Business question and answer method and device, equipment, storage medium and program product

Abstract

The invention provides a business question and answer method which can be applied to the technical field of artificial intelligence. The business question-answering method comprises the following steps: acquiring an initial question of a user, and determining a target question which needs to be input into a business question-answering model; inputting the target question into the business question and answer model to obtain a corresponding target answer result; displaying the target answer result; wherein the business question and answer model is obtained by training the target large language model according to a training set in a preset business knowledge base; the target large language model is obtained by performing architecture adjustment on the initial large language model based on a fine adjustment technology. The invention further provides a business question answering device and equipment, a storage medium and a program product.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically to a business question-answering method, apparatus, device, storage medium, and program product. Background Art

[0002] A question-answering system (Q&A) is an AI system that uses natural language processing (NLP) to answer user questions. Its fundamental concept is to use natural language understanding and generation techniques to understand and answer questions in natural language text. It typically consists of two main components: a question and an answer. A question is input into the system, which then uses NLP to understand the question and find one or more answers, representing the answers in natural language and returning them to the user. In recent years, large language models have become increasingly mainstream. These models possess extensive knowledge and powerful language understanding capabilities, enabling them to handle the complex semantics and context of natural language. As a result, Q&A systems have experienced significant development in the era of large models, becoming a key application in the field of AI.

[0003] However, while the use of large language models in question-answering systems has significant advantages, it also has obvious disadvantages: it requires a large amount of computing resources, including computing power and storage resources; large language models require a large amount of data for training. If this data has privacy and security issues or the data quality itself is not high, that is, the question-answering system itself has the defect of incorrect answers, it may have a negative impact on the model; large language models require a lot of training time, and the training process requires a lot of time and energy for effective scheduling and optimization. Summary of the Invention

[0004] In view of the above problems, the present application provides a business question-answering method, apparatus, device, storage medium and program product.

[0005] According to the first aspect of the present application, a business question-answering method is provided, comprising: obtaining the user's initial question and determining a target question to be input into a business question-answering model; inputting the target question into the business question-answering model to obtain a corresponding target answer result; and displaying the target answer result; wherein, the business question-answering model is obtained by training a target large language model based on a training set in a preset business knowledge base; and the target large language model is obtained by structurally adjusting the initial large language model based on fine-tuning technology.

[0006] According to an embodiment of the present application, the user's initial question is obtained, and the target question to be input into the business question and answer model is determined, including: under the condition that it is determined that there are historical conversation records, by calculating the similarity between the user's initial question and each historical question, the questions in all historical questions that are higher than a preset similarity threshold are determined as supplementary questions in the target question, and the target question includes the supplementary question and the initial question; under the condition that it is determined that there are no historical conversation records, the user's initial question is determined as the target question to be input into the business question and answer model.

[0007] According to an embodiment of the present application, by calculating the similarity between the user's initial question and each historical question, questions among all historical questions that are higher than a preset similarity threshold are determined as supplementary questions in the target question, including: matrix representation of the initial question and historical questions based on a pre-trained word vector model; using cosine similarity to calculate the similarity between the vector representation of the initial question and the vector representation of all historical questions to obtain a similarity score between the initial question and each historical question; randomly selecting any of the top three historical questions with the similarity score as a supplementary question; splicing the supplementary question and the initial question to obtain the target question.

[0008] According to an embodiment of the present application, the process of adjusting the architecture of the initial large language model based on fine-tuning technology includes: after the initial large language model is determined, adding a bypass structure to each stacked block in the core architecture of the initial large language model, and adding a regularization layer after all fully connected layers in the initial large language model.

[0009] According to an embodiment of the present application, a business question-answering model is obtained by training a target large language model based on a training set in a preset business knowledge base. The training process includes: using grid search to optimize the ignore ratio of the regularization layer to obtain the target large language model; based on a pre-trained word vector model, performing a matrix representation on each data item in the training set, where the data item is a business-related question-answer pair; inputting the matrix representation of each data item into the target large language model to complete the training and obtain a business question-answering model.

[0010] According to an embodiment of the present application, the method also includes: after the matrix representation of each data item is input into the target large language model, a matrix representation of each predicted answer is obtained; and the target large language model is adjusted according to the loss function between the matrix representation of each predicted answer and the true answer result in the question-answer pair.

[0011] According to an embodiment of the present application, displaying the target answer result includes: converting the target answer result into a corresponding target display form according to the target display mode; obtaining the user's feedback result and adjusting the target display form in real time.

[0012] According to an embodiment of the present application, the method further includes: obtaining the user's authorization to input the user's initial question; and obtaining the user's initial question after obtaining the user's authorization to input the initial question.

[0013] The second aspect of the present application provides a business question-and-answer device, which includes: an initial question acquisition module, used to obtain the user's initial question and determine the target question to be input into the business question-and-answer model; a result acquisition module, used to input the target question into the business question-and-answer model to obtain the corresponding target answer result; a result display module, used to display the target answer result; wherein, the business question-and-answer model is obtained by training a target large language model according to a training set in a preset business knowledge base; and the target large language model is obtained by structurally adjusting the initial large language model based on fine-tuning technology.

[0014] The third aspect of the present application provides an electronic device, comprising: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0015] The fourth aspect of the present application further provides a computer-readable storage medium having a computer program or instructions stored thereon, which implements the steps of the above method when the computer program or instructions are executed by a processor.

[0016] The fifth aspect of the present application further provides a computer program product, comprising a computer program or instructions, which implement the steps of the above method when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above contents and other objects, features and advantages of the present application will become more apparent through the following description of the embodiments of the present application with reference to the accompanying drawings, in which:

[0018] Figure 1 The following schematically illustrates an application scenario of the business question-answering method according to an embodiment of the present application;

[0019] Figure 2 A flowchart of a business question-and-answer method according to an embodiment of the present application is schematically shown;

[0020] Figure 3 Schematically shows a flow chart for determining a target problem according to an embodiment of the present application;

[0021] Figure 4 Schematically shows a flow chart for determining supplementary questions according to an embodiment of the present application;

[0022] Figure 5 A flowchart of training a target large language model according to an embodiment of the present application is schematically shown;

[0023] Figure 6 A schematic diagram of the structure of a business question-answering device according to an embodiment of the present application is shown; and

[0024] Figure 7 A block diagram of an electronic device suitable for implementing a business question-and-answer method according to an embodiment of the present application is schematically shown. DETAILED DESCRIPTION

[0025] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present application. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present application. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present application.

[0026] The terms used herein are only for describing specific embodiments and are not intended to limit the present application. The terms "comprise," "include," etc. used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0027] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0028] When expressions such as "at least one of A, B, and C, etc." are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0029] In the technical solution of this application, the user information involved (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0030] In the scenario of using personal information for automated decision-making, the methods, devices, and systems provided in the embodiments of the present application all provide users with corresponding operation portals for users to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge and skills, and have reached a certain level of professionalism.

[0031] An embodiment of the present application provides a business question-answering method, including: obtaining the user's initial question and determining the target question to be input into the business question-answering model; inputting the target question into the business question-answering model to obtain the corresponding target answer result; displaying the target answer result; wherein, the business question-answering model is obtained by training a target large language model based on a training set in a preset business knowledge base; and the target large language model is obtained by structurally adjusting the initial large language model based on fine-tuning technology.

[0032] Through the embodiments of this application, the method can automatically parse the initial question posed by the user and, when necessary, introduce relevant questions from historical conversation records. This not only reduces the model's computational cost and improves the efficiency of responses, but also enhances the accuracy and relevance of answers. Furthermore, by fine-tuning the architecture of the initial large language model, not only can the negative impact of limited data on the model be effectively reduced, but it can also further enhance the model's understanding of specific business domain knowledge, providing users with more accurate answers.

[0033] Figure 1 The application scenario diagram of the business question-and-answer method according to an embodiment of the present application is schematically shown.

[0034] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, and a server 105. A network 104 is used as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or optical fiber cables.

[0035] A user may use a first terminal device 101, a second terminal device 102, or a third terminal device 103 to interact with a server 105 via a network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, or the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (for example only).

[0036] The first terminal device 101 , the second terminal device 102 , and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0037] The server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process received data such as user requests, and feed back processing results (e.g., web pages, information, or data obtained or generated based on user requests) to the terminal devices.

[0038] It should be noted that the business question and answer method provided in the embodiment of the present application can generally be executed by the server 105. Accordingly, the business question and answer device provided in the embodiment of the present application can generally be set in the server 105. The business question and answer method provided in the embodiment of the present application can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the business question and answer device provided in the embodiment of the present application can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0039] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0040] The following will be based on Figure 1 The scene described by Figures 2 to 5 The business question-and-answer method according to an embodiment of the present application is described in detail.

[0041] Figure 2 The flowchart of the business question-and-answer method according to an embodiment of the present application is schematically shown.

[0042] like Figure 2 As shown, according to an embodiment of the present application, the business question-and-answer method specifically includes operations S210 to S230.

[0043] In operation S210 , the user's initial question is obtained, and a target question to be input into the business question-answering model is determined.

[0044] In an embodiment of the present application, before obtaining the user's initial question, the user's authorization to input the user's initial question is obtained; after obtaining the user's authorization to input the initial question, the user's initial question is obtained.

[0045] Specifically, before obtaining the user's initial question, the user's authorization to input the user's initial question may be obtained. For example, before operation S210, a request to obtain the user's initial question may be issued to the user. If the user agrees or authorizes obtaining the user's initial question, operation S210 is performed.

[0046] For example, when a user asks an initial question on an online platform app on a terminal device, the backend determines whether there is a historical conversation record for the user's question. Based on the existence of the historical conversation record, the target question that needs to be input into the business question-answering model corresponding to the initial question is determined, and the target answer result is then obtained and returned to the user. The following is a detailed description of an embodiment of determining the target question.

[0047] Figure 3 The flowchart of determining the target problem according to an embodiment of the present application is schematically shown.

[0048] like Figure 3 As shown, according to an embodiment of the present application, the user's initial question is obtained, and the target question to be input into the business question-answering model is determined. This process specifically includes operations S310 to S320.

[0049] In operation S310, under the condition that it is determined that there are historical conversation records, by calculating the similarity between the user's initial question and each historical question, questions with a similarity higher than a preset threshold are determined among all historical questions as supplementary questions in the target question, and the target question includes the supplementary question and the initial question.

[0050] For example, if a user enters an initial question, such as "I want to know how to change my credit card payment date," into the customer service Q&A box on a bank app, the backend system will check whether there are any historical conversation records. Based on different scenarios, the backend system will determine the target question to enter into the business Q&A model. If there are historical conversation records, such as "What is the credit card bill date?" or "How can I quickly change my bill date?", the system will search for similar historical questions in the historical conversation records, identify supplementary questions, and finally combine them with the initial question as the target question. The determination of supplementary questions is explained in detail below.

[0051] Figure 4 The flowchart of determining supplementary questions according to an embodiment of the present application is schematically shown.

[0052] like Figure 4 As shown, according to an embodiment of the present application, by calculating the similarity between the user's initial question and each historical question, questions in all historical questions that are higher than a preset similarity threshold are determined as supplementary questions in the target question. This process specifically includes operations S410 to S440.

[0053] In operation S410 , the initial question and the historical questions are represented by a matrix based on the pre-trained word vector model.

[0054] According to an embodiment of the present application, a pre-trained word embedding model is first used to convert the initial question ("I want to know how to change my credit card payment date") and the historical questions ("What is the credit card bill date?", "How can I quickly change my bill date?") into a matrix representation. The vocabulary of each question is then mapped into a high-dimensional space to form a corresponding vector.

[0055] In operation S420 , the similarity between the vector representation of the initial question and the vector representations of all historical questions is calculated using cosine similarity to obtain a similarity score between the initial question and each historical question.

[0056] According to an embodiment of the present application, cosine similarity is used to calculate the similarity between the vector representations of the initial question and all previous questions. This calculation yields similarity scores. For example, the similarity score for question 1 ("What is the credit card bill date?") is 0.88, while the similarity score for question 2 ("How can I quickly change the bill date?") is 0.90.

[0057] In operation S430 , a historical question ranked among the top three in similarity scores is randomly selected as a supplementary question.

[0058] In operation S440 , the supplementary question and the initial question are concatenated to obtain a target question.

[0059] Furthermore, based on the similarity score, the system randomly selects one of the top three historical questions as a supplementary question. For example, here the system selects historical question 2 ("How can I quickly change the bill date?") as the supplementary question. Finally, the selected supplementary question is concatenated with the initial question to generate the target question. This enriches the target question and provides more comprehensive information, helping the subsequent business question-answering model to provide more accurate and personalized answers.

[0060] In operation S320 , under the condition that it is determined that no historical conversation record exists, the user's initial question is determined to be a target question input into the business question-answering model.

[0061] For example, if a user enters an initial question such as "I want to know how to change the credit card repayment date" in the customer service question and answer box in a bank APP, the background system will check whether there is a historical conversation record. If there is no historical conversation record, the initial question "I want to know how to change the credit card repayment date" will be directly set as the target question.

[0062] Through the embodiments of the present application, the similarity between the current question and the user's historical questions is calculated based on the conversation records, and the historical questions are selected as supplementary information for the current question instead of using all historical records as model input, so as to reduce the model calculation cost and improve the efficiency of the reply.

[0063] In operation S220, the target question is input into the business question-answering model to obtain a corresponding target answer result.

[0064] According to an embodiment of the present application, the business question-answering model is obtained by training a target large language model based on a training set in a preset business knowledge base. The target large language model is obtained by adjusting the architecture of the initial large language model based on fine-tuning technology.

[0065] First, the architecture of the initial large language model is adjusted to obtain the target large language model, which is explained in detail.

[0066] In an embodiment of the present application, the process of adjusting the architecture of the initial large language model based on fine-tuning technology includes: after the initial large language model is determined, adding a bypass structure to each stacked block in the core architecture of the initial large language model, and adding a regularization layer after all fully connected layers in the initial large language model.

[0067] Specifically, a standard large language model is selected as the initial model. Within each stacked block of the initial model, a bypass structure is designed. This bypass structure can be understood as a simple additive connection that passes the input directly to the output, preserving the original information during training. For example, within each encoder layer in the core architecture of the large language model, a bypass connection is added, connecting the input feature vector directly to the output feature vector of that layer.

[0068] Furthermore, a regularization layer can be added after all fully connected layers to reduce the risk of model overfitting. For example, a regularization layer can be inserted after each fully connected layer and a certain dropout rate (for example, 0.5) can be set to improve the generalization ability of the model.

[0069] According to an embodiment of the present application, after completing the addition of the regularization layer, the adjusted target large language model can also be trained and fine-tuned using a specific task dataset, using an appropriate loss function and optimization algorithm. During the fine-tuning process, hyperparameters such as learning rate and training rounds are adjusted by monitoring the performance on the validation set. Finally, the target large language model can also be performance evaluated and tested to compare the indicators before and after the adjustment. The following is a detailed description of the process of training the target large language model.

[0070] Figure 5 The flowchart of training a target large language model according to an embodiment of the present application is schematically shown.

[0071] like Figure 5 As shown, according to an embodiment of the present application, the business question-answering model is obtained by training the target large language model based on the training set in the preset business knowledge base. This process specifically includes operations S510 to S530.

[0072] In operation S510 , a grid search is used to optimize the ignore ratio of the regularization layer to obtain a target large language model.

[0073] According to an embodiment of the present application, the ignore ratio range of the regularization layer is first set, for example, from 0.2 to 0.5. A grid search method is used to experiment with each possible ignore ratio. For each ratio value, training is performed using the same training set, and performance indicators such as loss and accuracy on the validation set are evaluated after each round of training. The model performance under different ignore ratios is then recorded, and the ignore ratio that performs best on the validation set is ultimately selected to determine the structure of the target large language model.

[0074] In operation S520, a matrix representation is performed on each data item in the training set based on the pre-trained word vector model, where the data item is a business-related question-answer pair.

[0075] Exemplarily, a pre-trained word vector model is used to process each question-answer pair in the training set. For example, for each question-answer pair, each word in the question and answer is converted into a corresponding word vector, and these word vectors are integrated into a matrix representation. For another example, average pooling or maximum pooling methods can be used to integrate the vectors of multiple words into an overall vector representation of the question and answer for subsequent model training.

[0076] Exemplarily, the preset business knowledge base can be understood as selecting channels according to needs to collect information about banking business and integrating relevant records of existing banking business processes to construct a knowledge base related to banking business. Each data item in the training set of the preset business knowledge base includes a question-answer pair of query question information and corresponding question response results.

[0077] It should be noted that before using the training set for model training, each data item will be cleaned. Specifically, for each data item in the banking business-related knowledge base, sensitive question and answer information will be removed, such as all question and answer pairs containing sensitive information such as the user's account name, password, and asset details, as well as some data items with invalid responses.

[0078] In operation S530 , the matrix representation of each data item is input into the target large language model to complete the training and obtain a business question answering model.

[0079] In an embodiment of the present application, after the matrix representation of each data item is input into the target large language model, the matrix representation of each predicted answer is obtained; the target large language model is adjusted based on the loss function between the matrix representation of each predicted answer and the true answer result in the question-answer pair.

[0080] Specifically, the matrix representation of each question-answer pair obtained in step S520 is input into the target large language model for training. During the training process, a matrix representation of the predicted answer is generated based on the input, and the loss between the predicted answer and the true answer is calculated. For example, a cross-entropy loss function is used to evaluate the accuracy of the predicted answer. Based on the calculated loss, the parameters of the target large language model are adjusted through a backpropagation algorithm to reduce the loss and improve the model's predictive ability.

[0081] Furthermore, after each training cycle, the model performance is evaluated using a retained validation set to check the match between the predicted answers and the true answers. If the model performance does not meet expectations, further adjustments are made to hyperparameters such as the learning rate and batch size, and training is repeated until the model performance stabilizes and achieves satisfactory results.

[0082] It should be understood that due to reasons such as the security of banking services and customer privacy, the amount of data in the question-and-answer knowledge base for banking services is difficult to support the training requirements of a large model, and fine-tuning the pre-trained large language model based on fine-tuning technology can greatly reduce the above-mentioned problems. Through the embodiments of the present application, by fine-tuning the structure of the large language model, the parameters that need to be updated during the training process are reduced, and a banking business question-and-answer model is obtained by training based on the established business knowledge database. It not only effectively reduces the negative impact of the model caused by the small amount of data, but also further enhances the model's understanding of knowledge in specific business areas, providing users with more accurate answers. In addition, by adjusting the architecture of the large language model through fine-tuning technology, the system can flexibly adapt to different business scenarios and handle complex business question-and-answer tasks.

[0083] According to the above embodiment, the initial large language model is structurally adjusted based on the fine-tuning technology to obtain the target large language model, and then the target large language model is trained according to the training set in the preset business knowledge base to obtain a business question and answer model, so that the business question and answer model has efficient question and answer capabilities and can meet the needs of specific business scenarios.

[0084] Furthermore, the target question is input into the business question-answering model to obtain the corresponding target answer result.

[0085] In operation S230, the target answer result is displayed.

[0086] According to an embodiment of the present application, displaying the target answer result includes: converting the target answer result into a corresponding target display form according to the target display mode; obtaining the user's feedback result and adjusting the target display form in real time.

[0087] For example, the target answer output by the model is converted into a text representation and then displayed to the user through a display box on the terminal app. Furthermore, user feedback can be obtained and the target display format can be adjusted in real time. This allows the display of the target answer to dynamically adapt to user needs, improving the interactivity of the question-and-answer session and user satisfaction.

[0088] Through the embodiments of the present application, it is possible to obtain user feedback in real time and adjust the target display format according to the feedback results, so that the information display is more in line with user needs, thereby improving user satisfaction.

[0089] Based on the above business question and answer method, this application also provides a business question and answer device. Figure 6 The device is described in detail.

[0090] Figure 6 The structural block diagram of the business question-and-answer device according to an embodiment of the present application is schematically shown.

[0091] like Figure 6 As shown, the business question-answering device 600 of this embodiment includes an initial question acquisition module 610 , a result acquisition module 620 and a result display module 630 .

[0092] The initial question acquisition module 610 is used to acquire the user's initial question and determine the target question to be input into the business question-answering model. In one embodiment, the initial question acquisition module 610 can be used to perform the operation S210 described above, which will not be repeated here.

[0093] The result acquisition module 620 is used to input the target question into the business question-answering model to obtain the corresponding target answer result. In one embodiment, the result acquisition module 620 can be used to perform the operation S220 described above, which will not be repeated here.

[0094] The result display module 630 is used to display the target answer result. In one embodiment, the result display module 630 can be used to perform the operation S230 described above, which will not be repeated here.

[0095] According to an embodiment of the present application, the business question-answering model is obtained by training the target large language model based on the training set in the preset business knowledge base; and the target large language model is obtained by adjusting the architecture of the initial large language model based on fine-tuning technology.

[0096] In an embodiment of the present application, the initial question acquisition module 610 includes a first submodule for acquiring a supplementary question and a second submodule for acquiring a supplementary question.

[0097] The first submodule for obtaining supplementary questions is configured to, upon determining the existence of historical conversation records, calculate the similarity between the user's initial question and each historical question, and identify those historical questions that exceed a preset similarity threshold as supplementary questions for the target question. The target question includes the supplementary question and the initial question. In one embodiment, the first submodule for obtaining supplementary questions can be used to perform operation S310 described above and will not be further described here.

[0098] The second submodule for obtaining a supplementary question is used to determine that the user's initial question is the target question in the input business question-answering model when it is determined that no historical conversation record exists. In one embodiment, the second submodule for obtaining a supplementary question can be used to perform operation S320 described above and will not be repeated here.

[0099] In an embodiment of the present application, the first submodule for obtaining a supplementary question includes a matrix representation unit, a similarity calculation unit, a supplementary question determination unit, and a target question determination unit.

[0100] The matrix representation unit is used to perform matrix representation on the initial question and the historical question based on the pre-trained word vector model. In one embodiment, the matrix representation unit can be used to perform the operation S410 described above, which will not be repeated here.

[0101] A similarity calculation unit is configured to calculate the similarity between the vector representation of the initial question and the vector representations of all historical questions using cosine similarity, thereby obtaining a similarity score between the initial question and each historical question. In one embodiment, the similarity calculation unit may be configured to perform operation S420 described above, which will not be further described herein.

[0102] The supplementary question determination unit is configured to randomly select any one of the top three historical questions with the similarity scores as the supplementary question. In one embodiment, the supplementary question determination unit may be configured to perform the operation S430 described above, which will not be described in detail here.

[0103] The target question determination unit is configured to combine the supplementary question with the initial question to obtain the target question. In one embodiment, the target question determination unit may be configured to perform the operation S440 described above, which will not be described in detail herein.

[0104] In an embodiment of the present application, the result acquisition module 620 includes a target large language model training submodule.

[0105] Furthermore, the target large language model training submodule includes a target large language model acquisition unit, a matrix representation unit and a training unit.

[0106] The target large language model acquisition unit is configured to optimize the ignore ratio of the regularization layer using grid search to obtain the target large language model. In one embodiment, the target large language model acquisition unit can be configured to perform operation S510 described above, which will not be described in detail here.

[0107] The matrix representation unit is used to perform a matrix representation on each data item in the training set based on the pre-trained word vector model, where the data item is a business-related question-answer pair. In one embodiment, the matrix representation unit can be used to perform operation S520 described above, which will not be repeated here.

[0108] The training unit is used to input the matrix representation of each data item into the target large language model to complete the training and obtain the business question answering model. In one embodiment, the training unit can be used to perform the operation S530 described above, which will not be repeated here.

[0109] According to embodiments of the present application, any multiple modules among the initial question acquisition module 610, the result acquisition module 620, and the result display module 630 can be combined into a single module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module. According to embodiments of the present application, at least one of the initial question acquisition module 610, the result acquisition module 620, and the result display module 630 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware through any other reasonable means of circuit integration or packaging, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or any appropriate combination of any of these. Alternatively, at least one of the initial question acquisition module 610, the result acquisition module 620, and the result display module 630 can be at least partially implemented as a computer program module that, when executed, can perform the corresponding functionality.

[0110] Figure 7 A block diagram of an electronic device suitable for implementing a business question-and-answer method according to an embodiment of the present application is schematically shown.

[0111] like Figure 7 As shown, an electronic device 700 according to an embodiment of the present application includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage unit 708 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to the embodiment of the present application.

[0112] Various programs and data required for the operation of the electronic device 700 are stored in the RAM 703. The processor 701, ROM 702, and RAM 703 are connected to each other via a bus 704. The processor 701 performs various operations of the method flow according to the embodiment of the present application by executing the programs in the ROM 702 and / or RAM 703. It should be noted that the programs may also be stored in one or more memories other than the ROM 702 and the RAM 703. The processor 701 may also perform various operations of the method flow according to the embodiment of the present application by executing the programs stored in one or more memories.

[0113] According to an embodiment of the present application, electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to bus 704. Electronic device 700 may also include one or more of the following components connected to I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 708 including a hard disk; and a communication section 709 including a network interface card such as a LAN card or modem. Communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to I / O interface 705 as needed. Removable media 711, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 710 as needed, so that computer programs read from the removable media can be installed into storage section 708 as needed.

[0114] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when the one or more programs are executed, the method according to the embodiments of this application is implemented.

[0115] According to an embodiment of the present application, a computer-readable storage medium may be a non-volatile computer-readable storage medium, and may include, for example, but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction change execution system, apparatus, or device. For example, according to an embodiment of the present application, a computer-readable storage medium may include the ROM 702 and / or RAM 703 described above and / or one or more memories other than ROM 702 and RAM 703.

[0116] The embodiments of the present application also include a computer program product, which includes a computer program containing program code for executing the method shown in the flowchart. When the computer program product is executed in a computer system, the program code is used to enable the computer system to implement the business question-and-answer method provided in the embodiments of the present application.

[0117] The computer program executes the above functions defined in the system / device of the embodiment of the present application when the computer program is executed by the processor 701. According to the embodiment of the present application, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0118] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 709, and / or installed from a removable medium 711. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0119] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 709, and / or installed from a removable medium 711. When the computer program is executed by the processor 701, the above-described functions defined in the system of the embodiment of the present application are performed. According to the embodiment of the present application, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0120] According to an embodiment of the present application, the program code for executing the computer program provided by the embodiment of the present application can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0121] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of the boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0122] Those skilled in the art will appreciate that the features described in the various embodiments of this application may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in this application. In particular, the features described in the various embodiments of this application may be combined and / or coupled in various ways without departing from the spirit and teachings of this application. All such combinations and / or couplings fall within the scope of this application.

[0123] The embodiments of the present application have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present application. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be advantageously used in combination. Without departing from the scope of the present application, those skilled in the art may make various substitutions and modifications, and these substitutions and modifications should all fall within the scope of the present application.

Claims

1. A business question-answering method, characterized in that: The method comprises: Obtain the user's initial question and determine the target question to be input into the business question-answering model; Input the target question into the business question-answering model to obtain the corresponding target answer result; Display the target answer result; The business question-answering model is obtained by training a target large language model based on a training set in a preset business knowledge base; and The target large language model is obtained by adjusting the architecture of the initial large language model based on fine-tuning technology.

2. The business question-answering method according to claim 1, characterized in that: The process of obtaining the user's initial question and determining the target question to be input into the business question-answering model includes: If it is determined that there are historical conversation records, by calculating the similarity between the user's initial question and each historical question, questions with a similarity above a preset threshold among all historical questions are determined as supplementary questions in the target question, where the target question includes the supplementary question and the initial question; Under the condition that it is determined that no historical conversation record exists, the user's initial question is determined to be the target question input into the business question-answering model.

3. The business question-answering method according to claim 2, characterized in that: The method of calculating the similarity between the user's initial question and each historical question and determining questions with a similarity higher than a preset threshold among all historical questions as supplementary questions in the target question includes: Based on the pre-trained word vector model, the initial question and historical questions are represented by a matrix; Calculate the similarity between the vector representation of the initial question and the vector representations of all historical questions using cosine similarity to obtain a similarity score between the initial question and each historical question; Randomly select one of the top three historical questions with the same similarity scores as a supplementary question; The target question is obtained by concatenating the supplementary question with the initial question.

4. The business question-answering method according to claim 1, wherein: The process of adjusting the architecture of the initial large language model based on the fine-tuning technology includes: After the initial large language model is determined, a bypass structure is added to each stacked block in the core architecture of the initial large language model, and a regularization layer is added after all fully connected layers in the initial large language model.

5. The business question-answering method according to claim 4, characterized in that: The business question-answering model is obtained by training the target large language model based on the training set in the preset business knowledge base. The training process includes: Optimizing the ignore ratio of the regularization layer using grid search to obtain the target large language model; Based on the pre-trained word vector model, each data item in the training set is represented by a matrix, where the data item is a business-related question-answer pair; The matrix representation of each data item is input into the target large language model to complete the training and obtain the business question-answering model.

6. The business question-answering method according to claim 4, characterized in that: The method further comprises: After the matrix representation of each data item is input into the target large language model, a matrix representation of each predicted answer is obtained; The target large language model is adjusted according to a loss function between the matrix representation of each predicted answer and the true answer result in the question-answer pair.

7. The business question-answering method according to claim 1, wherein: The displaying of the target answer result includes: According to the target display mode, converting the target answer result into a corresponding target display form; Obtain user feedback and adjust the target display format in real time.

8. The business question-answering method according to claim 1, wherein: The method further comprises: Obtain user authorization to enter the user's initial question; After obtaining the user's authorization to input the initial question, the user's initial question is obtained.

9. A business question-answering device, characterized in that: The device comprises: The initial question acquisition module is used to obtain the user's initial question and determine the target question to be input into the business question-answering model; A result acquisition module is used to input the target question into the business question-answering model to obtain the corresponding target answer result; A result display module, used to display the target answer result; The business question-answering model is obtained by training a target large language model based on a training set in a preset business knowledge base; and The target large language model is obtained by adjusting the architecture of the initial large language model based on fine-tuning technology.

10. An electronic device comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

12. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images