Intelligent question and answer method, equipment, storage medium and device

Abstract

The invention relates to the technical field of computers, and discloses an intelligent question and answer method, equipment, a storage medium and a device, and the method comprises the steps: searching private data related to a question of a user from a preset private document according to the received question of the user; and inputting the private data related to the question of the user and the question of the user into a large language model to obtain an answer. When the big language model answers the question of the user, the big language model refers to the input private data related to the question of the user, so that the accuracy of the obtained answer is relatively high when the question related to the private data is answered; and the large language model has very strong semantic understanding and creation capabilities and can generate continuous texts similar to human dialogues, so that the invention provides the intelligent question and answer method which can realize smooth dialogues with people and accurately answer questions related to private data.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an intelligent question-answering method, device, storage medium and apparatus. Background Art

[0002] With the advent of the digital age, intelligent question-answering has become one of the important means for enterprises to improve the quality and efficiency of customer service. Its role can not only greatly reduce the workload of customer service personnel and shorten customer waiting time, but also improve user experience and loyalty, and enhance the competitiveness of enterprises.

[0003] However, traditional machine learning methods have some limitations in the application of intelligent question answering. For example, they cannot handle the complex logic and semantics in user questions. At the same time, they require a large amount of data to train and improve the model. After the document data is added or updated, the model needs to be retrained due to insufficient generalization ability, which is very costly.

[0004] The birth of large language models has brought new opportunities and challenges to intelligent question answering. Large language models can generate and optimize texts infinitely, have strong semantic understanding and creativity, and can generate continuous texts similar to human conversations. Compared with traditional machine learning methods, when using large language models for intelligent question answering, they have better generalization and migration capabilities, do not require training, and can reduce costs.

[0005] However, the training data of the open source big language model does not involve the internal private data of the enterprise or group. Therefore, when the big language model is used to answer questions related to private data, the answers given by the big language model are not accurate.

[0006] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0007] The main purpose of the present invention is to provide an intelligent question-answering method, device, storage medium and apparatus, aiming to solve the problem that large language models are inaccurate when answering questions related to private data.

[0008] To achieve the above-mentioned purpose, the present invention provides an intelligent question-answering method, comprising: searching for private data related to the user's question from a preset private document based on the received user's question; inputting the private data related to the user's question and the user's question into a large language model to obtain an answer.

[0009] Optionally, the preset private document includes business problems and private data corresponding to the business problems; searching for private data related to the user's problem from the preset private document based on the received user's problem includes: extracting the business problem from the received user's problem; and searching for private data corresponding to the business problem from the preset private document based on the business problem.

[0010] Optionally, before searching for private data corresponding to the business question from the preset private document according to the business question, it also includes: determining whether the user is asking the question for the first time; if the user is asking the question for the first time, executing the step of searching for private data corresponding to the business question from the preset private document according to the business question.

[0011] Optionally, after determining whether it is the first time for the user to ask a question, the method further includes: if it is not the first time for the user to ask a question, determining whether a business question extracted from the user's question appears in the user's historical chat records; if the business question appears in the user's historical chat records, inputting the user's question and the user's historical chat records into a large language model to obtain an answer.

[0012] Optionally, after determining whether the business question extracted from the user's question appears in the user's historical chat records, it also includes: if the business question does not appear in the user's historical chat records, searching for private data corresponding to the business question from the preset private document based on the business question; and inputting the private data corresponding to the business question, the user's question and the user's historical chat records into a large language model to obtain an answer.

[0013] Optionally, the historical chat record of the user and the received question of the user belong to the same round of dialogue.

[0014] Optionally, the private data corresponding to the business problem in the preset private document includes any one or any combination of the following: business problem name, business problem type, business problem reference document, business problem reference link.

[0015] Optionally, before searching for private data related to the user's question from a preset private document based on the received user's question, it also includes: determining whether the received user's question is a first-category business question; if the received user's question is a first-category business question, then executing the step of searching for private data related to the user's question from a preset private document based on the received user's question, wherein the preset private document includes relevant data of the first-category business question.

[0016] Optionally, determining whether the received user question is a first-category business question includes: determining whether the received user question is a first-category business question according to a pre-trained classification model.

[0017] Optionally, after determining whether the received user question is a first-category business question, the method further includes: if the received user question is not a first-category business question, prompting the user to input a question related to the first-category business question.

[0018] Optionally, the private document includes vulnerability information related documents.

[0019] In addition, to achieve the above-mentioned purpose, the present invention also proposes an intelligent question and answer device, which includes a memory, a processor, and an intelligent question and answer program stored in the memory and executable on the processor, and the intelligent question and answer program is configured to implement the intelligent question and answer method as described above.

[0020] In addition, to achieve the above-mentioned purpose, the present invention also proposes a storage medium, on which an intelligent question and answer program is stored, and when the intelligent question and answer program is executed by a processor, the intelligent question and answer method as described above is implemented.

[0021] In addition, to achieve the above-mentioned purpose, the present invention also proposes an intelligent question-answering device, including: a query module, a processing module, and a large language model interface module; the query module is used to search for private data related to the user's question from a preset private document based on the received user's question; the processing module is used to input the private data related to the user's question and the user's question into the large language model interface module to obtain an answer.

[0022] Optionally, the preset private document includes business problems and private data corresponding to the business problems; the query module is specifically used to extract business problems from the questions received from the user, and search for private data corresponding to the business problems from the preset private document based on the business problems.

[0023] Optionally, the query module is specifically used to determine whether the user is asking a question for the first time before searching for private data corresponding to the business question from the preset private document based on the business question; if the user is asking a question for the first time, then executing the step of searching for private data corresponding to the business question from the preset private document based on the business question.

[0024] Optionally, the query module is also used to, after determining whether the user is asking a question for the first time, determine whether the business question extracted from the user's question appears in the user's historical chat records if it is not the first time for the user to ask a question; the processing module is also used to input the user's question and the user's historical chat records into the large language model interface module to obtain an answer if the business question appears in the user's historical chat records.

[0025] Optionally, the query module is also used to, after determining whether the business question extracted from the user's question appears in the user's historical chat records, search for private data corresponding to the business question from the preset private document according to the business question if the business question does not appear in the user's historical chat records; the processing module is also used to input the private data corresponding to the business question, the user's question and the user's historical chat records into the large language model interface module to obtain an answer.

[0026] Optionally, the historical chat record of the user and the received question of the user belong to the same round of dialogue.

[0027] Optionally, the private data corresponding to the business problem in the preset private document includes any one or any combination of the following: business problem name, business problem type, business problem reference document, business problem reference link.

[0028] Optionally, the query module is specifically used to determine whether the received user question is a first-category business question before searching for private data related to the user's question from a preset private document based on the received user's question; if the received user question is a first-category business question, then executing the step of searching for private data related to the user's question from a preset private document based on the received user's question, wherein the preset private document includes relevant data of the first-category business question.

[0029] Optionally, the query module is specifically used to determine whether the received user question is a first-category business question based on a pre-trained classification model.

[0030] Optionally, the processing module is further configured to, after determining whether the received user question is a first-category business question, prompt the user to input a question related to the first-category business question if the received user question is not a first-category business question.

[0031] Optionally, the private document includes vulnerability information related documents.

[0032] The present invention discloses an intelligent question-answering method, which first searches for private data related to the user's question from a preset private document based on the received user's question, and then inputs the private data related to the user's question and the user's question into a large language model together to obtain an answer. Since the private data related to the user's question found in the present invention is input into the large language model together with the user's question, the large language model will refer to the input private data related to the user's question when answering the user's question. Therefore, when answering questions related to private data, the accuracy of the answers obtained is higher. And because the large language model has strong semantic understanding and creative capabilities, it can generate continuous text similar to human conversations. Therefore, the present invention provides an intelligent question-answering method that can communicate fluently with people and accurately answer questions related to private data. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of the structure of the hardware operating environment of the intelligent question-answering device involved in the embodiment of the present invention;

[0034] Figure 2 Schematic diagram of the intelligent question-answering method of the present invention;

[0035] Figure 3 A schematic diagram of a flow chart of a first embodiment of an intelligent question-answering method of the present invention;

[0036] Figure 4 A schematic diagram of a flow chart of a second embodiment of the intelligent question-answering method of the present invention;

[0037] Figure 5 Schematic diagram of the flow chart of the third embodiment of the intelligent question-answering method of the present invention;

[0038] Figure 6 Schematic diagram of the fourth embodiment of the intelligent question-answering method of the present invention;

[0039] Figure 7 Schematic diagram of the structure of the intelligent question-answering device of the present invention.

[0040] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0041] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0042] First, the nouns involved in the present invention are explained in a unified manner, as follows:

[0043] 1. A large language model refers to a deep learning model trained with a large amount of text data, which can generate natural language text or understand the meaning of language text. Large language models can handle a variety of natural language tasks, such as text classification, question answering, and dialogue, and are an important path to artificial intelligence. Large language models include generative pre-trained transformers (GPT), general language models (GLM), etc., such as GPT-4, ChatGPT, ChatGLM, etc.

[0044] 2. Private data refers to data produced by an individual or a single legal person. This data is kept by the producer himself, used internally, not disclosed to the public, and does not involve the interests of third parties.

[0045] 3. The BERT model refers to a pre-trained language representation model (Bidirectional Encoder Representation from Transformers, BERT). The training method of the BERT model belongs to the unsupervised fine-tuning method, which is mainly divided into the following two steps: (1) pre-training, training the model on a large amount of unlabeled data of various tasks; (2) fine-tuning, adding an output layer to the BERT model according to a specific downstream task, initializing the model with pre-trained parameters, and then fine-tuning the model parameters on a labeled dataset for the task.

[0046] 4. Vulnerability: In the computer field, a vulnerability refers to a defect in the specific implementation of hardware, software, protocol or system security policy, which can enable attackers to access or damage the system without authorization.

[0047] Reference Figure 1 , Figure 1 It is a schematic diagram of the structure of the hardware operating environment of the intelligent question-answering device involved in the embodiment of the present invention.

[0048] like Figure 1As shown, the intelligent question-answering device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), and the user interface 1003 may also include a standard wired interface and a wireless interface. The wired interface of the user interface 1003 may be a USB interface in the present invention. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (Wireless-Fidelity, Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM) memory, or a stable memory (Non-volatile Memory, NVM), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0049] Those skilled in the art will understand that Figure 1 The structure shown in the figure does not constitute a limitation on the intelligent question-answering device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0050] like Figure 1 As shown, the memory 1005 identified as a computer storage medium may include an operating system, a network communication module, a user interface module, and an intelligent question-answering program.

[0051] exist Figure 1 In the intelligent question and answer device shown, the network interface 1004 is mainly used to connect to the background server and communicate data with the background server; the user interface 1003 is mainly used to connect to the user device; the intelligent question and answer device calls the intelligent question and answer program stored in the memory 1005 through the processor 1001, and executes the intelligent question and answer method provided by the embodiment of the present invention.

[0052] Based on the above hardware structure, an embodiment of the intelligent question-answering method of the present invention is proposed.

[0053] In order to answer questions related to private data raised by users, a pre-trained machine learning model can be used to answer them. The problems with this method are: it cannot handle the complex logic and semantics in user questions, the fluency and naturalness of conversations with people are not good, and after the document data is added or updated, the machine learning model needs to be retrained, which is costly. However, open source large language models such as ChatGPT and ChatGLM do not require training, have good generalization and migration capabilities, and have high fluency and naturalness in conversations with people. Therefore, open source large language models are often used to answer user questions. However, since the training data of open source large language models does not involve private data within enterprises or groups, when using open source large language models to answer questions related to private data, the answers given by the large language models are not accurate.

[0054] In view of this, the present invention proposes an intelligent question-answering method, such as Figure 2 As shown, the intelligent question-answering method in the present invention is implemented by combining a large language model with data retrieval technology. First, private data related to the user's question is searched from a preset private document based on the received user's question. Then, the private data related to the user's question is input into the large language model together with the user's question to obtain an answer. Since the private data related to the user's question found in the present invention is input into the large language model together with the user's question, the large language model will refer to the input private data related to the user's question when answering the user's question. Therefore, when answering questions related to private data, the accuracy of the answer obtained is higher. And because the large language model has strong semantic understanding and creative capabilities, it can generate continuous text similar to human conversations. Therefore, the present invention provides an intelligent question-answering method that can communicate fluently with people and accurately answer questions related to private data.

[0055] The intelligent question-answering method provided by an embodiment of the present invention is described in detail below with reference to the accompanying drawings.

[0056] Figure 3 This is a flow chart of the first embodiment of the intelligent question-answering method of the present invention. The intelligent question-answering method in this embodiment includes:

[0057] Step S10: searching for private data related to the user's question from a preset private document according to the received user's question.

[0058] In this embodiment, the user's question can be asked by inputting text on the display screen, or by voice questioning.

[0059] The preset private documents are stored in the memory, and the private documents include private data within the enterprise or group. In practice, the private documents include documents related to vulnerability information, documents related to employee handbooks, or documents related to technical information. In this embodiment, the preset private documents can be stored in the memory in the form of a database, or can be directly stored in the memory in the form of a document.

[0060] The private data corresponding to the business problem in the preset private document includes any one or any combination of the following: business problem name, business problem type, business problem reference document, business problem reference link. For example, if the private document includes vulnerability information related documents, then the business problem is a vulnerability-related problem, and the private data corresponding to the vulnerability-related problem may include vulnerability name, vulnerability type, vulnerability reference document, vulnerability reference link, vulnerability repair suggestion, etc.

[0061] Realistically, when searching for private data related to a user's question, a search algorithm may be used to search; if semantic search is required, a neural network model may be trained and the trained neural network model may be used to search.

[0062] Among them, the search algorithm can adopt the term frequency-inverse document frequency index statistical method (Term Frequency-Inverse Document Frequency, TF-IDF), hereinafter referred to as TF-IDF. TF-IDF is a commonly used weighted technology for information retrieval and text mining, which is used to evaluate the importance of a word to a document set or one of the documents in a corpus. The importance of a word increases in direct proportion to the number of times it appears in a document, but at the same time it decreases in inverse proportion to the frequency of its appearance in the corpus. The higher the importance of private data in the preset private document, the greater the relevance of the private data to the user's question. It can be sorted according to the importance, and the private data ranked in the top few places can be determined as private data related to the user's question; a threshold can also be set to determine the private data with an importance score exceeding this threshold as private data related to the user's question.

[0063] The search algorithm may also use the best matching algorithm BM25, where BM is the abbreviation of Best Match and 25 refers to the 25th algorithm iteration. BM25 is the most mainstream algorithm for calculating the similarity score between the question and the document in the current information indexing field. The higher the similarity score between the private data in the preset private document and the user's question, the greater the relevance of the private data to the user's question. A threshold may be set to determine private data with a similarity score exceeding this threshold as private data related to the user's question.

[0064] The search algorithm can also use the web page level algorithm PageRank, which gives each web page a positive real number to indicate the importance of the web page, forming a vector as a whole. The higher the PageRank value, the more important the web page is, and it may be ranked at the front in the ranking of Internet searches. For a web page, if the PageRank value pointing to the web page is higher, the probability of randomly jumping to the web page is higher, and the higher the PageRank value of the web page, the more important the web page is. The higher the PageRank value of the web page where the private data in the preset private document is located, the greater the relevance of the private data to the user's question. A threshold can be set to determine the private data whose PageRank value of the web page where the private data is located exceeds this threshold as private data related to the user's question.

[0065] The neural network model may be a word vector model Word2Vector that converts words into vectors; or it may be a vector encoder trained using a deep learning algorithm, such as BERT.

[0066] Realistically, the private document preset in this embodiment supports online updating of data.

[0067] Step S20: input the private data related to the user's question and the user's question into the large language model to obtain an answer.

[0068] The large language model used in this embodiment is an open source model and does not require a training model. The large language model can be ChatGPT, which can be implemented by calling the ChatGPT interface; if for data security and privacy protection considerations, you can also choose to deploy an open source ChatGPT-like large language model yourself, such as using the open source ChatGLM model for private deployment. The large language model should choose a model that supports longer input lengths as much as possible to support situations where the user inputs longer questions.

[0069] After the large language model understands the user's question, it will refer to the input private data related to the user's question and answer it in combination with the user's question. Therefore, when answering questions related to private data, the accuracy of the answers given is relatively high. And because the large language model has strong semantic understanding and creative capabilities, it can generate continuous text similar to human conversations. Therefore, the present invention adopts a method that combines large language models and data retrieval technology to realize an intelligent question-answering method that can fluently communicate with people and accurately answer questions related to private data.

[0070] Compared with traditional machine learning methods, the intelligent question-answering method based on large language models does not require training and has better generalization and migration capabilities. In other words, large language models have better adaptability to fresh samples. Even if the data in the preset private documents increases, decreases or is updated, there is no need to retrain the model, which can greatly reduce costs.

[0071] Figure 4 This is a flow chart of a second embodiment of the intelligent question-answering method of the present invention. The intelligent question-answering method in this embodiment includes:

[0072] Step S101: extracting business questions from received user questions.

[0073] In this example, a regular expression matching method can be used to extract business problems from the questions received from the users. The regular expression matching method is a string matching method that is used to match and search whether a known string contains the substring you need. In this embodiment, the regular expression can be determined based on the business problem in the preset private document. For example, if the business problem in the preset private document is the attendance system or the reward system, the regular expression can be the same as the business problem, which is "attendance system" or "reward system." Assuming that the question received from the user is "What is our company's attendance system?", the business problem extracted using the regular expression "attendance system" is "attendance system."

[0074] Realistically, the business problem can be any one or any combination of text, numbers, English letters or special characters. For example, the business problem can be CVE-2021-2429, performance appraisal management, salary and benefits, etc. The corresponding regular expressions can be "CVE-2021-2429", "performance appraisal", "salary and benefits".

[0075] Step S102: searching for private data corresponding to the business problem from preset private documents according to the business problem.

[0076] In this embodiment, the preset private document includes the correspondence between business questions and private data, and each business question corresponds to the private data related to the business question. Therefore, after extracting the business question from the received user's question, the private data corresponding to the business question is directly searched in the preset private document according to the business question, which can improve the query speed, especially when the data in the private document is complex, which greatly improves the query efficiency.

[0077] In practice, an index directory can be created for business problems in a preset private document, and the index directory includes a number of business problems and storage locations of private data corresponding to the business problems. After extracting the business problem from the received user's question, the storage location of the private data corresponding to the business problem can be directly searched in the index directory according to the business problem, so as to read the private data corresponding to the business problem at the storage location.

[0078] Step S201: input the private data corresponding to the business question and the user's question into the large language model to obtain an answer.

[0079] It can be understood that this embodiment is Figure 3 One implementation of the embodiment shown is as follows: Figure 3 Step S10 in the embodiment may include step S101 and step S102 in the present embodiment. Figure 3 Step S20 in the example may specifically be: Figure 4 Step S201 in .

[0080] Figure 5 : is a flow chart of a third embodiment of the intelligent question-answering method of the present invention. In this embodiment, the intelligent question-answering method includes:

[0081] Step S01: Determine whether the received user's question is a first-category business question. If the received user's question is a first-category business question, execute step S10; if the received user's question is not a first-category business question, execute step S02.

[0082] In this embodiment, it is determined whether the received user's question belongs to the first type of business question. For example, the first type of business question may be a security question or an employee handbook question.

[0083] In practice, determining whether the received user's question is a first-category business question includes: determining whether the received user's question is a first-category business question according to a pre-trained classification model. In this embodiment, the pre-trained classification model is used to determine whether the user's question is a first-category business question, which has high reliability and fast processing speed.

[0084] In this embodiment, the pre-trained classification model adopts the BERT model. The training method of the BERT model is as follows: use the questions generated by GPT-4 for testing, the training set data must be more than the test set data, the ratio of positive and negative samples in the training set and the test set is 1:1, and the model training only trains the classification accuracy. When the classification accuracy is greater than the preset threshold, the BERT model training is completed.

[0085] In an example, the first type of business problem is a security problem, and the pre-trained classification model is a BERT model. When training the BERT model, the training set has 14,856 data items and the test set has 3,000 data items. The ratio of positive and negative samples in the training set and the test set is 1:1. The specific parameters of the trained BERT model are as follows:

[0086] {

[0087] "vocab_size":130528,

[0088] "hidden_size":768,

[0089] "num_hidden_layers":4,

[0090] "num_attention_heads":12,

[0091] "intermediate_size":3072,

[0092] "hidden_act":"gelu",

[0093] "hidden_dropout_prob":0.1,

[0094] "attention_probs_dropout_prob":0.1,

[0095] "max_position_embedding":64,

[0096] "type_vocab_size":2,

[0097] "initializer_range": 0.02,

[0098] "layer_norm_eps":1e-12,

[0099] "pad_token_id":θ

[0100] }

[0101] Among them, the parameters such as the dictionary size vocab_size, the hidden layer vector dimension hidden_size, the number of hidden layers num_hidden_layers, and the number of attention mechanism heads num_attention_heads are the parameters adjusted after training, and the other parameters are the default parameters and are not adjusted during the training process. The BERT model uses the word segmenter and dictionary of the large language model. In this example, the classification accuracy of the trained BERT model is as high as 99.03%.

[0102] Step S10: searching for private data related to the user's question from a preset private document according to the received user's question, wherein the preset private document includes data related to the first type of business question.

[0103] If the received user's question is a first type of business question, step S10 is executed. In this embodiment, step S10 is Figure 3 Step S10 of the embodiment shown is substantially the same, except that the private document preset in step S10 of the embodiment includes relevant data of the first type of business problem, and Figure 3The preset private document in step S10 of the illustrated embodiment may include data related to the first category of business issues, or may not include data related to the first category of business issues, which is not limited in this embodiment.

[0104] In this embodiment, by determining whether the received user's question is a first-category business question, the user's questions that belong to the first-category business question are screened out. The intelligent question-answering method of this embodiment processes and answers the user's questions that belong to the first-category business question, and does not answer the user's questions that do not belong to the first-category business question, thereby avoiding the problem of wasting resources caused by answering other irrelevant questions that do not belong to the first-category business question.

[0105] Step S20: Input the private data related to the user's question and the user's question into the large language model to obtain an answer. Figure 3 Step S20 of the illustrated embodiment is substantially the same and will not be described again in this embodiment to avoid repetition.

[0106] Step S02: prompting the user to input questions related to the first category of business questions. In this embodiment, if the received user question is not a first category of business question, the user is prompted to input a question of the correct category, and a prompt text may be displayed on the display interface for prompting.

[0107] Figure 6 : is a flow chart of a fourth embodiment of the intelligent question-answering method of the present invention. In this embodiment, the intelligent question-answering method includes:

[0108] Step S101: extracting business questions from received user questions.

[0109] In this embodiment, step S101 and Figure 4 The step S101 of the illustrated embodiment is substantially the same and will not be described again in this embodiment to avoid repetition.

[0110] Step S03: Determine whether the user is asking a question for the first time. If it is determined to be yes, execute step S102; if it is determined to be no, execute step S04.

[0111] In this embodiment, when receiving a user's question, the user's identity identifier can be recorded synchronously, and whether the user is asking a question for the first time can be determined based on the user's identity identifier. Specifically, whether the user is asking a question for the first time can be determined by determining whether the user's identity identifier is recorded in the historical chat record.

[0112] In this embodiment, step S101 is shown before step S03 for the convenience of drawing. However, in actual applications, as long as step S101 and step S03 are both executed before step S04, step S101 and step S03 can be executed synchronously, and step S101 can also be executed after step S03.

[0113] Step S102: searching for private data corresponding to the business problem from preset private documents according to the business problem.

[0114] Step S201: input the private data corresponding to the business question and the user's question into the large language model to obtain an answer.

[0115] In this embodiment, if the user asks a question for the first time, step S102 and step S201 are executed. Figure 4 Step S102 of the embodiment shown is substantially the same as step S201. Figure 4 Step S201 of the illustrated embodiment is substantially the same and will not be described again in this embodiment to avoid repetition.

[0116] Step S04: Determine whether the business question extracted from the user's question appears in the user's historical chat record. If it is determined to be yes, execute step S05, if it is determined to be no, execute step S102.

[0117] Step S05: Input the user's question and the user's historical chat records into the large language model to obtain the answer.

[0118] In this embodiment, if the business question extracted from the user's question appears in the user's historical chat records, it means that the large language model has already searched for private data related to the above business question in the process of historical question and answer. In other words, private data related to the above business question already exists in the historical chat records. Therefore, the user's question and the user's historical chat records can be directly input into the large language model to obtain the answer. There is no need to re-search for private data related to the business question from the preset private documents based on the business question, which can improve the problem processing speed.

[0119] Step S102: searching for private data corresponding to the business problem from preset private documents according to the business problem.

[0120] In this embodiment, step S102 and Figure 4 Step S102 of the illustrated embodiment is substantially the same and will not be described again in this embodiment to avoid repetition.

[0121] Step S06: Input the private data corresponding to the business question, the user's question and the user's historical chat records into the large language model to obtain an answer.

[0122] In this embodiment, if the business question extracted from the user's question does not appear in the user's historical chat records, it means that the large language model has not searched for private data corresponding to the above business question during the historical question-answering process, that is, the private data related to the above business question does not exist in the historical chat records. Therefore, it is necessary to execute step S102 to search for private data corresponding to the business question from the preset private document according to the business question, and then execute step S06 to input the private data corresponding to the business question, the user's question and the user's historical chat records into the large language model to obtain the answer. The large language model can answer according to the user's historical chat records, while ensuring the accuracy of the answer, and improving the continuity of the user's dialogue with the large language model.

[0123] The prompt word template of the large language model in this embodiment is as follows:

[0124] "a+'Please answer based on this text:'+input(), history"

[0125] Among them, a is the private data related to the user's question searched in the preset private document, the content in the brackets after input is the user's question, and history is the user's historical chat record. The large language model can support the three forms of input of the above steps S201, S05 and S06.

[0126] It is worth mentioning that the historical chat records and the user's questions mentioned in this embodiment can belong to the same round of dialogue, which can avoid too many historical chat records and the long time to search for historical chat records, which affects the speed of problem processing. Specifically, the dialogue between the user and the large language model during the entire period from opening to closing the dialogue interface can be set as the same round of dialogue, and the chat content belonging to the same round of dialogue can be determined by recording the opening time and closing time of the dialogue interface. If the user reopens the dialogue interface, it is regarded as a new round of dialogue.

[0127] Realistically, after a round of conversation between the user and the large language model ends, that is, after the conversation interface is closed, the content of the chat history can be cleared to avoid resource occupation.

[0128] It is feasible that if no business question can be extracted from the received user's question, it is determined whether the user is asking the question for the first time. If it is not the first time, the business question in the most recent question and answer in the historical chat record is used as the business question in the current conversation to ensure the continuity of the conversation; if it is the first time to ask the question, the user is prompted to enter the correct question.

[0129] It should be understood that the division of the above embodiments is to help those skilled in the art understand the present invention, rather than to limit the implementation of the present invention to the exemplified embodiments. Those skilled in the art can obviously make various equivalent modifications or changes based on the above embodiments, and the embodiments formed after such modifications or changes also fall within the protection scope of the embodiments of the present invention.

[0130] The specific implementation of the intelligent question-answering method of each embodiment provided by the present invention must be combined with actual private documents.

[0131] The following takes the case where the user's question is a security question and the preset private document includes a business question and private data related to the business question as an example to illustrate the implementation process of the intelligent question-answering method of the present invention:

[0132] (1) Handle user issues.

[0133] First determine whether the user's question is a security question. If the question entered by the user is not related to security, remind the user that this intelligent question-answering method only supports security-related questions.

[0134] (2) If the user's question is a security question, relevant private data is searched in the preset private documents based on the user's question.

[0135] The preset private documents in this example are vulnerability information related documents, which include 8,000 Common Vulnerabilities and Exposures (CVE) documents, hereinafter referred to as CVE vulnerability documents.

[0136] CVE vulnerability documents include business issues and private data corresponding to the business issues. The following takes a business issue and its corresponding private data as an example:

[0137] {"instruction":"CVE-2021-2429",

[0138] "Outputs":"Vulnerability name: CVE-2021-2429"

[0139] Vulnerability Type: Denial of Service (DoS)

[0140] Vulnerability details and impact: This is a vulnerability found in the InnoDB component in MySQL Server. Affected versions are 8.0.25 and earlier. This vulnerability is difficult to exploit and allows an unauthenticated attacker with network access via multiple protocols to attack MySQL Server. Successful attacks of this vulnerability may cause the MySQL server to become unresponsive or frequently crash (complete DoS).

[0141] Vulnerability attack path: Access via network (AV:N)

[0142] Vulnerability attack difficulty or complexity: High (AC:H)

[0143] Vulnerability PoC reference document: There is currently no publicly available PoC reference document.

[0144] Repair suggestions and steps:

[0145] Update to the latest version of MySQL Server.

[0146] Limit network access to the MySQL Server, using a firewall or other control measures.

[0147] Monitor server performance and logs for potential attacks.

[0148] Scripts to detect whether this vulnerability exists: There is currently no publicly available detection script.

[0149] Data confidentiality impact: Low (the vulnerability does not affect data confidentiality)

[0150] Data security impact: Medium (the attack may cause MySQL server service interruption)

[0151] Data permission requirement: None (PR:N)

[0152] Vulnerability patch link: Please visit [Oracle official website]

[0153] (https: / / www.oracle.com / security-alerts / cpujul2021.html) Download and apply the relevant patches.

[0154] Documentation about this vulnerability: [Oracle July 2021 Critical Patch Update]

[0155] (https: / / www.oracle.com / security-alerts / cpujul2021.html).

[0156] Details of affected components or scope of impact: InnoDB component in Oracle MySQL Server. The affected versions are 8.0.25 and earlier.

[0157] General advice: Update MySQL Server to the latest version as soon as possible and restrict network access to the MySQL server. At the same time, continue to pay attention to security updates and patch information released by Oracle.

[0158] "}

[0159] In the above example, the content "CVE-2021-2429" after "instruction" is a business problem, and the content after "Outputs" is the private data corresponding to the business problem "CVE-2021-2429". Private data includes: vulnerability name, vulnerability type, vulnerability details and impact, vulnerability attack path, vulnerability attack difficulty or complexity, vulnerability PoC reference document, repair suggestions and steps, scripts for detecting whether the vulnerability exists, data permission requirements, vulnerability repair patch links, documents about the vulnerability, details of the affected related components or the scope of impact, and overall reference opinions.

[0160] A regular expression matching algorithm is used to extract business questions from the user's questions. If there is no business question in the user's question, it is determined whether it is the first time the user asks the question. If it is not the first time, the business question in the most recent question and answer in the historical chat record is used as the business question in the current conversation to ensure the continuity of the conversation. If it is the first time the user asks the question, the user is prompted to enter the correct question.

[0161] (3) Leverage large language models to encode relevant private data and user questions.

[0162] If the user asks a question for the first time, the private data corresponding to the business question and the user's question are input into the large language model.

[0163] If it is not the first time for the user to ask a question, and the business question extracted from the user's question appears in the user's historical chat records, the user's question and the user's historical chat records are input into the large language model.

[0164] If this is not the first time that the user asks a question, and the business question extracted from the user's question does not appear in the user's historical chat records, the private data corresponding to the business question, the user's question, and the user's historical chat records are input into the large language model.

[0165] (4) The large language model gives answers to the user’s questions.

[0166] Next, we will continue to use the intelligent question-and-answer method in the above example to illustrate the implementation effect of this solution.

[0167] First, we take the question "Give me a fix for CVE-2021-2429" as an example and directly input it into the large language model ChatGLM. The answer we get is as follows:

[0168] Robot answer: CVE-2021-2429 is a vulnerability for Adobe Flash Player that can lead to remote code execution (RCE) attacks. To fix this vulnerability, you can take the following steps:

[0169] 1. Update Adobe Flash Player: Adobe has released a fix for CVE-2021-2429 and recommends that users update to the latest version immediately.

[0170] 2. Disable automatic updates for Flash Player: If you are using an operating system or application, you can disable the automatic update feature for Flash Player to avoid installing unreleased fixes.

[0171] 3. Install the Adobe Flash Player repair patch: Users can download the latest Flash Player repair patch from the Adobe official website and install it into the operating system or application.

[0172] 4. Check the Flash Player version: Users should check the Flash Player version and make sure it has been updated to a version that contains the fix for CVE-2021-2429.

[0173] 5. Update network settings: To prevent RCE attacks, users can update their network settings to ensure that they do not have unencrypted sensitive data such as usernames and passwords.

[0174] 6. Strengthen security settings: Users should strengthen other security settings, such as limiting network access to the public network, using strong passwords, and changing passwords regularly.

[0175] Please note that the patch that fixes CVE-2021-2429 may not be the only solution as the vulnerability may target multiple applications and operating systems. Therefore, users are advised to take the above steps to minimize the risk of being affected by the vulnerability.

[0176] It can be seen that the repair suggestion in the above answer is obviously wrong, and it does not recognize the business issue CVE-2021-2429.

[0177] Secondly, we take the question "Give me a fix for CVE-2021-2429" as an example, input it into the intelligent question-answering method in the above example, and get the following answer:

[0178] User: Give me a fix for CVE-2021-2429.

[0179] Robot answer: CVE-2021-2429 is a denial of service (DoS) vulnerability targeting MySQL Server 8.0.25 and earlier versions. The following are the repair suggestions and steps:

[0180] 1. Update to the latest version of MySQL Server.

[0181] 2. Limit network access to MySQL Server, using firewalls or other control measures.

[0182] 3. Monitor server performance and logs to detect potential attacks.

[0183] 4. Script to detect whether the vulnerability exists: There is currently no public PoC reference document.

[0184] Data confidentiality impact: Low (the vulnerability does not affect data confidentiality)

[0185] Data security impact: Medium (the attack may cause the HySQL server to be interrupted)

[0186] General advice: Update MySQL Server to the latest version as soon as possible and restrict network access to the MySQL server. At the same time, continue to pay attention to security updates and patch information released by Oracle.

[0187] 2. A security question-and-answer robot based on ChatGLM’s reading comprehension ability, allowing ChatGLM to answer user questions based on our own materials. It can be seen that for the same vulnerability, it can give a correct answer this time.

[0188] Finally, we asked a series of questions about the business issue of CVE-2021-2429, and the results of the questions and answers are shown below:

[0189] User: What type of vulnerability is CVE-2021-2429?

[0190] Robot answer: CVE-2021-2429 is a denial of service (DoS) vulnerability caused by a security vulnerability in versions 8.0.25 and earlier of the MySQL Server InnoDB component. An attacker can exploit this vulnerability to attack the MySQL server through network access via multiple protocols. The scope of this vulnerability includes the InnoDB component in Oracle MySQL Server, so it may pose a threat to all applications using MySQL Server.

[0191] User: Please help me list the components affected by this vulnerability.

[0192] Robot answer: The components affected by the CVE-2021-2429 vulnerability include:

[0193] 1. InnoDB storage engine: InnoDB is an important data storage engine in MySQL, used to store table data. This component is one of the most important components in MySQL Server because it determines the storage and query performance of the table.

[0194] 2.MySQL User Authentication Module: The MySQL User Authentication Module is used to verify the identity of the user and authorize the user to access the MySQL server. If this module is compromised, the attacker can bypass the authentication with credentials such as username and password and access the MySQL server.

[0195] 3.MySQL console: MySQL console is the user interface of MySQL server. If the console is attacked, the attacker can access HAySQL server and perform arbitrary operations.

[0196] 4MySQL connection pool: MySQL connection pool is used to manage clients connecting to the MySQL server. If this component is attacked, the attacker can create a large number of connections and keep them, thereby occupying the resources of the MySQL server and reducing performance.

[0197] 5.MySQL log: MySQL log is used to record the operation and events of the server. If the log files are stored in an unsecured location or the log records are tampered with, attackers can obtain sensitive information from them. These components are very important parts of the MySQL server. If they are attacked, they may pose a threat to all applications using MySQL.

[0198] User: Is there any reference fix patch?

[0199] Robot answer: Yes, there are some reference fix patches for your reference:

[0200] 1. MySQL 8.0.26 and above patches, which resolve the CVE-2021-2429 vulnerability.

[0201] 2. The fix patches provided by the Oracle official website are applicable to MySQL Server versions above 8.0.26.

[0202] You can download and install the corresponding patch to fix the vulnerability according to the MySQL version you are using and the update information provided by the Oracle official website. Please note that before installing the patch, please make sure that your MySQL server has been updated to the latest version and restrict network access to the MySQL server to ensure security.

[0203] User: Give me some suggestions for fixes.

[0204] Robot answer: The repair suggestions are as follows:

[0205] 1. Update to the latest version of MySQL Server: If your MySQL server is using version 8.0.25 or lower, please update to version 8.0.26 or higher as soon as possible. Updating to the latest version ensures that you use the latest fixes and resolves known vulnerabilities.

[0206] 2. Limit network access to the MySQL server: Use firewalls or other control measures to limit network access to the MySQL server to ensure the security of the server.

[0207] 3. Monitor server performance and logs: Regularly monitor server performance and logs to ensure that there are no undiscovered vulnerabilities or attacks on the MySQL server.

[0208] 4. Perform security audits: Perform security audits on the MySQL server regularly to identify possible issues and take necessary steps to fix them.

[0209] 5. Use security tools: Use security tools, such as vulnerability scanners, to detect and fix vulnerabilities on the MySQL server.

[0210] Please note that fixing a vulnerability does not guarantee complete prevention of an attack. Before taking any action, make sure you understand the risks and take appropriate security measures to protect your data and applications.

[0211] User: Very useful, thank you.

[0212] Robot answer: You're welcome, if you have any other questions or need help, feel free to ask me. I will do my best to answer your questions.

[0213] An embodiment of the present invention further provides a storage medium on which an intelligent question-answering program is stored. When the intelligent question-answering program is executed by a processor, any embodiment of the intelligent question-answering method described above is implemented.

[0214] Reference Figure 7, the embodiment of the present invention also provides an intelligent question-answering device, the intelligent question-answering device comprising: a query module 10, a processing module 20, and a large language model interface module 201;

[0215] The query module 10 is used to search for private data related to the user's question from a preset private document according to the received user's question.

[0216] In this embodiment, the user's question can be asked by inputting text on the display screen, or by voice questioning.

[0217] The preset private documents are stored in the memory, and the private documents include private data within the enterprise or group. In practice, the private documents include documents related to vulnerability information, documents related to employee handbooks, or documents related to technical information. In this embodiment, the preset private documents can be stored in the memory in the form of a database, or can be directly stored in the memory in the form of a document.

[0218] The private data corresponding to the business problem in the preset private document includes any one or any combination of the following: business problem name, business problem type, business problem reference document, business problem reference link. For example, if the private document includes vulnerability information related documents, then the business problem is a vulnerability-related problem, and the private data corresponding to the vulnerability-related problem may include vulnerability name, vulnerability type, vulnerability reference document, vulnerability reference link, vulnerability repair suggestion, etc.

[0219] Realistically, when searching for private data related to a user's question, the query module 10 may use a search algorithm to search; if semantic search is required, a neural network model may be trained and the trained neural network model may be used to search.

[0220] Realistically, the private document preset in this embodiment supports online updating of data.

[0221] The processing module 20 is used to input the private data related to the user's question and the user's question into the large language model interface module 201 to obtain an answer.

[0222] The large language model used in this embodiment is an open source model and does not require a training model. The large language model interface module 201 can be a ChatGPT interface module. If for data security and privacy protection considerations, you can also choose to deploy an open source ChatGPT-like large language model yourself, such as using an open source ChatGLM interface module. The large language model chooses a model that supports a longer input length as much as possible to support the situation where the user inputs a longer question.

[0223] After the large language model understands the user's question, it will refer to the input private data related to the user's question and answer it in combination with the user's question. Therefore, when answering questions related to private data, the accuracy of the answers given is relatively high. And because the large language model has strong semantic understanding and creative capabilities, it can generate continuous text similar to human conversations. Therefore, the intelligent question-answering device in the present invention can achieve fluent conversations with people and accurately answer questions related to private data.

[0224] Compared with traditional machine question-answering devices, intelligent question-answering devices based on large language models do not require training and have better generalization and migration capabilities. In other words, the large language model interface module 201 has better adaptability to fresh samples. Even if the data in the preset private documents increases, decreases or is updated, there is no need to retrain the model, which can greatly reduce costs.

[0225] Other embodiments or specific implementations of the intelligent question-and-answer device of the present invention may refer to the above-mentioned method embodiments and will not be described in detail here.

[0226] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0227] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0228] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory image (ROM) / random access memory (RAM), a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0229] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

[0230] The present invention discloses A1, an intelligent question-answering method, the intelligent question-answering method comprising:

[0231] Searching for private data related to the user's question from a preset private document according to the received user's question;

[0232] The private data related to the user's question and the user's question are input into a large language model to obtain an answer.

[0233] A2. The intelligent question-answering method as described in A1, wherein the preset private document includes a business question and private data corresponding to the business question;

[0234] The step of searching for private data related to the user's question from a preset private document according to the received user's question includes:

[0235] Extracting business questions from the questions received from the user;

[0236] According to the business problem, private data corresponding to the business problem is searched from the preset private document.

[0237] A3. The intelligent question-answering method as described in A2, before searching the preset private document for private data corresponding to the business question according to the business question, further comprising:

[0238] Determine whether the user is asking a question for the first time;

[0239] If the user is asking a question for the first time, the step of searching the preset private document for private data corresponding to the business question is performed based on the business question.

[0240] A4. The intelligent question-answering method as described in A3, after determining whether the user asks a question for the first time, further comprising:

[0241] If it is not the first time for the user to ask a question, determining whether the business question extracted from the user's question appears in the user's historical chat record;

[0242] If the business question appears in the historical chat records of the user, the user's question and the historical chat records of the user are input into the large language model to obtain an answer.

[0243] A5. The intelligent question-answering method as described in A4, after determining whether the business question extracted from the user's question appears in the user's historical chat record, further comprises:

[0244] If the business question does not appear in the historical chat records of the user, searching for private data corresponding to the business question from the preset private document according to the business question;

[0245] The private data corresponding to the business question, the user's question and the user's historical chat records are input into a large language model to obtain an answer.

[0246] A6. In the intelligent question-answering method as described in A4 or A5, the historical chat records of the user and the questions received from the user belong to the same round of dialogue.

[0247] A7. In the intelligent question-answering method as described in any one of A2 to A5, the private data corresponding to the business question in the preset private document includes any one or any combination of the following: business question name, business question type, business question reference document, and business question reference link.

[0248] A8. The intelligent question-answering method as described in A1, before searching for private data related to the user's question from a preset private document according to the received user's question, further comprising:

[0249] Determine whether the received user's question is a first-category business question;

[0250] If the received user question is a first type of business question, the step of searching for private data related to the user's question from a preset private document based on the received user question is performed, wherein the preset private document includes relevant data of the first type of business question.

[0251] A9. In the intelligent question-answering method described in A8, determining whether the received user's question is a first-category business question comprises:

[0252] Determine whether the received user's question is a first-category business question based on a pre-trained classification model.

[0253] A10. The intelligent question-answering method as described in A8, after determining whether the received user's question is a first-category business question, further comprising:

[0254] If the received user's question is not a first-category business question, the user is prompted to input a question related to the first-category business question.

[0255] A11. In the intelligent question-answering method as described in A1, the private document includes a document related to vulnerability information.

[0256] The present invention also discloses B12, an intelligent question and answer device, which includes: a memory, a processor, and an intelligent question and answer program stored in the memory and executable on the processor, and when the intelligent question and answer program is executed by the processor, the intelligent question and answer method as described above is implemented.

[0257] The present invention also discloses C13, a storage medium, on which an intelligent question-answering program is stored, and when the intelligent question-answering program is executed by a processor, the intelligent question-answering method as described above is implemented.

[0258] The present invention also discloses D14, an intelligent question-answering device, comprising: a query module, a processing module, and a large language model interface module;

[0259] The query module is used to search for private data related to the user's question from a preset private document according to the received user's question;

[0260] The processing module is used to input the private data related to the user's question and the user's question into the large language model interface module to obtain an answer.

[0261] D15. The intelligent question-answering device as described in D14, wherein the preset private document includes a business question and private data corresponding to the business question;

[0262] The query module is specifically used to extract business questions from the received user questions, and search for private data corresponding to the business questions from the preset private documents according to the business questions.

[0263] D16. In the intelligent question-and-answer device as described in D15, the query module is specifically used to determine whether the user is asking a question for the first time before searching for private data corresponding to the business question from the preset private document based on the business question; if the user is asking a question for the first time, then executing the step of searching for private data corresponding to the business question from the preset private document based on the business question.

[0264] D17. The intelligent question-answering device as described in D16, wherein the query module is further used to determine whether the business question extracted from the user's question appears in the user's historical chat record after determining whether the user asks the question for the first time, if the user does not ask the question for the first time;

[0265] The processing module is also used to input the user's question and the user's historical chat records into the large language model interface module to obtain an answer if the business question appears in the user's historical chat records.

[0266] D18. The intelligent question-answering device as described in D17, wherein the query module is further used for searching private data corresponding to the business question from the preset private document according to the business question after determining whether the business question extracted from the user's question appears in the user's historical chat record if the business question does not appear in the user's historical chat record;

[0267] The processing module is also used to input the private data corresponding to the business question, the user's question and the user's historical chat records into the large language model interface module to obtain an answer.

[0268] D19. In the intelligent question-answering device as described in D17 or D18, the historical chat records of the user and the questions received from the user belong to the same round of dialogue.

[0269] D20. In an intelligent question-and-answer device as described in any one of D15 to D18, the private data corresponding to the business question in the preset private document includes any one or any combination of the following: business question name, business question type, business question reference document, and business question reference link.

[0270] D21. In the intelligent question-and-answer device as described in D14, the query module is specifically used to determine whether the received user question is a first-category business question before searching for private data related to the user's question from a preset private document based on the received user's question; if the received user question is a first-category business question, then executing the step of searching for private data related to the user's question from a preset private document based on the received user's question, wherein the preset private document includes relevant data of the first-category business question.

[0271] D22. In the intelligent question-answering device as described in D21, the query module is specifically used to determine whether the received user question is a first-category business question based on a pre-trained classification model.

[0272] D23. In the intelligent question-and-answer device as described in D21, the processing module is also used to prompt the user to enter a question related to the first category of business questions if the received user question is not a first category of business question after determining whether the received user question is a first category of business question.

[0273] D24. In the intelligent question-and-answer device as described in D14, the private document includes a document related to vulnerability information.

Claims

1. An intelligent question-answering method, characterized in that: include: Searching for private data related to the user's question from a preset private document according to the received user's question; The private data related to the user's question and the user's question are input into a large language model to obtain an answer.

2. The intelligent question-answering method as claimed in claim 1, characterized in that: The preset private document includes the business problem and the private data corresponding to the business problem; The step of searching for private data related to the user's question from a preset private document according to the received user's question includes: Extracting business questions from the questions received from the user; According to the business problem, private data corresponding to the business problem is searched from the preset private document.

3. The intelligent question-answering method as claimed in claim 2, characterized in that: Before searching the preset private document for private data corresponding to the business problem according to the business problem, the method further includes: Determine whether the user is asking a question for the first time; If the user is asking a question for the first time, the step of searching the preset private document for private data corresponding to the business question is performed based on the business question.

4. The intelligent question-answering method as claimed in claim 3, characterized in that: After determining whether the user asks a question for the first time, the method further includes: If it is not the first time for the user to ask a question, determining whether the business question extracted from the user's question appears in the user's historical chat record; If the business question appears in the historical chat records of the user, the user's question and the historical chat records of the user are input into the large language model to obtain an answer.

5. The intelligent question-answering method as claimed in claim 4, characterized in that: After determining whether the business question extracted from the user's question appears in the user's historical chat record, the method further includes: If the business question does not appear in the historical chat records of the user, searching for private data corresponding to the business question from the preset private document according to the business question; The private data corresponding to the business question, the user's question and the user's historical chat records are input into a large language model to obtain an answer.

6. The intelligent question-answering method according to any one of claims 2 to 5, characterized in that: The private data corresponding to the business problem in the preset private document includes any one or any combination of the following: business problem name, business problem type, business problem reference document, and business problem reference link.

7. The intelligent question-answering method as claimed in claim 1, characterized in that: Before searching for private data related to the user's question from a preset private document according to the received user's question, the method further includes: Determine whether the received user's question is a first-category business question; If the received user question is a first type of business question, the step of searching for private data related to the user's question from a preset private document based on the received user question is performed, wherein the preset private document includes relevant data of the first type of business question.

8. An intelligent question-answering device, characterized in that: The intelligent question-answering device comprises: a memory, a processor, and an intelligent question-answering program stored in the memory and executable on the processor. When the intelligent question-answering program is executed by the processor, the intelligent question-answering method according to any one of claims 1 to 7 is implemented.

9. A storage medium, characterized in that: The storage medium stores an intelligent question-answering program, and when the intelligent question-answering program is executed by the processor, the intelligent question-answering method according to any one of claims 1 to 7 is implemented.

10. An intelligent question-answering device, characterized in that: include: Query module, processing module, and large language model interface module; The query module is used to search for private data related to the user's question from a preset private document according to the received user's question; The processing module is used to input the private data related to the user's question and the user's question into the large language model interface module to obtain an answer.

Images