than just verifying the semantic similarity of word vectors; it should also be analyzed in conjunction with transaction features in actual business scenarios.

[0138] To further ensure model accuracy, the system can perform specialized validation on a specific set of suspicious transaction signature terms. This process involves defining a set of known suspicious transaction signature terms, such as "multiple transfers" and "large cash withdrawals," and then evaluating the model's recognition of these signature terms. By comparing the similarity between the word vectors generated by the model for these signature terms and the standard word vectors in the vocabulary, we can accurately determine the model's performance in real-world scenarios.

[0139] In addition to verification based on cosine similarity, the system can also be evaluated through coverage and vocabulary distribution characteristics. Coverage refers to the proportion of matches between generated word vectors and words in the vocabulary. If the generated word vectors can cover most of the important words in the vocabulary, it indicates that the model has a high level of semantic capture ability. In addition, by analyzing the distribution of word vectors in high-dimensional space, especially whether suspicious transaction feature words form clear clusters, the model's recognition performance can be further evaluated. If the model can concentrate semantically similar word vectors in a certain area of ​​space, while words with larger semantic differences are distributed in different areas, it indicates that the model training is successful and can accurately distinguish different transaction features.

[0140] Finally, the model validation results serve as a basis for adjusting model parameters. By analyzing the accuracy and coverage of generated word embeddings, the system can identify deficiencies in model training, such as low recognition accuracy for certain characteristic words or insufficient capture of important transaction characteristics. This validation information is fed back into the model training process, allowing parameters such as word embedding dimension, context window size, and number of negative sampling to be adjusted to further optimize the model and achieve even more accurate capture of transaction characteristics.

[0141] In summary, the verification process in step S105 ensures that the generated word vectors can accurately reflect the semantic relationships in the bank transaction records, especially with high accuracy in identifying suspicious transaction features.

[0142] Furthermore, after the training is completed, the generated word vectors are verified to evaluate the accuracy of the Word2Vec model in identifying transaction features, including:

[0143] By comparing the word vectors generated by the model with the word vectors of known suspicious transaction feature words, the cosine similarity between the two is calculated to evaluate the recognition accuracy of the model.

[0144] In the automatic identification of suspicious transaction characteristics in banks, the generated word vectors need to be verified after training to evaluate the Word2Vec model's accuracy in identifying transaction characteristics. This verification process is performed by calculating the cosine similarity between the generated word vectors and the word vectors of known suspicious transaction characteristics.

[0145] Specifically, the system compares the word vectors generated by the model with the vectors of known suspicious transaction characteristics in the vocabulary. Cosine similarity measures the similarity between two vectors and is calculated as the dot product of the vectors divided by the product of their moduli. Cosine similarity values ​​range from -1 to 1, where 1 indicates that the two vectors are identical, 0 indicates they are completely unrelated, and -1 indicates they are completely opposite.

[0146] This method quantifies the similarity between the generated word vectors and known high-risk terms, thereby assessing the model's accuracy in identifying these characteristic terms in suspicious transactions. A high cosine similarity value indicates that the model is effectively identifying the characteristic term. Conversely, a low similarity indicates that the model's recognition of the term is not ideal, and further model optimization or parameter adjustment may be needed.

[0147] Furthermore, after the training is completed, the generated word vectors are verified to evaluate the accuracy of the Word2Vec model in identifying transaction features, which also includes:

[0148] If the cosine similarity between the generated word vector and the word vector of the known suspicious transaction feature vocabulary exceeds the first threshold, the model is judged to have a high recognition accuracy; if the cosine similarity is lower than the second threshold, the model parameter adjustment mechanism is triggered.

[0149] In the automatic identification of suspicious transaction characteristics in banks, after model training is complete, the generated word vectors are verified to assess the Word2Vec model's accuracy in identifying transaction characteristics. During this verification process, the system evaluates the model's performance by calculating the cosine similarity between the generated word vectors and the word vectors of known suspicious transaction characteristics.

[0150] Specifically, if the cosine similarity between the generated word vector and a known suspicious transaction feature vocabulary exceeds a pre-set first threshold, the model demonstrates high recognition accuracy and is able to effectively capture the characteristics of these suspicious transactions. At this point, the system will determine that the model's performance meets the expected standard.

[0151] However, if the cosine similarity falls below the second threshold, the model performs poorly in identifying specific suspicious transaction signature words, possibly due to improper parameter settings. Therefore, the system triggers a model parameter adjustment mechanism to further optimize model performance by modifying model parameters such as vocabulary size, word vector dimensions, or learning rate. This mechanism ensures the accuracy and robustness of the model in practical applications, enabling it to effectively identify potentially suspicious transactions in various transaction scenarios. Using this automated adjustment mechanism, those skilled in the art can optimize the model in real time to improve the system's overall transaction signature recognition capabilities.

[0152] Furthermore, the constructed vocabulary includes at least 5,000 transaction-related words, wherein the transaction-related words include banking business terms, transaction pattern words, and descriptive words of potentially suspicious transaction behaviors.

[0153] The method for automatically identifying suspicious bank transaction characteristics builds a vocabulary containing at least 5,000 words related to banking transactions. This vocabulary includes not only common banking terms but also covers different types of transaction patterns and terms used to describe potentially suspicious transactions.

[0154] Specifically, banking business terminology encompasses vocabulary related to everyday financial operations, such as deposits, withdrawals, and transfers. Transaction pattern vocabulary describes common transaction methods or patterns, such as regular transfers and cross-border transactions. Descriptive vocabulary for suspicious transaction behavior describes characteristic terms that frequently appear in risky transaction analysis, such as "frequent transfers" and "large cash withdrawals." These terms form the foundation for feature extraction and classification in automatic recognition methods. By building and utilizing this comprehensive vocabulary, the model can more accurately identify and judge potential risky behaviors in banking transactions.

[0155] Furthermore, the vocabulary is dynamically updated according to the frequency of historical transaction data, and high-frequency transaction feature words are given priority during each training to ensure that the model has good adaptability to the latest transaction patterns.

[0156] In the automated identification of suspicious bank transaction signatures, the vocabulary is dynamically updated. This dynamic update is based on the frequency of vocabulary in historical transaction data. Specifically, during each model training, the system analyzes the frequency of vocabulary in transaction records and prioritizes frequently occurring transaction signature vocabulary to ensure optimal representation and training of these vocabulary within the model.

[0157] Through this dynamic update mechanism, the vocabulary can promptly reflect the latest transaction patterns and trends, enabling the model to not only adapt to historical transaction data but also quickly capture new suspicious transaction behavior characteristics. This update method ensures that the model maintains high recognition capabilities and accuracy in the ever-changing banking transaction environment, effectively addressing new transaction patterns and potentially risky behaviors. Through this dynamic update mechanism, technicians can maintain the real-time and adaptability of the model, thereby improving the overall performance of the system.

[0158] Furthermore, after generating word vectors through the Word2Vec model and verifying the model recognition accuracy, the Spearman coefficient is used to evaluate the effect of adjusting the model parameters. The Spearman coefficient is calculated according to the following formula (3):

[0159]

[0160] in, represents the Spearman correlation coefficient, which is used to measure the correlation of model performance before and after model parameter adjustment; represents the square of the difference between the rankings of two parameters; Indicates the training round.

[0161] In this example, after using the Word2Vec model to generate word vectors and verifying the model's recognition accuracy, the model parameters need to be adjusted to further optimize the model's performance. To evaluate the effectiveness of these parameter adjustments, a method for calculating the Spearman coefficient is introduced. This coefficient can quantify the correlation between performance changes before and after model parameter adjustments, helping developers determine whether the parameter adjustments are effective.

[0162] Spearman coefficient A nonparametric statistical metric that measures the correlation between two variables (in this example, the performance results before and after model parameter adjustments). It is calculated based on the variable's ranking rather than its actual value, making it particularly useful for capturing nonlinear relationships when evaluating the effects of model parameter adjustments.

[0163] In formula (3), each symbol has a specific meaning. First, Represents the Spearman correlation coefficient, whose value range is between -1 and 1. A value close to 1 indicates a high positive correlation, which means that the model performance is significantly improved after the parameter adjustment; a value close to -1 indicates a high negative correlation, which means that the model performance decreases after the adjustment; and a value close to 0 indicates that there is no significant change in performance before and after the adjustment.

[0164] In the formula is the sum of squares of ranking differences, Representative The difference in ranking before and after model parameter adjustments in a group of data. Specifically, for a given round of training data, the system calculates the performance ranking before and after the model parameter adjustments. If the performance after the adjustment is higher than before the adjustment, the ranking is higher, and vice versa. By calculating the ranking difference before and after the parameter adjustment for each training round and squaring it, the influence of negative ranking differences on the total can be eliminated, ensuring that positive and negative differences are evenly reflected in the results.

[0165] The sum of squares reflects the total amount of ranking differences. A larger ranking difference indicates a greater fluctuation in model performance after parameter adjustment, while a smaller sum of squares means a smaller ranking change, indicating that parameter adjustment has a more stable effect on model performance. The system can intuitively quantify the performance difference between the model before and after adjustment over multiple training rounds.

[0166] Next, Is the total number of training rounds. This value reflects the number of ranking results produced by the model during multiple rounds of training. In practical applications, the system evaluates the stability and consistency of the model through multiple training rounds. It provides a more comprehensive assessment, making the results of the Spearman coefficient more credible.

[0167] In the formula Is a multiple of the ranking difference, multiplied by 6 is the constant in the Spearman coefficient calculation, used to normalize the final correlation result. It is used to standardize the sum of squares of ranking differences so that the result of the Spearman coefficient remains in the range of -1 to 1.

[0168] Using this formula, the system can effectively evaluate the model's relevance before and after parameter adjustments. If the Spearman coefficient is close to 1, it indicates that the parameter adjustment has significantly improved the model's performance, and the system will apply this adjustment method to subsequent optimization processes. If the Spearman coefficient is close to -1, it indicates that the current parameter adjustment strategy has had a negative impact on model performance, and the system needs to further explore and optimize the adjustment method. If the Spearman coefficient is close to 0, it indicates that the parameter adjustment has had little impact on model performance. In this case, it may be necessary to adjust other parameters or further analyze the overall structure and training strategy of the model.

[0169] In practice, the system iteratively adjusts model parameters such as vocabulary size, word vector dimension, context window size, number of negative samples, and learning rate. After each parameter adjustment, the system analyzes the positive and negative impact of the adjustment on model performance based on feedback from the Spearman coefficient, gradually optimizing the overall model. By incorporating the Spearman coefficient, the system can scientifically and quantitatively evaluate the effectiveness of parameter adjustments and ensure that the resulting Word2Vec model can efficiently and accurately identify the characteristics of suspicious bank transactions.

[0170] Step S106: According to the verification result of the Word2Vec model, the model parameters of the Word2Vec model are adjusted to obtain an adjusted Word2Vec model capable of identifying features of suspicious bank transactions.

[0171] In step S106, based on the results from the previous validation step, the Word2Vec model parameters are adjusted to optimize the model's performance and enhance its ability to identify suspicious bank transaction characteristics. The core of this step is to precisely adjust the model's key parameters through feedback from model evaluation, enabling the model to more accurately capture transaction characteristics in different scenarios.

[0172] First, the adjustment of model parameters involves multiple aspects, including word vector dimension, context window size, number of negative samples for negative sampling, learning rate, etc. The adjustment of each parameter has a direct impact on the performance of the model. For example, the dimension of the word vector determines the accuracy with which the vocabulary is represented in the semantic space. Lower dimensions may not be able to fully capture complex semantic relationships, while too high a dimension may increase the computational complexity of the model and introduce overfitting. Therefore, when adjusting the word vector dimension, the model needs to strike a balance between accuracy and computational efficiency. The word vector quality indicators obtained in the verification step, such as cosine similarity, can be used to evaluate the performance of the current word vector and determine whether the dimension needs to be increased or decreased.

[0173] The size of the context window It is also a crucial parameter. The window size determines how many words surrounding the central word are used as context words for training. If the window is too small, the model may not capture sufficient semantic information, especially in long transaction descriptions. However, if the window is too large, the model may introduce too much irrelevant information, resulting in increased noise. Therefore, adjustment of the window size should be based on an analysis of the semantic relevance between words during the verification process. For example, by observing the impact of the context window size on specific suspicious transaction feature words, it is possible to determine whether to expand or reduce the window size.

[0174] Number of negative samples is another parameter that requires adjustment. Negative sampling optimizes the model's computational efficiency by selecting a small number of words that are unrelated to the central word. If the number of negative samples is too small, the model may not fully learn how to distinguish positive samples from irrelevant words; conversely, if the number of negative samples is too large, the computational burden may increase. Therefore, the number of negative samples can be adjusted by observing how the model handles noisy data. If the model performs poorly with noisy words during validation, it may mean that the number of negative samples needs to be increased to help the model better learn how to ignore irrelevant information.

[0175] The learning rate is a key parameter that controls the rate at which model parameters are updated. During model training, the learning rate determines the magnitude of the model parameter update at each iteration. If the learning rate is too high, the model may jump too much in the parameter space and fail to converge to the optimal value stably; if the learning rate is too low, the training process may be too slow or even fall into a local optimum. Therefore, by analyzing the loss function in the verification process, The changing trend of the learning rate can determine whether the learning rate needs to be adjusted. If the loss of the model drops too quickly in the early stages and fluctuates greatly in the later stages, it may be necessary to appropriately reduce the learning rate to make the model converge more stably.

[0176] Loss Function The specific formula is as follows:

[0177]

[0178] In this formula, represents the total number of samples in the training data, and are the word vectors of the context words and the center word respectively, is the number of negative samples. Analyzing the gradual optimization process of the loss function, especially the changes in loss under different parameter settings, can help determine whether the model training is effective. If the loss function value continues to decrease after model adjustments, it indicates that the model is moving in the right direction during the optimization process. Conversely, if the loss function value fluctuates significantly or fails to steadily decrease, it indicates that certain parameters need further adjustment.

[0179] In addition, the size of the vocabulary also needs to be considered when tuning model parameters. Impact on model performance. A larger vocabulary can cover more transaction feature words, but it may also increase model training time. Therefore, by analyzing the vocabulary coverage during validation, you can determine whether the vocabulary size is appropriate. If most of the words in the vocabulary are irrelevant to the training data, you may need to streamline the vocabulary to improve training efficiency.

[0180] Ultimately, through repeated adjustment and validation of these model parameters, the Word2Vec model can more accurately identify suspicious transaction characteristics in various banking transaction scenarios. The adjusted model should demonstrate high word vector similarity and low loss function values ​​during validation, accurately capturing key characteristic vocabulary in bank transaction records that reflect potential risks. This optimization process not only ensures the robustness of the model but also improves its effectiveness in practical applications, enabling banks to detect and respond to potentially suspicious transactions earlier.

[0181] Furthermore, the model parameters include vocabulary size, word vector dimension, context window size, number of negative samples, and learning rate.

[0182] The design of model parameters for automatic identification of suspicious banking transaction signatures is crucial. These parameters determine the model's training process, the quality of word embeddings, and ultimately, the effectiveness of identifying suspicious transaction signatures. To ensure the model's efficient operation in complex banking transaction scenarios and accurate identification of potentially risky transactions, detailed model parameter configuration and adjustment are essential. These parameters primarily include vocabulary size, word embedding dimension, context window size, number of negative samples, and learning rate.

[0183] First, the size of the vocabulary directly impacts the quality of the model's word embedding training. The vocabulary is the set of words used by the model to represent transaction records, encompassing all terms used in bank transaction descriptions. When constructing the vocabulary, the size should be appropriately set based on the characteristics of the transaction data. If the vocabulary is too small, the model may overlook certain important terms, particularly when identifying suspicious transactions, potentially missing certain risk features. Conversely, an overly large vocabulary increases the model's computational workload, prolongs training time, and may even introduce unnecessary noise. Therefore, the vocabulary size should be dynamically adjusted based on the vocabulary distribution of the transaction data, the complexity of the transaction scenarios, and the frequency of use of specific terms.

[0184] Secondly, the word embedding dimension is another key parameter that determines how the model represents each word in a high-dimensional space. Word embedding converts words into fixed-length numerical vectors, which are used to represent the semantic relationships between words. In bank transaction descriptions, different words may represent different transaction behaviors or characteristics. Therefore, the choice of word embedding dimension directly affects the model's ability to capture these transaction characteristics. If the dimension is too low, the model may not be able to accurately represent subtle semantic differences between words. Especially when dealing with complex transaction features, the model may not be able to distinguish between similar transaction descriptions. If the dimension is too high, while more complex semantic relationships can be captured, it also increases computational complexity and may lead to overfitting.

[0185] The size of the context window is a key parameter influencing model training, determining how the model captures the semantic relationships between the central word and its surrounding words during training. A larger context window allows the model to consider more contextual words related to the central word. This helps the model better understand the relationships between words when transaction descriptions are long or contain complex semantic relationships. However, an overly large context window may introduce excessive irrelevant information, resulting in reduced model prediction accuracy. On the other hand, an overly small window may not fully capture the contextual relationships of the central word. This is especially true in certain transaction descriptions, where the relevance of words may require a larger context window.

[0186] The number of negative samples is a crucial parameter in the model optimization process. In the Word2Vec model, negative sampling is used to improve model training efficiency. The number of negative samples determines how the model optimizes predictions during training by comparing true context words with randomly selected, irrelevant words. If the number of negative samples is too small, the model may not be able to effectively distinguish between positive and irrelevant negative samples, resulting in insufficient generalization ability in identifying suspicious transaction signatures. If the number of negative samples is too large, while the model's generalization ability is enhanced, the computational effort and training time will also increase significantly.

[0187] Finally, the learning rate is a crucial factor in controlling the rate at which model parameters are updated. During model training, the learning rate determines the magnitude of the model parameter updates at each iteration. Excessively high learning rates can cause the model to jump too far in parameter space, preventing it from stably converging to the optimal solution. Excessively low learning rates can slow the model's convergence and even trap it in a local optimum. Therefore, the learning rate setting needs to be adjusted dynamically based on the model's training progress.

[0188] In the above embodiment, a method for automatically identifying suspicious bank transaction features is provided. Correspondingly, the present application also provides an apparatus for automatically identifying suspicious bank transaction features. Figure 2 , which is a schematic diagram of an embodiment of a device for automatically identifying suspicious bank transaction characteristics according to the present application. Since this embodiment, the second embodiment, is substantially similar to the method embodiment, its description is relatively brief. For relevant details, please refer to the description of the method embodiment. The following embodiments are merely illustrative.

[0189] A second embodiment of the present application provides an automatic identification device for suspicious bank transaction characteristics, comprising:

[0190] The collection unit 201 is configured to collect bank transaction records and perform text standardization on the transaction records. The transaction records include transfers, purchases, ATM withdrawals, over-the-counter withdrawals, financial products, precious metals, and loan records. The standardization includes standardizing date formats, unit amounts, and removing non-text symbols.

[0191] Processing unit 202 is configured to perform word segmentation on the text in the standardized transaction records to extract key information words; construct a vocabulary based on the key information words; wherein the vocabulary includes words that reflect transaction characteristics and is used for subsequent model training;

[0192] The construction unit 203 is used to train the Word2Vec model using the word segmentation processing results of the text in the transaction record in combination with the constructed vocabulary; wherein the Word2Vec model is implemented using the Skip-Gram model architecture. For a central word in a given transaction record text, the Skip-Gram model is trained to generate a word vector by predicting its context words to capture the semantic relationship between words; wherein the occurrence probability of each context word is calculated according to the following formula (1):

[0193]

[0194] in, Indicates the central word, represents the context word, where Indicates the position of the central word in the sentence. Represents context-dependent indexes; Representing context words word vectors; Indicates the central word word vectors; A word vector representing each word in the vocabulary; is the size of the vocabulary;

[0195] The training unit 204 is used to train the Word2Vec model using a negative sampling method, wherein the negative sampling method randomly selects words that are not related to the central word as negative samples to reduce the amount of computation required for the model to predict all words. The loss function provided by the following formula (2) is used during the training process:

[0196]

[0197] in, Represents the loss function, which is used to measure the prediction error of the model during training; Represents the total number of samples in the training data; Indicates the window size; Representing context words word vectors; Indicates the central word word vectors; Indicates the total number of negative samples; Represents the distribution of negative samples Randomly sample negative samples The expected value of probability; Represents negative samples word vectors; is the sigmoid function;

[0198] A verification unit 205 is used to verify the generated word vectors after training is completed, and evaluate the accuracy of the Word2Vec model in identifying transaction features;

[0199] The adjustment unit 206 is configured to adjust the model parameters of the Word2Vec model according to the verification result of the Word2Vec model to obtain an adjusted Word2Vec model capable of identifying features of suspicious bank transactions.

[0200] Although the present application is disclosed as above with the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A method for automatically identifying suspicious bank transaction characteristics, characterized in that: include: Collect bank transaction records and standardize the text in the transaction records; the transaction records include transfers, purchases, ATM withdrawals, over-the-counter withdrawals, financial products, precious metals, and loan records; the standardization includes standardizing date formats, unit amounts, and removing non-text symbols; Performing word segmentation on the text in the standardized transaction records to extract key information words; constructing a vocabulary based on the key information words; wherein the vocabulary includes words that reflect transaction characteristics and is used for subsequent model training; The Word2Vec model is trained using the word segmentation results of the transaction record text and the constructed vocabulary. The Word2Vec model is implemented using the Skip-Gram model architecture. For a given central word in the transaction record text, the Skip-Gram model is trained to generate word vectors by predicting its context words to capture the semantic relationship between words. The probability of each context word occurring is calculated according to the following formula 1: in, Indicates the central word, represents the context word, where Indicates the position of the central word in the sentence. Represents context-dependent indexes; Representing context words word vectors; Indicates the central word word vectors; A word vector representing each word in the vocabulary; is the size of the exchange table; The Word2Vec model is trained using a negative sampling method, where the negative sampling method randomly selects words that are unrelated to the central word as negative samples to reduce the amount of computation required to predict all words. The loss function provided by the following formula 2 is used during training: in, Represents the loss function, which is used to measure the prediction error of the model during training; Represents the total number of samples in the training data; Indicates the window size; Representing context words word vectors; Indicates the central word word vectors; Indicates the total number of negative samples; Represents the distribution of negative samples Randomly sample negative samples The expected value of probability; Represents negative samples word vectors; is the sigmoid function; After training is completed, the generated word vectors are verified to evaluate the accuracy of the Word2Vec model in identifying transaction features; According to the verification results of the Word2Vec model, the model parameters of the Word2Vec model are adjusted to obtain an adjusted Word2Vec model that can identify the characteristics of suspicious bank transactions; The word segmentation process of the text in the standardized transaction records to extract key information words includes: Use natural language processing tools to perform grammatical analysis on transaction record text, identify the part of speech of each word, and use context window technology to capture the central word and its related context words to ensure that each word has accurate semantic association in its context. Dynamically filter out irrelevant or noisy words based on the contextual vocabulary in the transaction scenario; Combined with the characteristic vocabulary of financial transactions, after vocabulary filtering, the words related to capital flow are preferentially extracted; Based on high-risk characteristic words in historical suspicious transaction records, the extracted capital flow-related words are weighted to enhance their importance in subsequent model training; The weighted processing of the extracted capital flow-related words based on the high-risk characteristic words in the historical suspicious transaction records includes: According to the following formula 4, the historical data weight of capital flow related words is calculated : in, Words related to capital flow The number of times the suspicious transaction appears in historical records; Indicates the total number of words in historical suspicious transaction records; Based on the risk level defined internally by the bank, the modified vocabulary weight is calculated using the following formula 5 : in, is the risk correction factor; Vocabulary related to fund flow within the bank The assigned risk level, with risk level values ​​ranging from 0 to 1, representing a degree of risk from low risk to high risk; Combined with the time characteristics in the transaction records, the vocabulary weight is further adjusted according to the following formula 6 to obtain the time decay weight : in, Indicates the timestamp of the current transaction; The reference time point includes high-risk periods when suspicious transactions are concentrated; is the time decay coefficient, which is used to measure the effect of time on the importance of vocabulary; After considering historical weight, risk correction and time decay, the final word weight is calculated using the following formula 7 : in: is the adjustment factor for controlling risk and time; It is a correction term used to control the smoothing effect in specific cases to prevent the weight from being too high; is a nonlinear adjustment index used to adjust the weighted effects in extreme cases.

2. The method for automatically identifying suspicious bank transaction characteristics according to claim 1, characterized in that: The model parameters include vocabulary size, word vector dimension, context window size, number of negative samples, and learning rate.

3. The method for automatically identifying suspicious bank transaction characteristics according to claim 1, characterized in that: After generating word vectors using the Word2Vec model and verifying the model recognition accuracy, the Spearman coefficient is used to evaluate the effect of adjusting the model parameters. The Spearman coefficient is calculated according to the following formula 3: in, represents the Spearman correlation coefficient, which is used to measure the correlation of model performance before and after model parameter adjustment; represents the square of the difference between the rankings of two parameters; Indicates the training round.

4. The method for automatically identifying suspicious bank transaction characteristics according to claim 1, characterized in that: After the training is completed, the generated word vectors are verified to evaluate the accuracy of the Word2Vec model in identifying transaction features, including: By comparing the word vectors generated by the model with the word vectors of known suspicious transaction feature words, the cosine similarity between the two is calculated to evaluate the recognition accuracy of the model.

5. The method for automatically identifying suspicious bank transaction characteristics according to claim 4, characterized in that: After the training is completed, the generated word vectors are verified to evaluate the accuracy of the Word2Vec model in identifying transaction features, which also includes: If the cosine similarity between the generated word vector and the word vector of the known suspicious transaction feature vocabulary exceeds the first threshold, the model is judged to have a high recognition accuracy; if the cosine similarity is lower than the second threshold, the model parameter adjustment mechanism is triggered.

6. The method for automatically identifying suspicious bank transaction characteristics according to claim 1, characterized in that: The constructed vocabulary includes at least 5,000 transaction-related words, including banking business terms, transaction pattern words, and descriptive words of potentially suspicious transaction behaviors.

7. The method for automatically identifying suspicious bank transaction characteristics according to claim 6, characterized in that: The vocabulary is dynamically updated according to the frequency of historical transaction data, and high-frequency transaction feature words are given priority during each training to ensure that the model has good adaptability to the latest transaction patterns.

8. A device for automatically identifying suspicious bank transaction characteristics, characterized in that: include: A collection unit is configured to collect bank transaction records and standardize the text in the transaction records; wherein the transaction records include transfers, purchases, ATM withdrawals, over-the-counter withdrawals, financial products, precious metals, and loan records; the standardization includes standardizing date formats, normalizing amount units, and removing non-text symbols; a processing unit configured to perform word segmentation on the text in the standardized transaction records to extract key information words; and construct a vocabulary based on the key information words; wherein the vocabulary includes words reflecting transaction characteristics for use in subsequent model training; A construction unit is used to train a Word2Vec model using the word segmentation results of the text in the transaction record and the constructed vocabulary. The Word2Vec model is implemented using the Skip-Gram model architecture. For a central word in a given transaction record text, the Skip-Gram model is trained to generate word vectors by predicting its context words to capture the semantic relationship between the words. The probability of occurrence of each context word is calculated according to the following formula 1: in, Indicates the central word, represents the context word, where Indicates the position of the central word in the sentence. Represents context-dependent indexes; Representing context words word vectors; Indicates the central word word vectors; A word vector representing each word in the vocabulary; is the size of the exchange table; A training unit is used to train the Word2Vec model using a negative sampling method, wherein the negative sampling method reduces the computational effort of the model to predict all words by randomly selecting words that are unrelated to the central word as negative samples; wherein the loss function provided by the following formula 2 is used during the training process: in, Represents the loss function, which is used to measure the prediction error of the model during training; Represents the total number of samples in the training data; Indicates the window size; Representing context words word vectors; Indicates the central word word vectors; Indicates the total number of negative samples; Represents the distribution of negative samples Randomly sample negative samples The expected value of probability; Represents negative samples word vectors; is the sigmoid function; The verification unit is used to verify the generated word vectors after training is completed and evaluate the accuracy of the Word2Vec model in identifying transaction features; An adjustment unit, configured to adjust model parameters of the Word2Vec model according to a verification result of the Word2Vec model, to obtain an adjusted Word2Vec model capable of identifying features of suspicious bank transactions; The processing unit is specifically configured to: perform grammatical analysis on the transaction record text using a natural language processing tool, identify the part of speech of each word, and capture the central word and its related context words using a context window technique to ensure that each word has accurate semantic association in its context; Dynamically filter out irrelevant or noisy words based on the contextual vocabulary in the transaction scenario; Combined with the characteristic vocabulary of financial transactions, after vocabulary filtering, the words related to capital flow are preferentially extracted; Based on high-risk characteristic words in historical suspicious transaction records, the extracted capital flow-related words are weighted to enhance their importance in subsequent model training; The processing unit is further configured to: According to the following formula 4, the historical data weight of capital flow related words is calculated : in, Words related to capital flow The number of times the suspicious transaction appears in historical records; Indicates the total number of words in historical suspicious transaction records; Based on the risk level defined internally by the bank, the modified vocabulary weight is calculated using the following formula 5 : in, is the risk correction factor; Vocabulary related to fund flow within the bank The assigned risk level, with risk level values ​​ranging from 0 to 1, representing a degree of risk from low risk to high risk; Combined with the time characteristics in the transaction records, the vocabulary weight is further adjusted according to the following formula 6 to obtain the time decay weight : in, Indicates the timestamp of the current transaction; The reference time point includes high-risk periods when suspicious transactions are concentrated; is the time decay coefficient, which is used to measure the effect of time on the importance of vocabulary; After considering historical weight, risk correction and time decay, the final word weight is calculated using the following formula 7 : in: is the adjustment factor for controlling risk and time; It is a correction term used to control the smoothing effect in specific cases to prevent the weight from being too high; is a nonlinear adjustment index used to adjust the weighted effects in extreme cases.

Images