Method and device for inducing to cancel order risk control responsibility judgment based on text classification algorithm

Abstract

The invention relates to the technical field related to online car-hailing, in particular to a method and device for inducing to cancel order risk control responsibility judgment based on a text classification algorithm. The method comprises the following steps: acquiring text data corresponding to communication content between a driver and a passenger; based on a prompt word extraction model, analyzing and refining the text data to obtain prompt words; inputting the text data and the cue word into a preset text classification model to obtain an indication word; wherein the text classification model is a deep learning model, and is obtained by training based on preset data; wherein the indication word is used for indicating order responsibility judgment details.

Description

Technical Field

[0001] This application relates to the technical field of online car-hailing, and specifically relates to a risk control liability judgment method and device for induced order cancellation based on text classification algorithms. Background Art

[0002] In the online car-hailing industry, order cancellation is a common and complex problem. Order cancellation not only affects the operation efficiency of the platform, but may also damage user experience and satisfaction. Order cancellation may be caused by various factors, including communication problems between drivers and passengers, changes in travel routes, conflicts in time arrangements, etc. After an order is cancelled, determining whether the responsible party is the driver, the passenger, or the platform is crucial for maintaining the fairness and impartiality of the platform.

[0003] Online car-hailing platforms generally use manual review to determine the responsible party for order cancellation. Manual review requires a large amount of time and human resources. Especially in the case of a high volume of orders, the review work becomes particularly heavy and inefficient. Summary of the Invention

[0004] In view of this, embodiments of this application are committed to providing a risk control liability judgment method and device for induced order cancellation based on text classification algorithms to analyze data security more comprehensively and effectively.

[0005] This application provides a risk control liability judgment method for induced order cancellation based on text classification algorithms, including: Obtain text data corresponding to the communication content between the driver and the passenger; Analyze and refine the text data to obtain prompt words; Input the text data and the prompt words into a preset text classification model to obtain indicator words; Among them, the text classification model is a deep learning model trained based on preset data; Among them, the indicator words are used to indicate the details of order liability judgment.

[0006] In some embodiments, training the text classification model includes: Obtain text data corresponding to the communication content between a preset number of drivers and passengers and the corresponding prompt words as sample data; Annotate the sample data based on preset liability judgment rules; Train the text classification model based on the annotated sample data.

[0007] In some embodiments, the prompt words include: background description, role description, task description, liability party identification rules / criteria.

[0008] In some embodiments, obtaining the text data corresponding to the communication content between the driver and the passenger includes: Automatically obtaining the call recording text under the corresponding order through the order ID; Using ASR technology to perform ASR recognition on the call recording, and translating the audio data of the call recording into text data; Adjusting the structure of the text data to convert it into a plain text dialogue structure that marks the roles of the driver and the passenger.

[0009] In some embodiments, before automatically obtaining the call recording text under the corresponding order through the order ID, it further includes: Obtaining the order ID sent by the business system; Wherein, the business system is used to screen and cancel orders, and send the order IDs of the cancelled orders.

[0010] In some embodiments, it further includes: Sending the indication word to the business system; The business system is used to process the corresponding order based on the indication word.

[0011] This application also provides a risk control liability judgment device for inducing order cancellation based on a text classification algorithm, including: An acquisition module, configured to obtain the text data corresponding to the communication content between the driver and the passenger; analyze and refine the text data to obtain a prompt word; A liability judgment module, configured to input the text data and the prompt word into a preset text classification model to obtain an indication word; Wherein, the text classification model is a deep learning model, which is trained based on preset data; Wherein the indication word is used to indicate the details of order liability judgment.

[0012] This application also provides an electronic device, including: A processor, and a memory for storing the executable program of the processor; The processor is configured to implement the risk control liability judgment method for inducing order cancellation based on the text classification algorithm as described above by running the program in the memory.

[0013] This application also provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is run by a processor, the processor is caused to execute the above-mentioned risk control liability judgment method for inducing order cancellation based on the text classification algorithm.

[0014] The present application also provides a computer program product, including a computer program which, when executed by a processor, implements the above-mentioned risk control liability judgment method for induced order cancellation based on a text classification algorithm.

[0015] A risk control liability judgment method for induced order cancellation based on a text classification algorithm provided by the present application first obtains text data corresponding to the communication content between a driver and a passenger; analyzes and refines the text data to obtain prompt words; inputs the text data and the prompt words into a preset text classification model to obtain indicator words; wherein, the text classification model is a deep learning model trained based on preset data; and wherein, the indicator words are used to indicate the details of order liability judgment. With such a setting, in the solution provided by the present application, the text data corresponding to the communication content between the driver and the passenger is obtained through automated means, reducing manual intervention and improving the speed and efficiency of data processing. Automated text acquisition can ensure that the communication content is quickly obtained after an order cancellation event occurs, providing the possibility for timely response and processing. Using a deep learning model to analyze the text data can more deeply understand the semantics of the communication content and improve the accuracy of liability judgment. The deep learning model can identify and learn complex features and patterns in the text, which is crucial for understanding and analyzing the subtle differences in driver-passenger communication. The prompt words obtained through the text classification model directly indicate the details of order liability judgment, providing clear guidance and basis for the liability judgment process. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above-mentioned and other objects, features, and advantages of the present application will become more obvious. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 is a flowchart showing the process of a risk control liability judgment method for induced order cancellation based on a text classification algorithm provided by an embodiment of the present application.

[0018] Figure 2 is a flowchart showing the model development process of the method provided by an embodiment of the present application.

[0019] Figure 3 is a flowchart showing the automatic liability judgment process of the method provided by an embodiment of the present application.

[0020] Figure 4 is a structural diagram showing a risk control liability judgment device for induced order cancellation based on a text classification algorithm provided by an embodiment of the present application.

[0021] Figure 5It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0022] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0023] In the online car-hailing industry, order cancellation is a common and complex problem. Communication between drivers and passengers may lead to order cancellation. In this case, it is necessary to determine the responsible party for the cancelled order. Currently, this process mainly relies on manual subjective judgment by listening to the driver-passenger call recordings. This method is not only inefficient but also may be misjudged, affecting the user experience and the platform's credibility. Therefore, how to efficiently and accurately determine the responsible party for order cancellation has become an urgent problem to be solved.

[0024] Currently, online car-hailing platforms usually use manual review to determine the responsible party for order cancellation. The specific process is as follows: 1. Recording collection: When an order is cancelled, the system will record and save the relevant driver-passenger call recordings.

[0025] 2. Manual review: Special customer service personnel or review teams will listen to these recordings and make judgments based on their own experience and the rules formulated by the company.

[0026] 3. Liability determination: According to the review results, determine whether the driver or the passenger is responsible for the cancelled order.

[0027] This method has the following problems: Low efficiency: Manual review requires a large amount of time and human resources.

[0028] Strong subjectivity: Different reviewers may have different understandings of the same recording, resulting in inconsistent liability judgment results.

[0029] High risk of misjudgment: Due to human factors, misjudgment may occur, thus affecting the user experience and the platform's credibility.

[0030] To solve the above problems, the present application provides a solution that realizes more efficient and accurate liability judgment by making full use of the advantages of text classification algorithm models in context semantic understanding and logical reasoning. The specific advantages are as follows: 1. Improve the accuracy of liability judgment: The text classification algorithm model can deeply understand the semantic details in the driver-passenger communication text and perform complex logical reasoning, so as to more accurately identify the responsible party and reduce misjudgment situations.

[0031] 2. High degree of automation: Compared with the traditional method that relies on manual review, this application realizes a high degree of automation, greatly reducing the labor cost and improving the processing efficiency.

[0032] 3. Strong adaptability: The text classification algorithm model has strong learning ability and can continuously optimize its own algorithm to adapt to diverse and complex data inputs in different scenarios, improving the system robustness.

[0033] 4. Reducing human bias: By using a unified text classification algorithm model for liability determination, it can effectively avoid unfair situations caused by personal subjective factors and improve the fairness of the platform.

[0034] 5. Real-time processing ability: With the powerful computing ability of the text classification algorithm model, it can analyze and determine liability for a large number of order cancellation events in real-time or near real-time, which helps to improve the user experience. In summary, the method proposed in this application has significant advantages over existing solutions in terms of improving the accuracy of liability determination, reducing costs, enhancing system adaptability, and improving the user experience.

[0035] After introducing the basic principle of this application, the various non-limiting embodiments of this application will be specifically introduced with reference to the accompanying drawings.

[0036] Refer to Figure 1 、 Figure 2 、and Figure 3 As shown, the method provided by this application includes the following content.

[0037] Step S110, obtaining text data corresponding to the communication content between the driver and the passenger; In this step, it is necessary to obtain relevant communication content from the interaction process between the driver and the passenger and convert it into text-form data. This may involve various methods, specifically depending on the source and form of the original data.

[0038] If the communication content exists in the form of a call recording, first, it is necessary to associate it with the relevant system through the order ID to obtain the call recording text under the corresponding order. This may require calling the corresponding database or storage system to ensure that the target recording text can be accurately obtained.

[0039] Next, in order to convert the audio-form call recording into processable text data, automatic speech recognition (ASR) technology needs to be used. The ASR system will recognize and translate the speech in the recording and convert it into text information in the form of words, phrases, sentences, etc.

[0040] Since the result of ASR translation may be a relatively complex structure, such as a character-level or sentence-level JSON format, its structure needs to be adjusted. The purpose of the adjustment is to convert it into a pure text dialogue structure that can clearly mark the roles of the driver and the passenger for subsequent analysis and processing.

[0041] Step S120: Analyze and refine the text data based on the prompt word extraction model to obtain prompt words. After obtaining appropriate text data, enter the analysis and refinement stage. This step aims to extract valuable information from the text data, which will serve as an important input for the subsequent text classification model.

[0042] The analysis process may involve semantic understanding of the text, grammatical analysis, and research on the dialogue patterns between the driver and the passenger. By comprehensively examining the text data, key elements and features related to liability determination are identified.

[0043] The extracted prompt words usually cover multiple aspects. For example, background description prompt words may include background information about the order generation, such as time, location, order type, etc., which helps to understand the context in which the dialogue occurs. Role description prompt words clearly distinguish the role-related characteristics of the driver and the passenger, such as the driver's service status (whether it is working hours, whether the vehicle is normal, etc.) and the passenger's travel needs (destination, urgency of travel time, etc.). Task description prompt words may involve information related to the tasks of order execution in the dialogue, such as whether there are special requirements, itinerary changes, etc. The liability party identification rule / criterion prompt words are the most crucial part, which contains the rules and criteria summarized from the text for judging the liability party. For example, certain specific expressions (such as the driver indicating inability to serve, the passenger requesting cancellation, etc.) may point to the liability of a certain party.

[0044] Step S130: Input the text data and the prompt words into a preset text classification model to obtain indicator words. Among them, the text classification model is a deep learning model, which is trained based on preset data; among them, the indicator words are used to indicate the details of order liability determination.

[0045] In this step, the processed text data and the extracted prompt words are used as inputs and provided to a preset text classification model.

[0046] The text classification model is a deep learning model, which is trained based on a large amount of preset data. These preset data usually include the text data of the communication between the driver and the passenger with the liability party already marked, as well as information such as the corresponding prompt words. During the training process, the model learns the internal connection and law between the text data, the prompt words and the liability party.

[0047] When new text data and prompting words are input, the model will process and analyze the input based on the knowledge and algorithms it has learned. Finally, the model will output indicating words, which are an indication of the details of order liability judgment. For example, the indicating words may directly point to specific liability judgment results such as the driver being liable, the passenger being liable, or both parties being non-liable, or provide some probability values or confidence information related to liability judgment to assist further liability judgment decisions.

[0048] Furthermore, the iterative optimization of the prompting word extraction model and the text classification model includes: Construct a prompting word extraction model and a text classification model; Prompting word extraction model: The purpose of this model is to extract key prompting words from the communication text between the driver and the passenger. The prompting words are an important basis for subsequent liability judgment by the text classification model, and they may include background descriptions, role descriptions, task descriptions, liability party identification rules / standards, etc. To construct this model, appropriate algorithms and architectures need to be selected. For example, rule-based methods, machine learning algorithms, or deep learning models can be used.

[0049] Text classification model: This is a deep learning model used to determine the liability party for order cancellation based on the input text data and prompting words. When constructing the model, its architecture needs to be determined, such as convolutional neural network (CNN), recurrent neural network (RNN) and its variants (such as LSTM, GRU), etc., as well as hyperparameters such as the number of model layers, the number of neurons, and activation functions.

[0050] 2. Obtain the first sample data; First sample data: It includes a certain amount of communication text data between the driver and the passenger, as well as the corresponding manual annotations for these text data. The content of the manual annotations is used to characterize the determination method of the prompting words, that is, experts manually mark the key information in the text according to experience and rules, and these annotations will be used as the basis for training the prompting word extraction model.

[0051] Data collection: The communication text between the driver and the passenger can be obtained from historical order records, and these texts may exist in the form of chat records, call recording transcriptions, etc. Ensure that the collected data is representative and covers various possible communication scenarios and liability situations.

[0052] Manual annotation: Professional personnel annotate the collected text data to mark the prompting words in it. The annotation process needs to follow certain rules and standards to ensure the consistency and accuracy of the annotation results.

[0053] 3. Train the prompting word extraction model based on the first sample data; Training process: Input the first sample data into the prompt extraction model. The model automatically adjusts its internal parameters by learning the correspondence between the text data and the manual annotations to improve the accuracy of extracting prompts. Some optimization algorithms, such as the gradient descent method, can be used during the training process to minimize the prediction error of the model.

[0054] Model evaluation: During the training process, it is necessary to evaluate the performance of the model. Some metrics, such as accuracy, recall, F1 value, etc., can be used to measure the effect of the model in extracting prompts. The evaluation results will be used to guide the adjustment and optimization of the model.

[0055] 4. Obtain the second sample data; Second sample data: It includes the communication text data between the driver and the passenger, as well as the prompts determined based on the prompt extraction model. These data will be used to train the text classification model.

[0056] Data generation: Use the trained prompt extraction model to process the new communication text data and automatically extract the prompts. These prompts will form the second sample data together with the corresponding text data.

[0057] 5. Train the text classification model based on the second sample data Training process: Input the second sample data into the text classification model. The model adjusts its internal parameters by learning the relationship between the text data, the prompts, and the responsible party to improve the accuracy of judging the responsibility for order cancellation. Optimization algorithms can also be used during the training process to minimize the prediction error.

[0058] Model evaluation: Evaluate the trained text classification model and use some metrics, such as accuracy, recall, F1 value, etc., to measure the performance of the model. The evaluation results will be used to determine whether the model meets the preset requirements and whether further iterative training is needed.

[0059] 6. Iterative training Purpose of iterative training: Continuously optimize the prompt extraction model and the text classification model to improve their performance until the evaluation results meet the preset requirements.

[0060] Iterative steps: Modify the manual annotation content: According to the evaluation results of the model, analyze the error situations of the prompt extraction model and the text classification model, and modify and improve the manual annotation content in the first sample data. For example, if the model performs poorly on certain types of text data, the prompts for these data can be re-annotated to make them more accurate and representative.

[0061] Retrain the prompt extraction model: Use the modified first sample data to retrain the prompt extraction model. This will enable the model to learn more accurate prompt extraction rules and improve the accuracy of its prompt extraction.

[0062] Regenerate the second sample data: Use the retrained prompt extraction model to process the new communication text data and generate new prompts. These new prompts will form new second sample data together with the corresponding text data.

[0063] Retrain the text classification model: Use the new second sample data to retrain the text classification model. This will enable the model to learn more accurate relationships between text data, prompts, and responsible parties, and improve the accuracy of its determination of responsible parties.

[0064] Re-evaluate: Evaluate the retrained text classification model to assess whether its performance has improved and whether it meets the preset requirements.

[0065] Repeat iteration: If the evaluation result still does not meet the preset requirements, continue iterative training and repeat the above steps until the model performance meets the requirements.

[0066] 7. Complete model training; When the evaluation result of the text classification model meets the preset requirements, it indicates that the model already has high accuracy and can effectively determine the responsible party for order cancellation based on the communication text data and prompts between the driver and the passenger. At this time, the training processes of the prompt extraction model and the text classification model are completed, and they can be applied to the actual order cancellation liability determination scenario to provide automated and intelligent risk control liability judgment services for the online car-hailing platform.

[0067] Specifically, the prompts include: background description, role description, task description, responsible party identification rules / criteria, and liability determination result description.

[0068] The prompts are key information output by the text classification model, used to guide and explain the analysis results of the driver-passenger communication content and how to apply these results to order cancellation liability determination. The following is a detailed explanation of these prompts: 1. Background description Purpose: Provide background information on the occurrence of the communication content to help understand the context of the conversation.

[0069] Content: Includes factors such as the time, location, weather conditions, and traffic conditions of the order cancellation that may affect communication and liability determination.

[0070] 2. Role description Purpose: Clearly define the roles of the two parties in the communication, namely the driver and the passenger.

[0071] Content: Describe the behaviors and statements of each role in communication, and how these behaviors and statements are related to liability determination.

[0072] 3. Task Description Purpose: Outline the task of the model, which is to identify and classify driver-passenger communication content to assist in liability determination.

[0073] Content: Detail how the model analyzes text data, identifies key information, and maps it to predefined liability determination rules.

[0074] 4. Liability Party Identification Rules / Criteria Purpose: Provide specific rules and criteria for identifying the liability party.

[0075] Content: Include a series of predefined conditions and metrics for determining whether the driver, passenger, or platform is responsible for order cancellation. For example, if the communication content shows that the driver fails to arrive at the agreed location on time, according to the rules, this may be determined as the driver's liability.

[0076] 5. Explanation of Liability Determination Results Purpose: Explain the liability determination results output by the model and provide the basis for liability determination.

[0077] Content: Include the identified liability party, the basis for liability determination (i.e., which communication content triggered specific liability rules), and how the liability determination results affect subsequent business processes (such as compensation, credit scoring, etc.).

[0078] Through these indicators, the text classification model can not only output the results of liability determination, but also provide detailed explanations and bases, making the liability determination process clearer, more transparent, and interpretable. This is crucial for improving user satisfaction and enhancing the fairness and credibility of the platform.

[0079] In practical applications, obtaining the text data corresponding to the communication content between the driver and the passenger includes: automatically obtaining the call recording text corresponding to the order through the order ID; using ASR technology to perform ASR recognition on the call recording and translate the audio data of the call recording into text data; adjusting the structure of the text data to convert it into a plain text dialogue structure that marks the roles of the driver and the passenger.

[0080] In practical applications, obtaining the text data corresponding to the communication content between the driver and the passenger is a key step, which involves extracting data from the order system, using speech recognition technology to convert audio data into text, and further processing the text data for analysis. The following are the detailed steps of this process: Step 1: Automatically obtain the call recording text corresponding to the order through the order ID 1. Order ID Identification: After the system receives an order cancellation event, it first identifies the order ID, which is the key identifier for obtaining relevant communication content.

[0081] 2. Data Retrieval: Using the order ID, the system automatically retrieves the call recording files related to the order from the database or storage system.

[0082] 3. Data Integrity: Ensure that the retrieved call recording files are complete, including all communication content related to the order cancellation.

[0083] Step 2: Use ASR technology to perform ASR recognition on the call recording 1. Application of ASR Technology: Apply automatic speech recognition (ASR) technology to process the call recording and convert the audio data into text data.

[0084] 2. Speech Transcription: The ASR system recognizes and transcribes the speech information in the recording into text. This step may involve the application of acoustic models and language models.

[0085] 3. Transcription Accuracy: Ensure a high accuracy rate of the ASR technology to reduce transcription errors and improve the accuracy of subsequent text analysis.

[0086] Step 3: Adjust the structure of the text data 1. Structured Processing: Perform structured processing on the text data output by the ASR technology to facilitate the identification and differentiation of the driver's and passenger's speeches.

[0087] 2. Role Marking: Clearly mark the roles of the driver and passenger in the text. This may involve using specific markers or tags to distinguish the speeches of different roles.

[0088] 3. Dialogue Structure Conversion: Convert the text of the call recording into a pure text dialogue structure, which usually includes timestamps, role identifiers, and speech content, etc.

[0089] Step 4: Convert to a pure text dialogue structure with marked driver and passenger roles 1. Dialogue Parsing: Parse the text output by the ASR, identify each speech part in the dialogue, and assign it to the corresponding speaker.

[0090] 2. Role Identification: Use natural language processing (NLP) techniques, such as entity recognition or pattern matching, to identify and differentiate the driver's and passenger's speeches.

[0091] 3. Format Unification: Ensure that all dialogue texts follow a unified format so that the subsequent text classification model can correctly read and analyze them.

[0092] Benefits in practical applications Automation and Efficiency: Automated processes reduce manual intervention and improve processing speed and efficiency.

[0093] Accuracy and Consistency: Structured and role-tagged text data improves the accuracy and consistency of model analysis.

[0094] Easy to Analyze: Data converted into a pure text dialogue structure is easier to analyze and process by subsequent text classification models.

[0095] Through these steps, the system can effectively convert the communication content between the driver and the passenger into text data that can be used in the automatic liability determination process, providing accurate input for subsequent text classification and liability determination.

[0096] Furthermore, before automatically obtaining the call recording text corresponding to the order through the order ID, it further includes: Obtaining the order ID sent by the business system; wherein, the business system is used to screen and cancel orders and send the order IDs of the cancelled orders.

[0097] In some embodiments, it further includes: sending the indicator word to the business system; the business system is used to process the corresponding order based on the indicator word.

[0098] First, after obtaining the indicator word in the previous steps, it is necessary to send it to the business system. This process involves data transfer and interface calls. It may transmit the indicator word in a specific data format (such as JSON format, etc.) from the operating environment where the model is located to the server or platform where the business system is located through a network communication protocol. After receiving the indicator word, the business system will process the corresponding order based on these indicator words. If the indicator word clearly indicates that the driver is at fault, the business system may take corresponding measures, such as deducting points, fining the driver, or restricting their order-taking permissions, etc., while updating the order status to cancelled due to driver responsibility and recording relevant information in the database. If the indicator word shows that the passenger is at fault, the business system may deduct credit points from the passenger, notify the passenger to pay relevant fees (if any), and update the order status to cancelled due to passenger responsibility. If the indicator word indicates that both parties are not at fault, the business system will update the order status to normal cancellation (caused by non-responsible parties), and may also conduct statistical analysis on relevant data for subsequent macro-control and optimization of order cancellation situations.

[0099] The following describes the solutions provided by this application in combination with various preferred embodiments: Based on the initial liability determination rules / standards summarized manually, design and develop the initial prompt engineering. Use the text classification algorithm model to assist in quickly annotating and collecting the driver-passenger communication text datasets of driver responsibility, passenger responsibility, platform responsibility, and normal communication between the driver and the passenger (no responsible party).

[0100] The data misrecognized by the manual annotation model is continuously used to supplement and improve the business liability judgment standard rules / criteria. Based on this, the prompt engineering is continuously iterated to improve the accuracy and recall rate of the text classification algorithm model in identifying the responsible party, achieving the expected model metrics of a driver-responsible recall rate of over 90% and an accuracy rate of over 95%.

[0101] Finally, the scale of the induced cancellation liability judgment rules / criteria is refined as follows: Rules / criteria for determining passenger liability: 5 items. The local rules are as follows: The driver clearly states that the platform does not provide the service of delivering items and suggests that the passenger call another car, which belongs to the passenger's responsibility.

[0102] The passenger asks the driver about the taxi fare, and the passenger thinks the price is too high. The driver asks the passenger to cancel the order, which belongs to the passenger's responsibility.

[0103] The driver indicates that they have arrived, are almost there, are near or across from the boarding point, will come as soon as possible, will go as soon as possible, cannot park at the boarding point, or have arrived at the door and cannot park, indicating that the driver has the willingness to serve, which belongs to the passenger's responsibility.

[0104] The driver waits for the passenger for more than 3 minutes, which is considered an overtime wait, and asks the passenger to cancel the order or call another car, which belongs to the passenger's responsibility.

[0105] The driver states that the online car-hailing service cannot pass through / go to the place where the taxi is, resulting in the passenger canceling the order, which belongs to the passenger's responsibility and is preferentially marked as category D.

[0106] The driver indicates that they cannot enter the specified location / community of the passenger, and the passenger proposes to cancel, which belongs to the passenger's responsibility.

[0107] The driver states that the online car-hailing service cannot pass through the passenger's boarding location or the passenger has mispositioned, which belongs to the passenger's responsibility.

[0108] The passenger is on the highway and the driver cannot pick up the passenger, which belongs to the passenger's responsibility.

[0109] The passenger states that they have already left, and the driver asks the passenger to cancel the order, which belongs to the passenger's responsibility.

[0110] The passenger states that the assigned order is too far and cancels the order, which belongs to the passenger's responsibility.

[0111] The driver and the passenger communicate about the location, and finally the driver indicates that the order has been cancelled, which belongs to the passenger's responsibility.

[0112] Rules / criteria for determining driver liability: 49 items. The local rules are as follows: If the driver states that they cannot pass through or come over or cannot take the order, it is determined as the driver's responsibility and marked as A.

[0113] If the driver does not know that they have received the order, it is determined as the driver's responsibility and marked as A.

[0114] If the driver is eating or using the toilet, it is determined to be the driver's responsibility and marked as A.

[0115] If the driver indicates going to refuel, charge the vehicle, having insufficient power, or the electricity is not enough, and the driver or the passenger proposes to cancel the order, or the passenger agrees to cancel, it belongs to the driver's responsibility and is marked as A.

[0116] If the driver temporarily increases the price, charges extra, overcharges for cross-city trips, overcharges for return trips, overcharges for highway tolls, or charges additional toll fees, regardless of whether the passenger proposes to cancel or whether the passenger agrees to pay, it belongs to the driver's responsibility and is preferentially marked as A.

[0117] If the driver temporarily increases the price, charges extra, overcharges for cross-city trips, overcharges for return trips, overcharges for highway tolls, or charges additional toll fees, and the passenger agrees, it belongs to the driver's responsibility and is preferentially marked as A.

[0118] Rules / criteria for determining platform responsibility: 4 items. The partial rules are as follows: During the driver-passenger communication process, if there is a situation where the driver and the passenger cannot find each other's locations, it is determined to be the platform's responsibility and marked as D.

[0119] If the driver indicates that the passenger's boarding location is difficult to find and suggests canceling the order, it belongs to the platform's responsibility and is marked as D.

[0120] During the driver-passenger communication process, if the driver is unable to serve due to force majeure (natural disasters / harsh weather / heavy rain / waterlogging on the road surface), it belongs to the platform's responsibility.

[0121] If the driver indicates that they have arrived at or near the boarding point (by the roadside), and during the driver-passenger communication about the location, they cannot find each other, it is determined to be the platform's responsibility.

[0122] Rules / criteria for determining no responsible party: 14 items. The partial rules are as follows: If the driver indicates a long distance or road construction and is unable to arrive on time, and requests / asks the passenger to wait, it belongs to normal communication and is marked as D.

[0123] If the passenger indicates that they have not yet disembarked from the plane / left the station, and the driver indicates that they cannot wait for a long time and suggests that the passenger cancel the order, it is marked as D.

[0124] If the passenger ordered the car for someone else and the positioning is inaccurate and the driver cannot find it, and the driver suggests that the passenger cancel the order, it is marked as D.

[0125] If the driver asks the passenger if they can wait, and the passenger agrees, and there is no mention of "cancellation", "price increase", "extra charge" in the conversation, it belongs to normal communication and is marked as D.

[0126] The driver informs the passenger that there is a parking fee for entering the parking lot to pick up someone. The passenger cancels the order. The driver and the passenger are respectively in the parking lot or upstairs and downstairs of the terminal building at the airport / railway station / high-speed railway station, and they can't find each other / can't get to each other. Communicating to cancel the order / re-hailing a car belongs to normal communication.

[0127] The model development process can refer to Figure 2 , and the details are as follows: Prompt engineering development: Prompt engineering includes background description, role description, task description, responsible party identification rules / criteria, and model return result description. That is: during the process of training a text classification model, iteratively optimize the determination method of the prompt. Model selection: Using the initial version of prompt engineering, use different text classification algorithm models to test the indicators on the same batch of test sets. Determine the model with the highest indicator as the model to be used for project launch. Based on this text classification algorithm model, continuously improve the business rules / criteria and iterate the prompt engineering until the expected indicators are achieved.

[0128] Prompt iterative development: For the prompt engineering of the specified version, develop the model service, use the test set to identify the responsible party, and manually label the actual responsible party. Analyze the cases where the model misidentifies, summarize the data characteristics, and supplement and improve the business rules / criteria. Based on the improved business rules / criteria, optimize the responsible party identification rules / criteria part of the prompt engineering.

[0129] Model effect testing: Continuously use the model constructed with the latest version of the iterated prompt engineering to batch test the test set to confirm whether the expected indicators can be achieved and meet the business acceptance conditions. If it meets, the model is delivered for business acceptance; otherwise, continue to iterate.

[0130] Model effect acceptance: The business side collects a new dataset as the validation set to verify whether the delivered model effect meets the expected indicators.

[0131] Model service deployment: Deploy the model data and build a standardized service interface for use in the automatic liability determination process.

[0132] Furthermore, referring to Figure 3 , the automatic liability determination process is as follows: The business system screens the cancelled orders and passes the order ID as a parameter to the automatic liability determination service.

[0133] The service automatically obtains the call recording text (one or more) under this order through the order ID.

[0134] Use ASR technology to perform ASR recognition on the call recording and translate the audio data into text data.

[0135] The ASR translation output structure is a complex word-level and sentence-level JSON format. It is necessary to structure the ASR result data and convert it into a plain text dialogue structure that marks the roles of the driver and passenger.

[0136] Use the converted plain text content of the driver-passenger communication as the model input, and use the model to perform full-text semantic understanding and logical reasoning on the text to identify the responsible party for this order.

[0137] Return the recognition result to the business system to perform the automatic liability judgment action.

[0138] In the solution provided by this application, based on historical data, the automatic liability judgment business rules / standards for the online car-hailing industry's induced cancellation are formulated; using the semantic understanding and reasoning capabilities of the text classification algorithm model, analyze the text content of the driver-passenger communication, and accurately identify the responsible party; use the text classification algorithm model to implement the automatic liability judgment method for the induced cancellation scenario. By making full use of the advantages of the text classification algorithm model in context semantic understanding and logical reasoning, more efficient and accurate liability judgment is achieved. The specific advantages are as follows: 1. Improve the accuracy of liability judgment: The text classification algorithm model can deeply understand the semantic details in the driver-passenger communication text and perform complex logical reasoning, so as to more accurately identify the responsible party and reduce misjudgment situations.

[0139] 2. High degree of automation: Compared with the traditional method that relies on manual review, this patent realizes a high degree of automation, greatly reducing the labor cost and improving the processing efficiency.

[0140] 3. Strong adaptability: The text classification algorithm model has strong learning ability and can continuously optimize its own algorithm to adapt to diverse and complex data inputs in different scenarios, improving the system's robustness.

[0141] 4. Reduce human bias: By using a unified text classification algorithm model for liability judgment, it is possible to effectively avoid unfair situations caused by personal subjective factors and improve the fairness of the platform.

[0142] 5. Real-time processing ability: With the powerful computing ability of the text classification algorithm model, it is possible to analyze and judge a large number of order cancellation events in real-time or near real-time, which helps to improve the user experience.

[0143] In summary, the method proposed in this patent has significant advantages over the existing solutions in terms of improving the accuracy of liability judgment, reducing costs, enhancing system adaptability, and improving the user experience.

[0144] The device embodiment of this application can be used to execute the method embodiment of this application. For the details not disclosed in the device embodiment of this application, please refer to the method embodiment of this application.

[0145] Figure 4 The following is a block diagram of a risk control liability judgment device for induced order cancellation based on a text classification algorithm provided by an embodiment of the present application. As Figure 4 shown, the device includes: An acquisition module 41, configured to acquire text data corresponding to the communication content between the driver and the passenger; A liability judgment module 42, configured to input the text data into a preset text classification model to obtain a prompt word; Wherein, the text classification model is a deep learning model, which is trained based on preset data; The prompt word is used to indicate the details of order liability judgment.

[0146] Next, refer to Figure 5 to describe an electronic device according to an embodiment of the present application. Figure 5 The following shows a block diagram of an electronic device according to an embodiment of the present application.

[0147] As Figure 5 shown, the electronic device 500 includes one or more processors 510 and a memory 520.

[0148] The processor 510 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 500 to perform desired functions.

[0149] The memory 520 may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage media, and the processor 510 may run the program instructions to implement the risk control liability judgment method for induced order cancellation based on the text classification algorithm of various embodiments of the present application described above and / or other desired functions. Various contents such as category correspondence relationships may also be stored in the computer-readable storage media.

[0150] In one example, the electronic device 500 may further include: an input device 530 and an output device 540, and these components are interconnected through a bus system and / or other forms of connection mechanisms (not shown).

[0151] In addition, the input device 530 may further include, for example, a keyboard, a mouse, an interface, and so on. The output device 540 may output various information to the outside, including analysis results and the like. The output device 540 may include, for example, a display, a speaker, a printer, a communication network, and a remote output device connected thereto, and so on.

[0152] Of course, for simplicity, Figure 5 only some of the components related to the present application in the electronic device are shown, and components such as a bus, an input / output interface, and so on are omitted. In addition, according to specific application scenarios, the electronic device may further include any other appropriate components.

[0153] In addition to the above methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions that, when run by a processor, cause the processor to execute the steps in the induced cancellation order risk control liability determination method based on a text classification algorithm according to various embodiments of the present application described in the "Exemplary Method" section of this specification.

[0154] The computer program product may be written in any combination of one or more programming languages to write program code for performing the operations of the embodiments of the present application. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's device, executed as an independent software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0155] In addition, an embodiment of the present application may also be a computer-readable storage medium, on which computer program instructions are stored, and the computer program instructions, when run by a processor, cause the processor to execute the steps in the induced cancellation order risk control liability determination method based on a text classification algorithm according to various embodiments of the present application described in the "Exemplary Method" section of this specification.

[0156] The computer-readable storage medium may adopt any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may include, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CDROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0157] The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the form disclosed herein. Although several example aspects and embodiments have been discussed above, those skilled in the art will recognize some of their variations, modifications, alterations, additions, and sub-combinations.

Claims

1. A risk control and accountability method for inducing order cancellation based on text classification algorithm, characterized in that: include: Obtain text data corresponding to the communication content between the driver and the passenger; Based on the prompt word extraction model, the text data is analyzed and refined to obtain prompt words; Input the text data and the prompt word into a preset text classification model to obtain an indicator word; Wherein, the text classification model is a deep learning model, which is obtained by training based on preset data; The indicator words are used to indicate the order judgment details; Wherein, the prompt word extraction model and the text classification model are obtained by iterative optimization in advance; Iteratively optimizing the prompt word extraction model and the text classification model includes: Constructing a prompt word extraction model and the text classification model; Acquire first sample data, the first sample data comprising: manual annotations corresponding to text data corresponding to the communication contents between a first preset number of drivers and passengers; wherein the manual annotation contents are used to characterize the determination method of the prompt words; Training the prompt word extraction model based on the first sample data; Acquire second sample data; wherein the second sample data includes: text data corresponding to the communication content between the driver and the passenger and the prompt word determined based on the prompt word extraction model; Training the text classification model based on the second sample data; and evaluating the trained text classification model to obtain an evaluation result; The prompt word extraction model and the text classification model are iteratively trained; the iterative training includes: modifying the manual annotation content in the first sample data; retraining the prompt word extraction model based on the modified first sample data, regenerating the second sample data based on the retrained prompt word extraction model, retraining the text classification model, and re-evaluating; repeating the iterative training steps; until the evaluation results meet the preset requirements, the corresponding prompt word extraction model and the text classification model are trained.

2. The risk control and accountability method for inducing order cancellation based on text classification algorithm according to claim 1 is characterized in that: The prompt words include: background description, role description, task description, and responsible party identification rules / standards.

3. The risk control and accountability method for inducing order cancellation based on text classification algorithm according to claim 1 is characterized in that: The obtaining of text data corresponding to the communication content between the driver and the passenger includes: Automatically obtain the call recording text of the corresponding order through the order ID; Using ASR technology, performing ASR recognition on the call recording, and translating the audio data of the call recording into text data; The text data is structurally adjusted and converted into a plain text dialogue structure for marking the roles of driver and passenger.

4. The risk control and accountability method for inducing order cancellation based on text classification algorithm according to claim 4 is characterized in that: Before automatically obtaining the call recording text under the corresponding order through the order ID, the method further includes: Get the order ID sent by the business system; The business system is used to screen the cancelled orders and send the order IDs of the cancelled orders.

5. The risk control and accountability method for inducing order cancellation based on text classification algorithm according to claim 1 is characterized in that: Also includes: Sending the indicator to a business system; The business system is used to process corresponding orders based on the indicator words.

6. A risk control and accountability device for inducing order cancellation based on a text classification algorithm, characterized in that: include: An acquisition module is used to acquire text data corresponding to the communication content between the driver and the passenger; based on the prompt word extraction model, the text data is analyzed and refined to obtain prompt words; A judgment module, used for inputting the text data and the prompt word into a preset text classification model to obtain an indicator word; Wherein, the text classification model is a deep learning model, which is obtained by training based on preset data; The indicator words are used to indicate the order judgment details; Iteratively optimizing the prompt word extraction model and the text classification model includes: Constructing a prompt word extraction model and the text classification model; Acquire first sample data, the first sample data comprising: manual annotations corresponding to text data corresponding to the communication contents between a first preset number of drivers and passengers; wherein the manual annotation contents are used to characterize the determination method of the prompt words; Training the prompt word extraction model based on the first sample data; Acquire second sample data; wherein the second sample data includes: text data corresponding to the communication content between the driver and the passenger and the prompt word determined based on the prompt word extraction model; Training the text classification model based on the second sample data; and evaluating the trained text classification model to obtain an evaluation result; The prompt word extraction model and the text classification model are iteratively trained; the iterative training includes: modifying the manual annotation content in the first sample data; retraining the prompt word extraction model based on the modified first sample data, regenerating the second sample data based on the retrained prompt word extraction model, retraining the text classification model, and re-evaluating; repeating the iterative training steps; until the evaluation results meet the preset requirements, the corresponding prompt word extraction model and the text classification model are trained.

7. An electronic device, characterized in that: include: A processor, and a memory for storing a program executable by the processor; The processor is used to implement the risk control and accountability method for inducing order cancellation based on a text classification algorithm as described in any one of claims 1 to 6 by running the program in the memory.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, enables the processor to execute the risk control and accountability method for inducing order cancellation based on a text classification algorithm as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, which, when executed by a processor, implements the risk control and accountability method for inducing order cancellation based on a text classification algorithm as described in any one of claims 1 to 5.

Images