Commodity auction selling and bidding transaction conversion method based on pattern recognition

Abstract

The invention relates to the technical field of commodity auction transaction, discloses a commodity auction and bidding transaction conversion method based on mode recognition, and aims to overcome the limitation of a traditional transaction mode. The method comprises the following steps: firstly, constructing a price monitoring model based on a recurrent neural network (RNN), monitoring a price dynamic state in real time, and capturing time sequence characteristics; and when the price reaches a preset conversion threshold value, triggering a transaction mode conversion model based on rule matching, processing a buyer-seller relationship and performing mode switching. Meanwhile, in a bidding mode, a bidding prediction model based on a gradient boosting decision tree GBDT is started to predict a next round of bidding so as to improve bidding efficiency. According to the invention, through combination of the deep learning algorithm and the machine learning model, intelligentization and adaptive adjustment of the transaction mode are realized, changes of market dynamics and participant demands can be accurately captured, and defects of a traditional transaction mode are avoided. The method is suitable for e-commerce and online auction markets, and can significantly improve transaction efficiency and user satisfaction.

Description

Technical Field

[0001] The present invention relates to the field of commodity auction transactions, and in particular to a method for commodity auction transaction conversion based on pattern recognition. Background Art

[0002] With the booming development of e-commerce and online auction markets, the commodity auction transaction model has gradually become one of the mainstream transaction methods in the market. Traditional auction transactions usually rely on fixed rules and timetables. However, in actual applications, this fixed model often cannot fully reflect the rapid changes in market dynamics and participant needs. Especially in an environment with large price fluctuations and complex market conditions, the traditional transaction model may lead to low transaction efficiency, low participant satisfaction, and even missed the best transaction opportunity.

[0003] In order to overcome these limitations, the industry has begun to explore more intelligent and adaptive trading models. Among them, trading conversion methods based on pattern recognition have gradually emerged. Pattern recognition technology can discover the laws and trends in price changes through real-time analysis and mining of market data, thereby providing decision support for the dynamic adjustment of trading models. Existing trading conversion methods based on pattern recognition still have many shortcomings. Existing price monitoring models often use simple threshold judgments or statistical methods, which are difficult to accurately capture the time series characteristics and complex patterns of price changes. This leads to the model being unable to make accurate judgments in time when prices fluctuate violently or market conditions are abnormal, thus affecting the timely conversion of trading modes; existing trading mode conversion models usually handle the relationship between buyers and sellers based on fixed rules, lacking flexibility and adaptability. In actual transactions, the roles and needs of buyers and sellers may change with changes in market conditions, and fixed rules often cannot fully meet such dynamic needs.

[0004] In the bidding mode, the accuracy of bid prediction is crucial to improving transaction efficiency and participant satisfaction. Existing bid prediction models often rely on simple statistical methods or machine learning algorithms, which are difficult to fully consider multi-dimensional feature data such as historical bid records, current bids, bidder information, etc., resulting in inaccurate and unreliable prediction results. Summary of the invention

[0005] The purpose of the present invention is to provide a method for commodity auction and purchase transaction conversion based on pattern recognition to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for commodity auction transaction conversion based on pattern recognition, the method comprising:

[0007] A1: Build a price monitoring model based on pattern recognition. This model uses the recurrent neural network (RNN) structure in the deep learning algorithm to monitor the price dynamics during the auction in real time. Its structure includes:

[0008] Input layer: receives real-time price data during the auction process;

[0009] Hidden layer: contains multiple RNN units, which are used to process the input price data sequence and capture the temporal characteristics of price changes;

[0010] Output layer: outputs the judgment result of whether the price reaches the preset conversion threshold;

[0011] A2: Build a transaction mode conversion model based on rule matching. This model is triggered when the output of the price monitoring model reaches the conversion threshold. It is used to handle the buyer-seller relationship, save the current bidding record of the auction mode, and initialize the new bidding record of the auction mode. Its rules include:

[0012] Identify situations where prices fall into negative territory;

[0013] Automatically switch the auction mode from selling to buying;

[0014] Adjusting buyer and seller roles;

[0015] A3: Build a bid prediction model based on the gradient boosting decision tree (GBDT). This model is enabled in the bidding mode and is used to predict the bids for the next round. Its structure includes:

[0016] Input layer: receives historical bidding records, current bidding, bidder information and other feature data;

[0017] Multiple weak learners: Decision trees are used as base learners to enhance prediction capabilities by gradually fitting residuals;

[0018] Output layer: outputs the predicted next round of bids;

[0019] A4: In the auction mode, receive user bids, implement price reduction strategies, and input price data into the price monitoring model;

[0020] A5: When the price monitoring model determines that the price reaches the conversion threshold, the transaction mode conversion model is triggered to switch modes and process the buyer-seller relationship;

[0021] A6: In the bidding mode, use the bid prediction model to predict the next round of bids and record the new bids;

[0022] A7: After the bidding ends, the winning bidder will be confirmed based on the bidding records in the bidding mode.

[0023] Preferably, the training step of the price monitoring model includes:

[0024] A101: Collect historical auction data, including price data, auction time and auction results;

[0025] A102: Preprocess the collected data, including removing outliers and standardizing them;

[0026] A103: Divide the preprocessed data into training set and validation set;

[0027] A104: Use the training set to train the RNN model and optimize the model parameters through the back propagation algorithm;

[0028] A105: Use the validation set to evaluate the model performance until the model reaches the preset accuracy in determining whether the price reaches the conversion threshold;

[0029] A106: Save the trained model parameters.

[0030] Preferably, the rules of the transaction mode conversion model are determined by the following steps:

[0031] A201: Analyze historical auction data to identify cases of trading mode transition when prices fall to negative numbers;

[0032] A202: Extract common features and conversion logic in cases;

[0033] A203: Encode the extracted features and logic into rules to form a rule set for the transaction mode conversion model.

[0034] Preferably, the structure of the transaction mode conversion model includes:

[0035] Input layer: receives the output of the pattern recognition model and the current auction status;

[0036] Judgment layer: Determine whether the bidding mode needs to be switched based on the output results of the pattern recognition model;

[0037] Processing layer: When conversion is required, it handles the buyer-seller relationship, saves existing bid records, and initializes new bid records;

[0038] Output layer: Outputs the converted bidding mode and new bidding records.

[0039] Preferably, the training step of the bid prediction model includes:

[0040] A301: Collect historical bidding data, including bidding time, bidding amount, current bidder’s bidding history, product information, other bidders’ bidding behavior and product scarcity;

[0041] A302: Perform feature engineering on the collected data to extract relevant features. The specific method is: for each feature in the collected data, calculate its Pearson correlation coefficient. The formula is as follows:

[0042]

[0043] Among them, r represents the Pearson correlation coefficient between feature x and feature y, X i and Y i Respectively represent the values ​​of feature x and feature y on the i-th sample, and They represent the mean of feature x and feature y respectively, and n represents the number of samples;

[0044] According to the calculated Pearson correlation coefficient, several candidate features with the highest absolute value of correlation coefficient are selected as features that have an important impact on bid prediction;

[0045] A303: Divide the feature data and the corresponding bidding results into a training set and a validation set;

[0046] A304: Use the training set to train the GBDT model and optimize the model performance through the gradient boosting algorithm;

[0047] A305: Use the validation set to evaluate the model prediction accuracy;

[0048] A306: Save the trained model parameters.

[0049] Preferably, after the bidding is completed, the steps of confirming the winning bidder include:

[0050] According to the bidding records in the auction mode, filter out the highest bid or the bid that meets specific conditions;

[0051] Confirm the highest bidder or the bidder that meets certain conditions as the winning bidder;

[0052] Generate a winning bid notification and send it to the winning bidder and all users who participated in the auction.

[0053] Preferably, when the bidding mode is switched to the auction mode, or the auction mode is switched to the bidding mode, the method further includes:

[0054] Notify all participants that the mode has changed, and update the trading interface to display the current bidding mode and bid records;

[0055] Participants are allowed to continue bidding according to the new bidding model until the auction ends.

[0056] Preferably, the method further includes an exception handling step for handling abnormal situations during the auction process, including network failures and system errors.

[0057] Preferably, the exception handling step includes:

[0058] Monitor the operation status of the auction system in real time and detect abnormal signals;

[0059] When an abnormal signal is detected, the auction will be suspended and the abnormal information will be recorded;

[0060] Take appropriate measures based on the type of abnormal information, including restarting the system and restoring data;

[0061] After the exception is handled, the auction is resumed and the subsequent auction process continues.

[0062] Preferably, in step A102, the Z-score method is used to standardize the data, and the mean of each numerical data is subtracted and divided by its standard deviation, so that the processed data conforms to the standard normal distribution. The formula for standardization is: Z = (X-μ) / σ, wherein X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] By constructing a price monitoring model based on a recurrent neural network (RNN), the present invention can capture the time series characteristics and complex patterns of price changes in real time, effectively solving the problem of inaccurate judgment of traditional price monitoring models when prices fluctuate violently or the market is abnormal. The RNN structure can process the input price data sequence, deeply explore the inherent laws of price changes, ensure that accurate judgments are made in a timely manner at critical moments, and provide strong support for the timely conversion of trading modes. The trading mode conversion model based on rule matching proposed in the present invention is automatically triggered when the output of the price monitoring model reaches the conversion threshold, realizing flexible handling of the buyer-seller relationship. The model can identify special situations such as prices falling to negative numbers, and automatically switch the auction mode, adjust the roles of buyers and sellers, meet the needs of dynamic adjustment of the roles and demands of buyers and sellers in actual transactions with market changes, and improve the flexibility and adaptability of transactions.

[0065] By constructing a bid prediction model based on a gradient boosted decision tree (GBDT), the present invention can fully consider multi-dimensional feature data such as historical bid records, current bids, bidder information, etc. in the bidding mode, effectively improving the accuracy of bid prediction. The GBDT model enhances the prediction ability by gradually fitting the residuals, ensuring the reliability and stability of the prediction results, providing traders with a more accurate bidding reference, and improving transaction efficiency.

[0066] The present invention organically combines the three key links of price monitoring, transaction mode conversion and bid prediction to form a complete transaction conversion method. In the auction mode, user bids are received in real time and price reduction strategies are implemented, while monitoring price dynamics; when the conversion threshold is reached, it automatically switches to the bidding mode and enables the bid prediction model; after the bidding is completed, the winning bidder is confirmed based on the bid record in the bidding mode. This process optimization reduces redundant links in the transaction process, improves transaction efficiency, and also improves the satisfaction and experience of participants. The implementation of the present invention helps to improve the transaction transparency and fairness of the e-commerce and online auction markets, and provides a more efficient and convenient trading environment for market participants through intelligent and adaptive transaction conversion methods. This not only promotes the activity and liquidity of market transactions, but also helps to improve the overall competitiveness and sustainable development capabilities of the market. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A diagram showing the processing steps of a method for converting commodity auction transactions based on pattern recognition according to the present invention;

[0068] Figure 2 Construct a flow chart for the transaction mode conversion model based on rule matching;

[0069] Figure 3 This is a flowchart for confirming the winning bidder and handling exceptions after the bidding ends. DETAILED DESCRIPTION

[0070] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0071] See also Figure 1-3 The present invention provides a technical solution: a method for commodity auction transaction conversion based on pattern recognition, the method comprising:

[0072] A1. Construct a price monitoring model based on pattern recognition: This step uses the recurrent neural network (RNN) structure in the deep learning algorithm, which is designed to monitor the price dynamics during the auction process in real time. The specific structure includes:

[0073] Input layer: configured to receive real-time price data during the auction process. This data is collected by the system in real time and input into the model.

[0074] Hidden layer: The design contains multiple RNN units, which can process the input price data sequence and capture the timing characteristics of price changes, so as to understand the dynamic laws of price changes.

[0075] Output layer: Set to output the judgment result of whether the price reaches the preset conversion threshold. The result will serve as the basis for subsequent transaction mode conversion.

[0076] A2. Build a transaction mode conversion model based on rule matching. When the output of the price monitoring model reaches the conversion threshold, this model is triggered. This model is used to handle the buyer-seller relationship, save the current bidding record of the auction mode, and initialize the new bidding record of the auction mode. Specific rules include:

[0077] Identify situations where the price drops to negative values ​​as one of the triggers for a mode switch.

[0078] When the triggering conditions are met, the bidding mode is automatically switched from selling to buying, realizing dynamic conversion of modes.

[0079] According to the needs of model transformation, adjust the roles of buyers and sellers to adapt to the new trading model.

[0080] A3. Build a bid prediction model based on the gradient boosted decision tree (GBDT). This model is enabled in the bidding mode and is used to predict the bids for the next round. The specific structure includes:

[0081] Input layer: It is configured to receive feature data such as historical bid records, current bids, bidder information, etc. These data provide a rich information basis for the model's prediction.

[0082] Multiple weak learners: Decision trees are used as base learners, and the prediction ability of the model is enhanced by gradually fitting the residuals to improve the accuracy of the prediction.

[0083] Output layer: It is set to output the predicted next round of bids to provide a basis for decision-making during the bidding process.

[0084] A4. In the auction mode, the following operations are performed: receiving user bids and adjusting prices according to the preset price reduction strategy. Inputting real-time price data into the price monitoring model to monitor price dynamics in real time.

[0085] A5. When the price monitoring model determines that the price reaches the conversion threshold, the following operations are performed: the transaction mode conversion model is triggered to switch the auction mode and process the buyer-seller relationship; the bid record of the current auction mode is saved, and a new bid record of the auction mode is initialized.

[0086] A6. In the bidding mode, perform the following operations: enable the bid prediction model to predict the next round of bids based on historical bid records, current bids, bidder information and other feature data; record new bids and update bid records.

[0087] A7.After the bidding is over, perform the following operations: confirm the winning bidder based on the bidding records in the bidding mode and complete the transaction process.

[0088] The present invention will be further described below in conjunction with Examples 1 to 4:

[0089] Embodiment 1:

[0090] When constructing a price monitoring model based on pattern recognition, the training step of the model has a very important impact on the performance of the model. This embodiment is used to elaborate on the specific implementation of the training step, which specifically includes:

[0091] A101. Data collection: Collect historical auction data and ensure that the data covers key information such as price data, auction time and auction results. These data will serve as the basis for model training to capture the laws and trends of price changes.

[0092] A102. Data preprocessing: Preprocess the collected data to ensure data quality and consistency. The preprocessing steps include:

[0093] Remove outliers: Identify and remove outliers in the data through statistical analysis and data cleaning techniques to avoid these outliers from having a negative impact on model training.

[0094] Standardization: The Z-score method is used to standardize the data. Specifically, each numerical data is subtracted from its mean and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is:

[0095]

[0096] Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0097] A103. Data partitioning: Divide the preprocessed data into a training set and a validation set. The training set is used for model training, and the validation set is used for model performance evaluation. Data partitioning should follow the principles of randomness and representativeness to ensure that the training set and validation set can fully reflect the characteristics of the overall data.

[0098] A104. Model training: Use the training set to train the RNN model. Optimize the model parameters through the back propagation algorithm so that the model can accurately capture the timing characteristics and laws of price changes. During the training process, attention should be paid to the convergence of the model and the training efficiency, and the training parameters and strategies should be adjusted in a timely manner.

[0099] A105. Model evaluation: Use the validation set to evaluate the performance of the model. By calculating the accuracy of the model in determining whether the price reaches the conversion threshold, the model's prediction and generalization capabilities are evaluated. If the model performance does not reach the preset accuracy, it is necessary to return to step A104 to continue optimizing the model parameters and training strategy.

[0100] A106. Model saving: When the model reaches the preset accuracy on the validation set, the trained model parameters are saved. These parameters will serve as the core part of the price monitoring model, used to monitor the price dynamics during the auction in real time and provide decision support for the dynamic adjustment of the transaction model.

[0101] Embodiment 2:

[0102] When constructing a transaction mode conversion model based on rule matching, the methods for determining transaction rules include:

[0103] A201. Analyze historical data: Conduct in-depth analysis of historical auction data, with special attention paid to cases of trading mode conversion when prices fell to negative numbers. Such cases are an important basis for formulating conversion rules and can reflect the market's response in different situations and the logic of trading mode conversion.

[0104] A202. Extract features and logic: Extract common features and conversion logic from the identified cases. These features include the magnitude, duration, and market participation of the price drop to negative numbers, while the conversion logic involves the adjustment of buyer and seller roles, the preservation and initialization of bid records, etc.

[0105] A203. Encoding rules: Encode the extracted features and logic into specific rules to form a rule set for the transaction mode conversion model. These rules are clear and can guide the model to make correct conversion decisions in specific situations.

[0106] The structure design of the transaction mode conversion model is as follows:

[0107] Input layer: It is configured to receive the output of the pattern recognition model and the current auction status. This information is the basis for the model to make conversion judgments.

[0108] Judgment layer: Based on the output of the pattern recognition model and the preset rule set, it is determined whether the bidding mode needs to be switched. The judgment layer is the core of the model and is responsible for making switching decisions.

[0109] Processing layer: When the judgment layer determines that a conversion is needed, it processes the buyer-seller relationship, saves the existing bid records, and initializes new bid records. The processing layer is used to ensure the smooth progress of the conversion process and the accurate recording of data.

[0110] Output layer: Outputs the converted bidding mode and new bidding records for subsequent processing and use by the system.

[0111] Processing after mode conversion: When the bidding mode is switched to the auction mode, or the auction mode is switched to the bidding mode, the following operations are performed:

[0112] Notify all participants: Promptly notify all participants that the mode has changed through system messages or interface prompts to ensure transparent and timely communication of information.

[0113] Update the trading interface: According to the converted auction mode, update the display content of the trading interface, including the current bidding mode and bid records, so that participants can clearly understand the current situation.

[0114] Allow continued bidding: After updating the trading interface, participants are allowed to continue bidding according to the new bidding model until the bidding ends, ensuring the continuity and fairness of the transaction.

[0115] Embodiment 3:

[0116] This embodiment describes in detail the specific implementation of the bid prediction model training steps, which specifically include:

[0117] A301. Data collection: Collect historical bidding data to ensure that the data covers key elements such as bidding time, bidding amount, current bidder's bidding history, product information, other bidders' bidding behavior, and product scarcity. These data will serve as the basis for model training to capture the characteristics and patterns of bidding behavior.

[0118] A302. Feature Engineering: Perform feature engineering on the collected data to extract features that have a significant impact on bid prediction. The specific methods are as follows:

[0119] Calculate the Pearson correlation coefficient: For each feature in the collected data, use the Pearson correlation coefficient formula to calculate its correlation with other features. The Pearson correlation coefficient formula is:

[0120]

[0121] Among them, r represents the Pearson correlation coefficient between feature x and feature y, X i and Y i Respectively represent the values ​​of feature x and feature y on the i-th sample, and They represent the mean of feature x and feature y respectively, and n represents the number of samples.

[0122] Feature selection: Based on the calculated Pearson correlation coefficient, select several candidate features with the highest absolute value of the correlation coefficient. These features are considered to be closely related to bid prediction and can effectively improve the prediction performance of the model.

[0123] A303. Data partitioning: The feature data processed by feature engineering and the corresponding bidding results are divided into training set and validation set. The training set is used for model training, and the validation set is used for model performance evaluation. Data partitioning should follow the principles of randomness and representativeness to ensure that the training set and validation set can fully reflect the characteristics of the overall data.

[0124] A304. Model training: Use the training set to train the GBDT (Gradient Boosting Decision Tree) model. The model performance is gradually optimized through the gradient boosting algorithm, so that the model can accurately capture the complex characteristics and rules of bidding behavior. During the training process, attention should be paid to the model's convergence, training efficiency, and overfitting problems, and the training parameters and strategies should be adjusted in a timely manner.

[0125] A305. Model evaluation: Use the validation set to evaluate the prediction accuracy of the model. By calculating the error between the prediction results of the model on the validation set and the actual bidding results, the prediction ability and generalization ability of the model are evaluated. If the model prediction accuracy does not meet the expected requirements, it is necessary to return to step A304 to continue optimizing the model parameters and training strategy.

[0126] A306. Model saving: When the model achieves satisfactory prediction accuracy on the validation set, the trained model parameters are saved. These parameters will serve as the core part of the bid prediction model, used to predict bidding behavior in the auction process in real time, and provide strong support for transaction decisions.

[0127] Embodiment 4:

[0128] This embodiment describes how to ensure that the winning bidder is accurately and efficiently confirmed after the bidding is completed, and effectively handle abnormal situations during the bidding process. The detailed steps are as follows:

[0129] The steps to identify the successful bidder include:

[0130] A401. Filtering bid records: Based on the bid records in the auction mode, use the sorting algorithm or specific condition filtering logic to filter out the highest bid or the bid that meets specific conditions from all bids. Specific conditions may include the bid reaching a certain preset threshold, the bid time meeting specific requirements, etc.

[0131] A402 Confirmation of the Winning Bidder: Based on the highest bid or bid that meets certain conditions, the corresponding bidder is confirmed as the winning bidder. This step may involve a re-examination of the bidder's qualifications to ensure that it meets the auction requirements.

[0132] A403. Generate and send winning notice: Automatically generate a winning notice, which should include key content such as the winning bidder’s information, winning bid amount, winning bid conditions, etc. Subsequently, the winning notice will be sent to the winning bidder and all users participating in the auction via email, SMS, system messages, etc., to ensure transparency and timely communication of information.

[0133] Exception handling methods include:

[0134] Real-time monitoring and anomaly detection: Real-time monitoring of the operation status of the auction system, including key indicators such as network status, server load, database connection, etc. Detect abnormal signals such as network delays, system crashes, data errors, etc. using anomaly detection algorithms or preset anomaly thresholds.

[0135] Suspend bidding and record abnormalities: Once an abnormal signal is detected, the bidding process will be suspended immediately to avoid the impact of the abnormality on the bidding results. At the same time, the abnormal information will be recorded in detail, including the time, type, and scope of impact of the abnormality, to provide a basis for subsequent processing.

[0136] Exception handling and recovery: Analyze the exception type based on the recorded exception information and take corresponding handling measures. For network failures, it may involve restarting network equipment and optimizing network configuration; for system errors, it may require system restart, data recovery or patch application. Ensure that the security and integrity of data are guaranteed during the processing.

[0137] Resume bidding: After the exception is handled, conduct a comprehensive test on the bidding system to ensure that all functions are restored to normal. Then, resume the bidding process and continue the subsequent bidding process, including bidding, confirming the winning bidder, etc. For auctions that are suspended due to exceptions, the bidding time should be adjusted reasonably to ensure that all participants have sufficient time to bid and make decisions.

[0138] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0139] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for commodity auction and purchase transaction conversion based on pattern recognition, characterized in that: The method comprises: A1: Build a price monitoring model based on pattern recognition. This model uses the recurrent neural network (RNN) structure in the deep learning algorithm to monitor the price dynamics during the auction in real time. Its structure includes: Input layer: receives real-time price data during the auction process; Hidden layer: contains multiple RNN units, which are used to process the input price data sequence and capture the temporal characteristics of price changes; Output layer: outputs the judgment result of whether the price reaches the preset conversion threshold; A2: Build a transaction mode conversion model based on rule matching. This model is triggered when the output of the price monitoring model reaches the conversion threshold. It is used to handle the buyer-seller relationship, save the current bidding record of the auction mode, and initialize the new bidding record of the auction mode. Its rules include: Identify situations where prices fall into negative territory; Automatically switch the auction mode from selling to buying; Adjusting buyer and seller roles; A3: Build a bid prediction model based on the gradient boosting decision tree (GBDT). This model is enabled in the bidding mode and is used to predict the bids for the next round. Its structure includes: Input layer: receives historical bidding records, current bidding, bidder information and other feature data; Multiple weak learners: Decision trees are used as base learners to enhance prediction capabilities by gradually fitting residuals; Output layer: outputs the predicted next round of bids; A4: In the auction mode, receive user bids, implement price reduction strategies, and input price data into the price monitoring model; A5: When the price monitoring model determines that the price reaches the conversion threshold, the transaction mode conversion model is triggered to switch modes and process the buyer-seller relationship; A6: In the bidding mode, use the bid prediction model to predict the next round of bids and record the new bids; A7: After the bidding ends, the winning bidder will be confirmed based on the bidding records in the bidding mode.

2. A method for converting commodity auction transactions based on pattern recognition according to claim 1, characterized in that: The training steps of the price monitoring model include: A101: Collect historical auction data, including price data, auction time and auction results; A102: Preprocess the collected data, including removing outliers and standardizing them; A103: Divide the preprocessed data into training set and validation set; A104: Use the training set to train the RNN model and optimize the model parameters through the back propagation algorithm; A105: Use the validation set to evaluate the model performance until the model reaches the preset accuracy in determining whether the price reaches the conversion threshold; A106: Save the trained model parameters.

3. The method for converting commodity auction transactions based on pattern recognition according to claim 1, characterized in that: The rules of the transaction mode conversion model are determined by the following steps: A201: Analyze historical auction data to identify cases of trading mode transition when prices fall to negative numbers; A202: Extract common features and conversion logic in cases; A203: Encode the extracted features and logic into rules to form a rule set for the transaction mode conversion model.

4. A method for converting commodity auction transactions based on pattern recognition according to claim 3, characterized in that: The structure of the transaction mode conversion model includes: Input layer: receives the output of the pattern recognition model and the current auction status; Judgment layer: Determine whether the bidding mode needs to be switched based on the output results of the pattern recognition model; Processing layer: When conversion is required, it handles the buyer-seller relationship, saves existing bid records, and initializes new bid records; Output layer: Outputs the converted bidding mode and new bidding records.

5. The method for converting commodity auction transactions based on pattern recognition according to claim 1, characterized in that: The training steps of the bid prediction model include: A301: Collect historical bidding data, including bidding time, bidding amount, current bidder’s bidding history, product information, other bidders’ bidding behavior and product scarcity; A302: Perform feature engineering on the collected data to extract relevant features. The specific method is: for each feature in the collected data, calculate its Pearson correlation coefficient. The formula is as follows: Among them, r represents the Pearson correlation coefficient between feature x and feature y, X i and Y i Respectively represent the values ​​of feature x and feature y on the i-th sample, and They represent the mean of feature x and feature y respectively, and n represents the number of samples; According to the calculated Pearson correlation coefficient, several candidate features with the highest absolute value of correlation coefficient are selected as features that have an important impact on bid prediction; A303: Divide the feature data and the corresponding bidding results into a training set and a validation set; A304: Use the training set to train the GBDT model and optimize the model performance through the gradient boosting algorithm; A305: Use the validation set to evaluate the model prediction accuracy; A306: Save the trained model parameters.

6. The method for converting commodity auction transactions based on pattern recognition according to claim 1, characterized in that: After the bidding is over, the steps to confirm the winning bidder include: According to the bidding records in the auction mode, filter out the highest bid or the bid that meets specific conditions; Confirm the highest bidder or the bidder that meets certain conditions as the winning bidder; Generate a winning bid notification and send it to the winning bidder and all users who participated in the auction.

7. The method for converting commodity auction transactions based on pattern recognition according to claim 1, characterized in that: When the bidding mode is switched to the auction mode, or the auction mode is switched to the bidding mode, the method further includes: Notify all participants that the mode has changed, and update the trading interface to display the current bidding mode and bid records; Participants are allowed to continue bidding according to the new bidding model until the auction ends.

8. The method for converting commodity auction transactions based on pattern recognition according to claim 7, characterized in that: The method also includes an exception handling step for handling abnormal situations during the auction process, including network failures and system errors.

9. The method for converting commodity auction transactions based on pattern recognition according to claim 8, characterized in that: The exception handling steps include: Monitor the operation status of the auction system in real time and detect abnormal signals; When an abnormal signal is detected, the auction will be suspended and the abnormal information will be recorded; Take appropriate measures based on the type of abnormal information, including restarting the system and restoring data; After the exception is handled, the auction is resumed and the subsequent auction process continues.

10. The method for converting commodity auction transactions based on pattern recognition according to claim 2, characterized in that: In step A102, the Z-score method is used to standardize the data. The mean of each numerical data is subtracted and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

Images