Form completion method and system, device, storage medium, and program product

Abstract

The embodiments of the present disclosure provide a form completion method and system, a device, a storage medium, and a program product. The method comprises the following steps: displaying a target interface, the target interface comprising a form to be completed, and the form to be completed comprising a first item to be completed and a first control; in response to a trigger operation on the first control, displaying, on the target interface, a first question raised to a virtual assistant; and displaying, on the target interface, a first reply of the virtual assistant to the first question, the first reply comprising first recommendation information for completing the first item to be completed. The technical solution provided by the embodiments of the present disclosure improves form completion efficiency.

Description

Form filling methods, systems, devices, storage media, and program products

[0001] This disclosure claims priority to Chinese Patent Application No. 202411523227.4, filed with the China Patent Office on October 29, 2024, entitled “Form Filling Method, System, Device, Storage Medium and Program Product”, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This disclosure relates to the field of Internet technology, and in particular to a form filling method, system, device, storage medium, and program product. Background Technology

[0003] With the continuous development of internet technology, internet applications have become widespread. More and more users are able to conduct various activities online, and in this process, they will encounter situations where they need to fill out online forms.

[0004] For example, in online trade (especially online international trade), when a buyer files a chargeback request with a bank for an order, the bank will require the merchant to provide evidence and defenses against the chargeback request within a specified time. The bank will review the evidence provided by the merchant and make a final ruling. To assist merchants in providing evidence and defenses, e-commerce platforms provide merchants with online forms for collecting defense materials, which the merchants fill out and then use to file a chargeback request with the bank.

[0005] However, online forms are currently inefficient to fill out. Summary of the Invention

[0006] In view of the above problems, this disclosure is made to provide a form filling method, system, device, storage medium and program product that solves or at least partially solves the above problems.

[0007] A first aspect of this disclosure provides a form completion method suitable for a terminal, including:

[0008] Display a target interface, the target interface including a form to be filled out, the form including a first field to be filled out and a first control;

[0009] In response to a triggering operation on the first control, a first question posed to the virtual assistant is displayed on the target interface;

[0010] The target interface displays the virtual assistant's first response to the first question, which includes first recommended information for filling in the first field to be filled.

[0011] A second aspect of this disclosure provides a form filling method applicable to the server side, including:

[0012] The receiving terminal sends a first target request, which includes first indication information and second prompt information. The first indication information is used to indicate the first item to be filled in of the form to be filled in, and the second indication information is used to indicate the target object. The form to be filled in is used to collect information related to the target object.

[0013] Based on the target object, a large language model is used to determine the first recommended information to fill in the first field to be filled;

[0014] The first recommendation information is returned to the terminal.

[0015] A third aspect of this disclosure provides a form filling system, comprising: a terminal and a server; wherein,

[0016] The terminal is configured to: display a target interface, the target interface including a form to be filled out, the form to be filled out being used to collect information related to a target object, the form to be filled out including a first field to be filled out and a first control; in response to a trigger operation on the first control, send a first target request to the server and display a first question to be asked of the virtual assistant on the target interface, the first target request including first instruction information and second instruction information, the first instruction information being used to indicate the first field to be filled out, the second instruction information being used to indicate the target object, and the first target request being used to request the server to provide the first recommendation information based on the target object using the large language model;

[0017] The server is configured to: receive a first target request sent by the terminal; determine the first recommendation information based on the target object using a large language model; and return the first recommendation information to the terminal.

[0018] The terminal is further configured to: receive the first recommendation information returned by the server; and display the first response from the virtual assistant to the first question on the target interface, wherein the first response includes the first recommendation information.

[0019] A fourth aspect of this disclosure provides an electronic device. The electronic device includes: a memory and a processor, wherein,

[0020] The memory is used to store programs;

[0021] The processor, coupled to the memory, is configured to execute the program stored in the memory to implement the method described in any of the preceding embodiments.

[0022] A fifth aspect of this disclosure provides a computer-readable storage medium storing a computer program that, when executed by a computer, enables the implementation of the methods described in any of the preceding claims.

[0023] A sixth aspect of this disclosure provides a computer program product including a computer program that, when executed by a processor, implements the method described in any of the preceding claims.

[0024] This disclosure provides an accompanying data entry scheme, in which a virtual assistant (or virtual robot) provides relevant entry suggestions for the fields to be filled in the form during the user's form entry process, thereby improving entry efficiency and reducing entry time. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 is a schematic diagram of model iteration provided in an embodiment of this disclosure;

[0027] Figure 2 is a flowchart illustrating a form filling method provided in an embodiment of this disclosure;

[0028] Figure 3 is a schematic diagram of the chargeback details interface provided in an embodiment of this disclosure;

[0029] Figure 4 is a schematic diagram of a form filling interface provided in an embodiment of this disclosure;

[0030] Figure 5 is a flowchart illustrating a form filling method provided in another embodiment of this disclosure;

[0031] Figure 6 is a schematic diagram of the structure of a form filling system provided in another embodiment of this disclosure;

[0032] Figure 7 is a schematic diagram of the form filling scheme provided in an embodiment of this disclosure;

[0033] Figure 8 is a structural block diagram of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0034] To enable those skilled in the art to better understand the present disclosure, the technical solutions of the present disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are merely some embodiments of the present disclosure, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present disclosure without creative effort are within the scope of protection of the present disclosure.

[0035] Furthermore, some processes described in the specification, claims, and accompanying drawings of this disclosure include multiple operations that appear in a specific order. These operations may be performed out of order or in parallel. Operation numbers such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be performed sequentially or in parallel. It should be noted that the terms "first," "second," etc., used herein are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0036] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0037] First, the terminology used in the embodiments of this disclosure will be explained. It should be understood that this explanation is for the purpose of providing a clearer understanding of the embodiments of this disclosure and does not necessarily constitute a limitation thereof.

[0038] Chargeback, also known as order cancellation or refusal, refers to a credit card holder's (buyer's) request to the bank (the credit card issuer) to refuse payment for an order within a certain period after payment. Chargeback is essentially a right granted by the bank to credit card holders, and this risk cannot be completely avoided regardless of the e-commerce platform used by the merchant (seller). When a buyer pays for an online transaction on an e-commerce platform using a credit card, the buyer can request a chargeback from the bank according to the bank's rules and time limits. Disputes regarding chargebacks are ultimately adjudicated by the bank.

[0039] Chargeback defense: Banks require merchants to provide corresponding evidence and defenses against chargeback requests initiated by buyers within a specified time. The bank will review the defense materials provided by the seller and make a final ruling. The timeliness and professionalism of the prepared defense materials determine whether a chargeback defense will be successful.

[0040] Comprehensive Chargeback Service (or Chargeback Protection Service): To better protect the rights and interests of merchants and reduce the losses caused by chargeback risks, e-commerce platforms provide comprehensive chargeback services to merchants who have activated credit protection services. Under the condition of meeting the requirements of the comprehensive chargeback service and having sufficient chargeback service limit, regardless of the final outcome of the appeal, the e-commerce platform will bear the loss of chargeback refund amount within the merchant's comprehensive chargeback service limit.

[0041] Large Language Models (LLMs) are deep learning models trained on large amounts of text data that are capable of generating and understanding natural language text. These models typically consist of billions or even hundreds of billions of parameters and are trained through deep neural networks to handle complex natural language tasks such as text classification, question answering, and dialogue.

[0042] A prompt is a way to interact with an LLM model. It guides the model to produce the desired output through language instructions. Prompts can be questions, instructions, or any form of text input, designed to motivate the model to generate a specific type of response or complete a specific task.

[0043] A cue word template is a pre-designed cue word framework with a specific structure and format. It includes fixed parts and placeholders. The positions occupied by the placeholders in the template are the spaces to be filled. By filling the placeholders, different cue words can be generated. Cue word templates help improve the efficiency and standardization of cue word generation, making interaction with LLM more convenient and effective.

[0044] To improve the efficiency of filling out online forms (hereinafter referred to as forms), this disclosure provides an accompanying data filling scheme, that is, during the process of users filling out forms, a virtual assistant (or virtual robot) is used to provide relevant filling suggestions for the items to be filled in the form, thereby improving filling efficiency and reducing filling time.

[0045] The virtual assistant mentioned in this disclosure is a software based on artificial intelligence (AI) technology, which can recognize and understand the user's language input, perform information search and processing, and output appropriate responses.

[0046] In one alternative implementation, the virtual assistant described above can be implemented based on an LLM model.

[0047] Typically, LLM models involve a pre-training phase and a fine-tuning phase. The pre-training phase is the initial stage of the LLM model's learning. During pre-training, the model is exposed to a large amount of unlabeled text data, such as books, articles, and websites. The goal is to capture the underlying patterns, structures, and semantic knowledge existing in the text corpus. By utilizing a large amount of unlabeled or weakly labeled data, the model is trained to obtain a preliminary model with general knowledge or capabilities. Pre-training allows the model to capture a wide range of useful features by learning from a large amount of general data before performing a specific task, thereby improving the model's performance and generalization ability on that specific task.

[0048] Although pre-trained models have learned rich general features and prior knowledge on large-scale datasets, these features and knowledge may not be fully applicable to a specific target task. Fine-tuning, by further training the pre-trained model on a small amount of labeled data relevant to the specific task, enables the model to learn specific features and patterns related to the target task, thereby better adapting to the specific task.

[0049] Besides improving the performance of LLM models through pre-training and fine-tuning, techniques such as prompt engineering and retrieval-augmented generation (RAG) can also enhance their performance. Prompt engineering is a method specifically designed to optimize language models, specifically by optimizing prompt templates. Its goal is to guide these models to generate more accurate and targeted output text by designing and adjusting prompt templates. Retrieval-augmented generation combines retrieval and generation techniques. It generates answers or content by referencing information from external knowledge bases (i.e., plug-in knowledge bases), offering strong interpretability and customizability. It is suitable for various natural language processing tasks such as question-answering systems, document generation, and intelligent assistants. RAG-based models have the advantages of strong versatility, the ability to achieve real-time knowledge updates, and the provision of more efficient and accurate information services through end-to-end evaluation methods.

[0050] Optionally, the LLM model can be derived from an existing pre-trained model, such as an existing Generative Pre-trained Transformer (GPT) model. This allows skipping the pre-training stage and directly improving the pre-trained model's performance on a specific task through one or more methods such as prompt word engineering, retrieval enhancement generation, and fine-tuning.

[0051] Typically, forms involve multiple fields to be filled in, each requiring different content. Optionally, to improve the model's ability to provide fill-in recommendations, different prompt word templates can be configured for different fields. The prompt word template for each field can be optimized through prompt word engineering.

[0052] Taking a chargeback defense scenario as an example, training corpus can be generated based on historical order defense cases. The training corpus can include: the defense materials submitted in the historical defense cases and the order information of the historical orders. The defense materials are historically collected from forms, including the information entered for each field in the form, and the order information of the historical orders, including but not limited to: order details, buyer and seller information, logistics information and vouchers, payment information and vouchers, the buyer's reasons for chargeback, and communication records between the buyer and seller regarding the order. Logistics vouchers can include shipping vouchers and delivery confirmation vouchers. The training corpus generated based on successful historical defense cases can be used as positive corpus, and the training corpus generated based on unsuccessful historical defense cases can be used as negative corpus.

[0053] Based on both positive and negative corpora, the prompt word template for each field to be filled can be optimized / adjusted to obtain a prompt word template that meets preset requirements. This way, when providing filling recommendations to users later, the LLM can be used to obtain the filling recommendations for the fields to be filled based on the prompt word target. The prompt word template optimization process can be implemented based on existing prompt word projects, and this embodiment does not specifically limit this process.

[0054] In practical applications, after the user (i.e., the merchant) fills out the form and submits it to the e-commerce platform, the platform can use an LLM model to assess whether the collected defense information is sufficient for a successful defense. In other words, before the e-commerce platform initiates a chargeback defense request with the bank, it first estimates whether the user's submitted defense information will be successful. In one example, the e-commerce platform can determine whether to provide chargeback protection services to the user based on the estimation results. In another example, the e-commerce platform can determine whether to remind the user to modify the form to improve the success rate of the defense based on the estimation results.

[0055] To improve the model's performance in assessing whether a defense can be successfully defended, the prompt words, external knowledge base, and / or the model can be optimized based on positive and negative corpora.

[0056] In one embodiment, the prompt word template can be optimized / adjusted based on positive and negative corpora to obtain a prompt word template that meets preset requirements. Thus, when evaluating whether the rebuttal data collected from the form can be successfully submitted, the evaluation results of the LLM model can be obtained based on this prompt word template.

[0057] In another embodiment, an external knowledge base (i.e., an external knowledge base) is attached to the model. This knowledge base includes positive and negative corpora. The positive corpora record successful defenses, while the negative corpora record unsuccessful defenses. Thus, when evaluating the success of defense data collected from the form, the collected data can be matched with the positive and negative corpora in the knowledge base. If a match is found, the defense result corresponding to the matched corpus is used as the evaluation result; otherwise, the larger model generates the evaluation result based on its parameters.

[0058] In another embodiment, the model parameters can be fine-tuned based on positive and negative corpora to obtain a model that meets preset performance requirements. During model fine-tuning, the training corpus can be divided into a training set and a test set. The training set is used to fine-tune the model parameters, and the test set is used to test or evaluate the model's prediction accuracy. When the prediction accuracy is greater than or equal to a preset threshold, it indicates that the model's performance meets the preset requirements, and fine-tuning can stop. Existing model fine-tuning methods can be used, and this disclosure does not specifically limit the methods used.

[0059] In practical applications, one or more of the following methods can be selected to improve model performance: optimizing prompt word templates, adding external knowledge bases, and fine-tuning the model. This disclosure does not specifically limit the methods used. For example, an external knowledge base can be added to the large model first; then the prompt word template can be optimized to obtain an optimized prompt word template; finally, the model can be fine-tuned based on the optimized prompt word template. It should be noted that if optimizing the prompt word template is sufficient to make the model's performance in evaluating whether a successful defense can be achieved to meet the preset performance requirements, then fine-tuning of the model is unnecessary; if optimizing the prompt word template is insufficient to make the model's performance in evaluating whether a successful defense can be achieved to meet the preset performance requirements, then fine-tuning of the model can continue.

[0060] Furthermore, after the model is deployed, it can be fine-tuned based on newly generated rebuttal cases, thereby iterating to produce a better model. For example, new training corpus can be generated based on newly generated rebuttal cases. This new training corpus can be used as the model's test set. By comparing the model's evaluation results on the test set with the actual rebuttal results on the test set, the model's accuracy on that test set can be evaluated. If the accuracy is lower than a preset threshold, the model can be further fine-tuned based on the new training corpus and the training corpus previously used for fine-tuning the model, thereby iterating to produce a better model.

[0061] As shown in Figure 1, the model is continuously iterated using the newly generated training corpus. The performance of model V2 is higher than that of model V1, and the performance of model V3 is higher than that of model V2.

[0062] Of course, the prompt word template can also be optimized and the knowledge base of the plug-in can be updated based on the newly generated training corpus. For example, the newly generated training corpus can be added to the plug-in knowledge base.

[0063] Forms typically require a large number of fields to be filled, which is time-consuming. To improve form filling efficiency, this disclosure provides a form filling method applicable to terminals, as shown in Figure 2. The method includes the following steps:

[0064] 201. Display the target interface.

[0065] The target interface includes a form to be filled out.

[0066] For example, a form to be filled out is used to collect information related to a target object. The target object differs in different application scenarios. For instance, if the form to be filled out is a company's annual report, then the target object is the target company. Alternatively, if the form to be filled out is a defense information form used to collect defense information against a chargeback application, and the chargeback application is against a target order, then the target object refers to the target order.

[0067] The data required for a form to be filled out varies depending on the application scenario. Therefore, the fields for filling out the form can be designed according to different application scenarios.

[0068] Taking the scenario of defending against non-payment as an example, as shown in Figure 4, the form filling interface (also known as the target interface) 40 includes a form to be filled 41 and a dialog box 42 for communicating with the non-payment defense assistant (also known as the virtual assistant).

[0069] In one optional implementation, in response to a user's (i.e., the merchant's) click on a target order in the order list interface, the order details interface of the target order is displayed, including a refund management control. In response to a user's triggering of the refund management control, the user enters the refund management interface, which includes a chargeback control. In response to a user's triggering of the chargeback control, the user enters the chargeback details interface 30 (as shown in Figure 3). As shown in Figure 3, the chargeback details interface 30 may include a control 31 for filling in defense information. In response to a user's triggering of the control 31 for filling in defense information, the form filling interface 40 shown in Figure 4 is displayed. Optionally, as shown in Figure 3, the chargeback details interface 30 may also include: order information 32 and chargeback details 33 of the target order. The order information may include the order number, total order price, prepayment information, and final payment information. The chargeback details information may include: chargeback type, chargeback reason, chargeback amount, and chargeback time. Optionally, as shown in Figure 3, the chargeback details interface 30 may also include: operation records related to chargebacks 34.

[0070] For example, as shown in Figure 4, the form to be filled 41 is located in the left area of ​​the form filling interface 40, and the dialog box 42 is located in the right area of ​​the form filling interface 40.

[0071] It should be noted that the layout of the form to be filled and the dialog box on the target interface can be set according to actual needs, and this embodiment does not impose specific limitations on this. For example, the form to be filled may be located in the upper area of ​​the form filling interface, and the dialog box may be located in the lower area of ​​the form filling interface; another example is that the form to be filled may be located in the right area of ​​the form filling interface, and the dialog box 42 may be located in the left area of ​​the form filling interface 40.

[0072] As shown in Figure 4, the form 41 to be filled out may include multiple fields. Taking a chargeback defense scenario as an example, as shown in Figure 4, the form 41 to be filled out may include: fields for filling in the reasons for the defense, fields for filling in the return address, fields for filling in the tracking number and tracking URL, fields for filling in the delivery time, fields for filling in communication proof materials such as emails and instant messaging, fields for filling in the shipping certificate, fields for filling in the delivery certificate, and fields for filling in the closing statement.

[0073] In practical applications, some fields are used to fill in text information, while others are used to add files.

[0074] When a field to be filled in is used to enter text information, it may include an input box (as shown in Figure 4, 415). Optionally, when a field to be filled in has the attribute of entering multiple pieces of information, it may include multiple input boxes, or it may include an add control for adding additional input boxes. For example, the field to be filled in for entering the tracking number and tracking URL in Figure 4 has the attribute of entering multiple pieces of information, and it corresponds to the "Add Tracking Number" control 416 shown in Figure 4. In response to a triggering operation on the add control, an input box can be added to the field to be filled in on the target interface.

[0075] As shown in Figure 4, when a field can be filled out with the assistance of a virtual assistant, a corresponding assistance control (412 as shown in Figure 4) can be set for that field in the form. When there are multiple fields that can be filled out with the assistance of a virtual assistant, a corresponding assistance control can be set for each field in the form, and a correspondence between the assistance control and the field can be established, with different fields corresponding to different assistance controls. For example, the assistance control for each field is displayed in association with that field, where the association can be achieved by displaying them in close proximity or by connecting them through a UI element used for association. As shown in Figure 4, the field 411 for filling out the defense statement is displayed close to its corresponding assistance control 412.

[0076] Optionally, the fields to be filled in may have an attribute that allows for manual input from the user. That is, the terminal can respond to an input operation on a field, receive the information entered during the input operation, and use that information as the input information for the field. In this way, if the virtual assistant cannot provide input recommendations or provides incorrect recommendations, the user can manually enter the relevant content to ensure accuracy.

[0077] The form to be filled out includes a first field to be filled in and a first control. The first field to be filled in can be any field in the form, and the first control refers to the assist control corresponding to the first field to be filled in.

[0078] 202. In response to a trigger operation on the first control, display a first question posed to the virtual assistant on the target interface.

[0079] The first question is used to ask for recommendations on how to fill in the first field to be filled in.

[0080] Optionally, in response to a trigger operation on the first control, the field to be filled corresponding to the first control (i.e., the first field to be filled) is determined based on a preset correspondence between controls and fields to be filled. Based on the first field to be filled, the first question is then determined. For example, a corresponding question can be configured for each field to be filled in advance, and a correspondence can be established between them. Subsequently, based on this correspondence, the question corresponding to the first field to be filled can be determined as the first question. After determining the first question, the first question posed to the virtual assistant is displayed on the target interface.

[0081] For example, the triggering action may include a click action.

[0082] For example, as shown in Figure 4, in response to a triggering operation on the assist control 412 corresponding to the field 412 for filling in the defense, the question 421 "Recommend a suitable defense for this defense" is displayed in the dialog box 42 of the target interface.

[0083] In this embodiment of the disclosure, users only need to click on the control corresponding to the field to be filled in to input a question for the virtual assistant to ask for filling recommendations for that field. Users do not need to manually edit the question, which reduces the user's cost and improves the filling efficiency.

[0084] 203. Display the virtual assistant's first response to the first question on the target interface.

[0085] The first response may include first recommendation information for filling in the first field to be filled.

[0086] For example, as shown in Figure 4, the virtual assistant's response 422 to question 421 can be displayed in the conversation box 42 of the target interface.

[0087] Optionally, the first response may also include the basis or reason for the recommendation, as shown in Figure 4. Response 422 includes "Recommendation Reason: ***". In this way, users can judge the accuracy of the virtual assistant's recommendation based on the recommendation reason.

[0088] Subsequently, users can complete the first field to be filled in based on the first response. For example, users can copy the first recommendation information and paste it into the first field to be filled in, thereby completing the task of filling in the first field.

[0089] In the technical solution provided by the embodiments of this disclosure, during the process of a user filling out a form, the user can use a virtual assistant to provide relevant filling suggestions for the items to be filled in the form by clicking the corresponding controls, which helps to improve filling efficiency and save time.

[0090] Applying the technical solutions provided in this disclosure to scenarios involving defense against non-payment can also help improve the professionalism of the information provided, thereby increasing the success rate of the defense.

[0091] In one alternative implementation, the virtual assistant may include an LLM model.

[0092] It should be noted that, in some embodiments, the virtual assistant's LLM model runs on the server. Therefore, the terminal can interact with the server to obtain the LLM model's filling recommendations for the first field to be filled, and then generate the virtual assistant's first answer to the first question based on the filling recommendations. In other embodiments, when the terminal's hardware resources meet the requirements for running the LLM model, the virtual assistant's LLM model can be deployed on the terminal. In this way, the virtual assistant on the terminal uses the locally deployed LLM model to obtain filling recommendations for the first field to be filled, and generates the virtual assistant's first answer to the first question based on the filling recommendations.

[0093] In one embodiment, the first response may further include the basis or reason for the recommendation, wherein the basis and reason for the recommendation may be provided by the LLM model. By displaying the basis or reason for the recommendation, the user can further determine whether the model recommendation is accurate. If accurate, the user adopts the model's recommended content; if inaccurate, the user manually inputs it.

[0094] When the first response includes the recommendation basis corresponding to the first recommendation information, a fourth control is displayed on the target interface; in response to a trigger operation on the fourth control, a details interface is displayed, wherein the details interface is used to display detailed information about the recommendation basis.

[0095] As shown in Figure 4, the target interface 40 can display a fourth control 423. In response to a trigger operation on the fourth control 423, a details interface (not shown) is displayed, which displays detailed information about the recommendation criteria. This allows users to easily view the details of the recommendation criteria. For example, as shown in Figure 4, the recommendation criteria for "logistics tracking number and tracking URL" include logistics vouchers. In response to the user's click operation on the "Click to view" control 423, a details interface is displayed, which can show information such as logistics voucher images.

[0096] In an optional implementation, the above method may further include:

[0097] 204. Display the second control on the target interface.

[0098] For example, the first recommendation information and the second control are displayed in a preset area of ​​the target interface.

[0099] In the target interface, the preset area is associated with the area where the first field to be filled in is located. In this context, "associated area" refers to areas that are geographically close or linked through interface elements.

[0100] For example, as shown in Figure 4, in the target interface, near the field 411 (i.e., the first field to be filled in) for filling in the defense reasons, the virtual assistant's recommended defense reasons 413 (i.e., the first recommended information) and the confirmation field filling control (i.e., the second control) 414 are displayed.

[0101] In one embodiment, when the first field to be filled in is used to add a file, the first recommendation information includes a first recommended file, wherein the file can be a document or an image. When the first recommended file is a document file, the filename of the document file is displayed on the target interface; when the first recommended file is an image, a thumbnail of the image (thumbnail 432 as shown in Figure 4) and / or the image name (image name 433 as shown in Figure 4) can be displayed. There can be one or more first recommended files. When there are multiple first recommended files, there are also multiple second controls ("Application" control 434 as shown in Figure 4). The multiple second controls correspond one-to-one with the multiple first recommended files. Different first recommended files correspond to different second controls, and the corresponding first recommended files and second controls can be displayed together on the target interface.

[0102] Optionally, when the number of first recommended files exceeds a preset threshold (e.g., 2), some of the first recommended files and the expansion control 435 can be displayed on the target interface. In response to a trigger operation on the expansion control 435, the first recommended files that have not yet been displayed in the form area can be displayed.

[0103] 205. In response to a trigger operation on the second control, fill in the first recommendation information into the first field to be filled.

[0104] For example, the triggering action could be a click action.

[0105] When the first field to be filled includes the first input box (that is, the first field to be filled is used to fill in text information), in response to the trigger operation on the second control, the first recommendation information is input into the first input box.

[0106] Optionally, the first recommendation information can be displayed in the first input box.

[0107] Optionally, after the first recommendation information is entered into the first field to be filled, the first recommendation information and the second control can be made to disappear from the target interface, that is, the first recommendation information and the second control are not displayed on the target interface.

[0108] As shown in Figure 4, the fields for filling in the tracking number and tracking URL include an input box 417 and an add control 416 for adding additional input boxes. When the virtual assistant recommends multiple pieces of information for the field (e.g., logistics information one and logistics information two as shown in Figure 4), the recommended information 419 (e.g., logistics information one as shown in Figure 4) can be displayed on the target interface, along with the "Confirm Fill" control 418 corresponding to the recommended information 419. In response to the trigger operation of the "Confirm Fill" control 418, the recommended information 419 is filled into the input box 417. In response to the trigger operation of the add control 416, an additional input box (not shown) is added to the field on the target interface, and other recommended information (other recommended information refers to recommended information that has not yet been filled in, such as logistics information two as shown in Figure 4) and their corresponding "Confirm Fill" controls (not shown) are displayed on the target interface. In response to the trigger operation of the control, the other recommended information is filled into the additional input box.

[0109] For example, on the target interface, other recommended information and their corresponding "confirm entry" controls are displayed instead of recommended information 419 and its corresponding "confirm entry" control 418. After the replacement, the target interface no longer displays recommended information 419 and its corresponding "confirm entry" control 418, but instead displays other recommended information and their corresponding "confirm entry" controls.

[0110] When the first field to be filled is used to add a file, the first recommendation information includes a first recommended file. In response to the trigger operation for the second control, the first recommended file is used as the file to be added to the first field to be filled.

[0111] Optionally, on the target interface, the added files for the first field to be filled in are displayed. For example, as shown in Figure 4, a thumbnail 431 of the added image for the first field to be filled in is displayed, or the filename of the added file is displayed.

[0112] In this embodiment of the disclosure, after the virtual assistant provides a filling recommendation, the user can trigger the operation through a control to fill in the recommended information into the field to be filled, thereby further improving the filling efficiency and user experience.

[0113] In one embodiment, the virtual assistant includes a large language model running on a server. The method described above may further include the following steps:

[0114] 206. In response to the trigger operation on the first control, send a first target request to the server.

[0115] The first target request includes first instruction information and second instruction information. The first instruction information is used to indicate the first item to be filled in, and the second instruction information is used to indicate the target object. The first target request is used to request the server to provide the first recommendation information based on the target object using the large language model.

[0116] For example, the first instruction information may include the name of the first field to be filled in. For example, the name of the field used to fill in the defense is the defense.

[0117] For example, the second indication information may include the object identifier of the target object. Taking a target order as an example, the object identifier may be the order number.

[0118] For example, the first target request mentioned above is generated by the terminal in response to a trigger operation on the first control.

[0119] In one optional implementation, after receiving the first target request, the server performs the following steps:

[0120] S1. Based on the target object, use a large language model to determine the first recommended information to fill in the first field to be filled.

[0121] S2. Return the first recommendation information to the terminal.

[0122] In step S1 above, object information of the target object can be obtained, a first target prompt word can be generated based on the object information, and the first target prompt word can be input into the LLM model to obtain first recommendation information. The first target prompt word is used to prompt the large language model to provide first recommendation information for filling in the first field to be filled based on the object information.

[0123] Taking a chargeback defense scenario as an example, the target object is the target order. Therefore, the object information is the order information, which includes, but is not limited to: order details, buyer and seller information, logistics information, logistics receipts, payment information, payment receipts, the buyer's reason for the chargeback, and communication records between the buyer and seller regarding the order. The order details may include, but are not limited to: product type, product name, product quantity, unit price, and total price.

[0124] It should be noted that the required object information differs depending on the field to be filled in.

[0125] Take the scenario of a defense against payment as an example:

[0126] For the fields to be filled in for the defense, order information such as the reason for the chargeback, logistics information, payment information, and product type can be obtained.

[0127] For the fields to be filled in, such as the tracking number and the tracking URL, you can obtain logistics information.

[0128] Logistics information can be obtained for the fields to be filled in for the delivery time.

[0129] Regarding shipping images, communication records between buyers and sellers can be accessed, from which previously uploaded images can be retrieved. Shipping images refer to logistics photos related to the shipping process.

[0130] The following section details a method for constructing a first-target prompt word, which includes the following steps:

[0131] S11. Obtain a first preset prompt word template that matches the first field to be filled in.

[0132] Based on the correspondence between the preset fields to be filled and the preset prompt templates, a first preset prompt template that matches the first field to be filled can be obtained.

[0133] Different preset prompt templates correspond to different fields to be filled in. The process of determining / optimizing the preset prompt templates can be found in the relevant content of the above embodiments, and will not be repeated here.

[0134] S12. Fill the object information of the target object into the first preset prompt word template to obtain the first target prompt word.

[0135] Using the previous example:

[0136] For the fields to be filled in for the defense arguments, order information such as the reason for chargeback, logistics information, payment information, and product type can be entered into the corresponding preset prompt word templates to obtain prompt words. Based on these prompt words, LLM selects the target defense argument from multiple alternative defense arguments.

[0137] For fields to be filled in, such as tracking numbers and tracking URLs, logistics information can be entered into corresponding preset prompt templates to generate prompts. Based on these prompts, LLM first identifies the tracking number, then the logistics company, and finally determines the corresponding tracking URL for that logistics company according to the preset correspondence between logistics companies and tracking URLs.

[0138] For the field to be filled in for the delivery time, logistics information can be entered into the corresponding preset prompt word template to obtain prompt words. The LLM model identifies the logistics information based on these prompt words to obtain delivery information and the delivery time.

[0139] For shipping images, images from communication records can be entered into corresponding preset prompt word templates to generate prompt words. The LLM model can then identify the image based on these prompt words to determine whether it is a shipping image.

[0140] In step S2 above, the first recommendation information can be sent to the terminal via the network. Optionally, the reasons or basis for the LLM recommendation of the first recommendation information can also be sent to the terminal.

[0141] 207. Receive the first recommendation information returned by the server.

[0142] For example, the terminal receives the first recommendation information and the reasons or basis for the recommendation returned by the server through the network.

[0143] 208. Based on the first recommendation information, determine the first response.

[0144] For example, the first recommendation information and the reasons or basis for the recommendation can be used as the first response.

[0145] In one embodiment, the form requires the user to input a summary text (hereinafter referred to as the summary text) to summarize the information collected in the form. Taking a defense against non-payment as an example, the form requires the user to input a concluding statement for the defense; the professionalism of the concluding statement has a significant impact on the defense outcome. Generally, editing a concluding statement is not only difficult but also very time-consuming. To improve the efficiency of inputting the summary text, the form also includes a second field and a third control, whereby the second field is used to input the summary text.

[0146] The summary text refers to a summary and overview of the data collected from the completed form. The above methods may also include:

[0147] 209. In response to a trigger operation on the third control, send a second target request to the server and display a second question to the virtual assistant on the target interface.

[0148] The second target request includes third instruction information, which is used to instruct the second field to be filled in. The second target request is used to request the server to provide second recommendation information based on the already filled information of the first field to be filled in, using the large language model. The second recommendation information is used to fill in the second field to be filled in.

[0149] For example, the third instruction information may include the name of the second field to be filled in. For example, the field for filling in the closing statement may be named "Closing Statement".

[0150] For example, the second target request mentioned above is generated by the terminal in response to a triggering operation on a third control.

[0151] In one optional implementation, after receiving the second target request, the server performs the following steps:

[0152] S3. Based on the information already filled in the first field to be filled, use a large language model to determine the second recommended information to fill in the second field to be filled.

[0153] S4. Return the second recommendation information to the terminal.

[0154] In S3 above, the server can obtain the filled-in information of other fields in the form (including the first field) besides the second field; based on the filled-in information of the other fields, it generates a second target prompt word and inputs the second target prompt word into the LLM model to obtain second recommendation information. The second target prompt word is used to prompt the large language model to provide second recommendation information for filling in the second field based on the filled-in information.

[0155] The following section details a method for constructing a second target cue word, which includes the following steps:

[0156] S11. Obtain a second preset prompt word template that matches the second field to be filled in.

[0157] In one embodiment, a second preset prompt template that matches the second item to be filled can be obtained based on the correspondence between preset items to be filled and preset prompt templates.

[0158] In one embodiment, the second preset prompt template may include a summary text example. This allows the model to refer to the summary text example when generating the summary text, helping to improve the accuracy and professionalism of the generated text.

[0159] In chargeback defense scenarios, the wording of the summary text may differ depending on the reason for chargeback. Therefore, different preset prompt templates can be configured in advance for different reasons for chargeback, and a correspondence between chargeback reasons and preset prompt templates can be established. The preset prompt template for each chargeback reason includes a summary text example that matches that reason. In this way, a preset prompt template that matches the target chargeback reason for the target order can be obtained later and used as a second preset prompt template.

[0160] Of course, in practical applications, the wording of the summary text may differ between different banks. Therefore, for the same chargeback reason, different preset prompt templates can be configured for different banks, establishing a correspondence between banks, chargeback reasons, and preset prompt templates. For the same chargeback reason, the summary text examples in the preset prompt templates for different banks will be different. Subsequently, based on the target bank involved in the target order (i.e., the bank used by the buyer for payment) and the target chargeback reason for the target order, a preset prompt template adapted to that target bank and target chargeback reason can be obtained as a second preset prompt template. This further improves the accuracy and professionalism of the generated summary text.

[0161] S12. Fill in the information already filled in the other fields to be filled in the form except for the second field to be filled into the second preset prompt word template to obtain the second target prompt word.

[0162] 210. Receive the second recommendation information returned by the server.

[0163] For example, the terminal receives the second recommendation information returned by the server via the network.

[0164] 211. Display the virtual assistant's second response to the second question on the target interface.

[0165] The second response includes the second recommendation information.

[0166] Alternatively, the above method may also include the following steps:

[0167] 212. Enter the second recommendation information into the input box of the second field to be filled in and display the second recommendation information in the input box.

[0168] As shown in Figure 4, the second field to be filled in 436 includes an input box 438, in which the summary text generated by the LLM model can be displayed.

[0169] Alternatively, the above method may also include the following steps:

[0170] 213. Receive user modification requests for the second recommended information in the input box.

[0171] It can receive user modifications such as word choice and sentence structure adjustments for the second recommended information in the input box.

[0172] 214. Based on the modification operation, modify the second recommended information in the input box.

[0173] In this embodiment, receiving manual modifications from the user to the summary text generated by the LLM model can further improve the professionalism of the summary text.

[0174] In this embodiment of the disclosure, the summary text generated by the model (i.e., the second recommendation information) belongs to AI Generated Content (AIGC).

[0175] Optionally, the aforementioned virtual assistant can also provide merchants with consulting services, such as: consultation on chargeback notices, consultation on filling out defense information, consultation on modifying defense information, consultation on reasons for non-coverage, and consultation on reasons for losing a case.

[0176] Figure 5 shows a flowchart of a form filling method provided in another embodiment of this disclosure. The execution entity of this method is the server. As shown in Figure 5, the method includes:

[0177] 501. Receive the first target request sent by the receiving terminal.

[0178] The first target request includes a first instruction information and a second prompt information. The first instruction information is used to indicate the first field to be filled in of the form to be filled in, and the second instruction information is used to indicate the target object. The form to be filled in is used to collect information related to the target object.

[0179] 502. Based on the target object, use a large language model to determine the first recommended information to fill in the first field to be filled.

[0180] 503. Return the first recommendation information to the terminal.

[0181] The specific implementation of steps 501 to 503 above can be found in the corresponding contents of the above embodiments, and will not be repeated here.

[0182] This disclosure provides a data entry scheme that uses an LLM model to provide relevant entry recommendations for fields to be filled in a form, thereby improving entry efficiency and reducing entry time.

[0183] In an alternative implementation, step 502 above, "determining the first recommended information for filling in the first field to be filled using a large language model," can be achieved by the following steps:

[0184] 5021. Obtain the object information of the target object.

[0185] 5022. Obtain the first preset prompt word template that matches the first field to be filled in.

[0186] 5023. Fill the object information of the target object into the first preset prompt word template to obtain the first target prompt word.

[0187] Wherein, the first target prompt word is used to prompt the large language model to provide first recommended information for filling in the first item to be filled, based on the object information.

[0188] 5024. Input the first target prompt word into the large language model to obtain the first recommendation information.

[0189] The specific implementation of steps 5021 to 5024 above can be found in the corresponding contents of the above embodiments, and will not be repeated here.

[0190] In one optional implementation, the above method may further include:

[0191] 504. Receive the second target request sent by the terminal.

[0192] The second target request includes a third instruction, which is used to instruct the second field to be filled in the form, and the second field to be filled in is used to fill in summary text.

[0193] 505. Obtain the second preset prompt word template that matches the second field to be filled.

[0194] 506. Fill in the already filled information of the first field to be filled into the second preset prompt word template to obtain the second target prompt word.

[0195] The second target prompt word is used to prompt the large language model to provide second recommended information for filling in the second field based on the information already filled in the first field to be filled.

[0196] 507. Input the second target prompt word into the large language model to obtain the second recommendation information.

[0197] 508. Return the second recommendation information to the terminal.

[0198] The specific implementation of steps 504 to 508 can be found in the corresponding contents of the above embodiments, and will not be repeated here.

[0199] In one alternative implementation, the second preset prompt template includes a summary text example.

[0200] In one alternative implementation, the target object includes a target order, the form to be filled out is used to collect defense information against a chargeback request, the chargeback request being made against the target order, and the defense information being related to the target order.

[0201] In one optional implementation, the above method further includes:

[0202] 509. Obtain the defense information collected from the form to be filled in and the third preset prompt word template.

[0203] The third preset prompt word template can be obtained through prompt word engineering optimization.

[0204] For example, the third preset prompt word template may include preset judgment criteria. The preset judgment criteria are used to determine whether the defense is successful or unsuccessful. The judgment criteria can be set according to actual needs, and this embodiment of the disclosure does not specifically limit them.

[0205] 510. Fill the defense information into the third preset prompt word template to obtain the third target prompt word.

[0206] The third preset prompt word template is used to prompt the large language model to predict the defense result of the defense materials.

[0207] For example, the third preset prompt word template is used to prompt the large language model to estimate the defense result of the defense materials according to preset judgment criteria.

[0208] 511. Input the third target prompt word into the large language model to predict the defense result of the defense data.

[0209] In one embodiment, a large language model can generate a defense result of the defense material based on a third target prompt word and using model parameters.

[0210] In another embodiment, the large language model includes an external knowledge base. The model can first match the rebuttal data in the third target prompt with the positive and negative corpora in the external knowledge base. If a positive corpus is matched, the rebuttal result of the rebuttal data is estimated to be a successful rebuttal; if a negative corpus is matched, the rebuttal result of the rebuttal data is estimated to be a failed rebuttal; if neither a positive nor a negative corpus is matched, the rebuttal result of the rebuttal data is generated using model parameters.

[0211] In one embodiment, if the defense results in a successful defense, it is determined that the target order will be provided with chargeback protection. If the defense results in a failed defense, it is determined that the target order will not be provided with chargeback protection.

[0212] Optionally, in the case of a chargeback defense, a defense document (such as a Word document or a PDF document) can be generated based on the defense information collected from the form. The defense document can then be submitted to the bank for a defense application.

[0213] In this embodiment, data is collected through an online form. The collected data is structured, which helps to generate a Word or PDF file for defense.

[0214] In practical applications, the accompanying defense assistant can provide merchants with guidance and suggestions on filling out forms, reducing the time merchants spend filling out forms from hours to minutes, effectively improving the user experience.

[0215] It should be noted that any steps in the method provided in this disclosure that are not fully described in detail can be found in the corresponding content of the above embodiments, and will not be repeated here. Furthermore, the method provided in this disclosure may include other parts or all of the steps in the above embodiments in addition to the steps described above; for details, please refer to the corresponding content of the above embodiments, and will not be repeated here.

[0216] Figure 6 shows a schematic diagram of a form filling system provided according to an embodiment of the present disclosure. The form filling system includes: a terminal 601 and a server 602; wherein,

[0217] The terminal 601 is configured to: display a target interface, the target interface including a form to be filled out, the form to be filled out for collecting information related to a target object, the form to be filled out including a first field to be filled out and a first control; in response to a trigger operation on the first control, send a first target request to the server and display a first question to be asked of the virtual assistant on the target interface, the first target request including first instruction information and second instruction information, the first instruction information being used to indicate the first field to be filled out, the second instruction information being used to indicate the target object, and the first target request being used to request the server to provide the first recommendation information based on the target object using the large language model;

[0218] The server 602 is configured to: receive a first target request sent by the terminal; determine the first recommendation information based on the target object using a large language model; and return the first recommendation information to the terminal.

[0219] The terminal 601 is further configured to: receive the first recommendation information returned by the server; and display the first answer from the virtual assistant to the first question on the target interface, wherein the first answer includes the first recommendation information.

[0220] This disclosure provides an accompanying data entry scheme, in which a virtual assistant provides relevant entry suggestions for the fields to be filled in the form during the user's form entry process, thereby improving entry efficiency and reducing entry time.

[0221] It should be noted that the specific implementation of the terminal, server, and interaction process between them provided in this embodiment can be found in the corresponding content of the above embodiments, and will not be repeated here.

[0222] Terminal: This can be a mobile terminal, fixed terminal, or portable terminal, such as mobile phones, sites, units, devices, multimedia computers, multimedia tablets, internet nodes, communicators, desktop computers, laptop computers, notebook computers, netbook computers, tablet computers, personal communication system devices, personal navigation devices, personal digital assistants, audio / video players, digital cameras / camcorders, positioning devices, television receivers, radio broadcast receivers, e-book devices, gaming devices, or any combination thereof, including accessories, peripherals, or any combination thereof. It is also foreseeable that terminal devices can support any type of user-facing interface device (e.g., wearable devices).

[0223] Server: A server can be one or more servers. A server can also be a physical server or a virtual server, etc. A server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, and big data and artificial intelligence platforms.

[0224] As shown in Figure 7, the form filling solution provided in this embodiment leverages database query capabilities, LLM model generation, dialogue, multimodal understanding, and intelligent agent technologies to offer rich technical expression capabilities. These capabilities include: structured form filling and recommendation, argument generation, closing statement generation, and voucher screenshot generation and understanding. These technical expression capabilities are used to generate or obtain filling recommendation information, such as payment vouchers, logistics vouchers, arguments, and closing statements.

[0225] Figure 8 shows a schematic diagram of the structure of an electronic device provided according to an embodiment of the present disclosure. As shown in Figure 8, the electronic device includes a memory 1101 and a processor 1102. The memory 1101 can be configured to store various other data to support operation on the electronic device. Examples of such data include instructions for any application or method operating on the electronic device. The memory 1101 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Electrically Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0226] The memory 1101 is used to store programs;

[0227] The processor 1102 is coupled to the memory 1101 and is used to execute the program stored in the memory 1101 to implement the methods provided in the above-described method embodiments.

[0228] Furthermore, as shown in Figure 8, the electronic device also includes other components such as a communication component 1103, a display 1104, a power supply component 1105, and an audio component 1106. Figure 8 only schematically shows some of the components and does not imply that the electronic device includes only the components shown in Figure 8.

[0229] Accordingly, this disclosure also provides a computer-readable storage medium storing a computer program, which, when executed by a computer, can implement the steps or functions of the methods provided in the above-described method embodiments.

[0230] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps or functions of the methods provided in the above-described method embodiments.

[0231] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0232] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM (Read Only Memory) / RAM (Random Access Memory), magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0233] Having described several aspects and embodiments of the technology of this disclosure, it should be understood that various changes, modifications, and improvements will readily occur to those skilled in the art. These changes, modifications, and improvements are intended to fall within the spirit and scope of the technology described herein. Therefore, it should be understood that the foregoing embodiments are presented by way of example only, and embodiments of this disclosure may be practiced in ways different from those specifically described within the scope of the appended claims and their equivalents. Furthermore, any combination of two or more features, systems, articles of manufacture, materials, and / or methods described herein that do not contradict each other is included within the scope of this disclosure. Moreover, as mentioned above, some aspects may be embodied as one or more methods. Actions performed as part of a method may be ordered in any suitable manner. Thus, embodiments in which actions are performed in an order different from the order shown can be constructed, which may include performing some actions simultaneously, even if shown as sequential actions in the illustrative embodiments.

Claims

1. A form filling method, applicable to a terminal, wherein, include: Display a target interface, the target interface including a form to be filled out, the form including a first field to be filled out and a first control; In response to a triggering operation on the first control, a first question posed to the virtual assistant is displayed on the target interface; The target interface displays the virtual assistant's first response to the first question, which includes first recommended information for filling in the first field to be filled.

2. The method according to claim 1, wherein, Also includes: The second control is displayed on the target interface; In response to a trigger operation on the second control, the first recommendation information is entered into the first field to be filled.

3. The method according to claim 2, wherein, The second control is displayed on the target interface, including: The first recommendation information and the second control are displayed in a preset area of ​​the target interface; In the target interface, the preset area is associated with the area where the first field to be filled is located.

4. The method according to any one of claims 1 to 3, wherein, The form to be filled out is used to collect information related to the target object.

5. The method according to claims 1 to 4, wherein, The virtual assistant is implemented based on a large language model running on the server; the method further includes: In response to a trigger operation on the first control, a first target request is sent to the server. The first target request includes first indication information and second indication information. The first indication information is used to indicate the first field to be filled in, and the second indication information is used to indicate the target object. The first target request is used to request the server to provide the first recommendation information based on the target object using the large language model. Receive the first recommendation information returned by the server; Based on the first recommendation information, the first response is determined.

6. The method according to claims 1 to 5, wherein, The form to be filled out includes a second field to be filled out and a third control. The second field to be filled out is used to fill in summary text. In response to a trigger operation on the third control, a second target request is sent to the server and a second question to be asked of the virtual assistant is displayed on the target interface. The second target request includes third instruction information, which is used to indicate the second field to be filled in. The second target request is used to request the server to provide second recommendation information based on the already filled information of the first field to be filled in using the large language model. The second recommendation information is used to fill in the second field to be filled in. Receive the second recommendation information returned by the server; The target interface displays the virtual assistant's second response to the second question, which includes the second recommendation information.

7. The method according to claim 4 or 5, wherein, The target object includes the target order, and the form to be filled out is used to collect defense information against the chargeback application, which is made in response to the target order.

8. The method according to any one of claims 1 to 7, wherein, The first response also includes the basis or reason for the recommendation corresponding to the first recommendation information.

9. The method according to claim 8, wherein, Also includes: When the first response includes the recommendation basis corresponding to the first recommendation information, a fourth control is displayed on the target interface; In response to a trigger operation on the fourth control, a details interface is displayed, wherein the details interface is used to display detailed information about the recommendation criteria.

10. The method according to claim 6, characterized in that, The method further includes: When the second recommendation information successfully matches the positive corpus in the external knowledge base, the recommendation confidence level is marked as high in the second response. The external knowledge base includes positive corpus of historical successful defense cases and negative corpus of historical failed defense cases. When the second recommendation information successfully matches the negative corpus in the external knowledge base, the recommendation confidence level is marked as low in the second response.

11. A form filling method, applicable to the server side, wherein, include: The receiving terminal sends a first target request, which includes first indication information and second prompt information. The first indication information is used to indicate the first item to be filled in of the form to be filled in, and the second indication information is used to indicate the target object. The form to be filled in is used to collect information related to the target object. Based on the target object, a large language model is used to determine the first recommended information to fill in the first field to be filled; The first recommendation information is returned to the terminal.

12. The method according to claim 11, wherein, Based on the target object, a large language model is used to determine the first recommended information for filling in the first field to be filled, including: Obtain the object information of the target object; Obtain a first preset prompt word template that matches the first field to be filled in; The object information of the target object is filled into the first preset prompt word template to obtain the first target prompt word. The first target prompt word is used to prompt the large language model to provide first recommended information for filling in the first item to be filled in based on the object information. The first target prompt word is input into the large language model to obtain the first recommendation information.

13. The method according to claim 11 or 12, wherein, Also includes: The terminal sends a second target request, which includes third indication information. The third indication information is used to indicate a second field to be filled in the form to be filled in. The second field to be filled in is used to fill in summary text. Obtain a second preset prompt word template that matches the second field to be filled in; The information already filled in the first field to be filled is entered into the second preset prompt word template to obtain the second target prompt word. The second target prompt word is used to prompt the large language model to provide second recommendation information for filling in the second field to be filled based on the information already filled in the first field to be filled. The second target prompt word is input into the large language model to obtain the second recommendation information; The second recommendation information is returned to the terminal.

14. The method according to claim 13, wherein, The second preset prompt template includes a summary text example.

15. The method according to any one of claims 11 to 14, wherein, The target object includes the target order, and the form to be filled in is used to collect defense information against the chargeback application. The chargeback application is made against the target order, and the defense information is related to the target order.

16. The method according to claim 11 or 15, wherein, Also includes: Obtain the defense information collected from the form to be filled in and the third preset prompt word template; The defense information is filled into the third preset prompt word template to obtain the third target prompt word. The third preset prompt word template is used to prompt the large language model to predict the defense result of the defense information. The third target prompt is input into the large language model to predict the defense outcome of the defense materials.

17. A form filling system, wherein, include: Terminal and server; among them, The terminal is configured to: display a target interface, the target interface including a form to be filled out, the form to be filled out being used to collect information related to a target object, the form to be filled out including a first field to be filled out and a first control; in response to a trigger operation on the first control, send a first target request to the server and display a first question to be asked of the virtual assistant on the target interface, the first target request including first instruction information and second instruction information, the first instruction information being used to indicate the first field to be filled out, the second instruction information being used to indicate the target object, and the first target request being used to request the server to provide the first recommendation information based on the target object using the large language model; The server is configured to: receive a first target request sent by the terminal; determine the first recommendation information based on the target object using a large language model; and return the first recommendation information to the terminal. The terminal is further configured to: receive the first recommendation information returned by the server; and display the first response from the virtual assistant to the first question on the target interface, wherein the first response includes the first recommendation information.

18. An electronic device, wherein, include: Memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is configured to execute the program stored in the memory to implement the method of any one of claims 1 to 10, or the method of any one of claims 11 to 16.

19. A computer-readable storage medium storing a computer program, wherein, When the computer program is executed by a computer, it can implement the method of any one of claims 1 to 10, or the method of any one of claims 11 to 16.

20. A computer program product comprising a computer program, wherein, When executed by a processor, the computer program implements the method of any one of claims 1 to 10, or the method of any one of claims 11 to 16.

Images