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

Abstract

The invention provides an intelligent question and answer method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence, in particular to the technical field of large models, intelligent question and answer and the like. According to the historical question and answer information, matching a target dialogue skill from a pre-constructed dialogue skill library; and in response to the matched target dialogue skill, generating the target reply information according to the target dialogue skill. Therefore, the target reply information is generated according to the target dialogue skill, the professionality and persuasion of the target reply information are improved, the user requirements are better met, the accuracy of the target reply information is improved in an intelligent question and answer scene, and the user experience is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, specifically to the technical fields of large models and intelligent question and answer, and in particular to an intelligent question and answer method, device, electronic device and storage medium. Background Art

[0002] In the existing intelligent marketing dialogue models, although the generation capability of the Large Language Model (LLM) can simulate the communication methods of sales personnel to a certain extent, it still has significant deficiencies in terms of professional sales skills and complex marketing logic. Especially when user demands are complex and business scenarios are diverse, the generated dialogue content often appears mechanical and lacks depth and pertinence. Summary of the invention

[0003] The present disclosure proposes an intelligent question-answering method, device, electronic device, storage medium, and computer program product.

[0004] According to the first aspect of the present disclosure, an intelligent question and answer method is proposed, including: obtaining historical question and answer information; matching target dialogue skills from a pre-built dialogue skill library based on the historical question and answer information; in response to matching the target dialogue skills, generating the target reply information based on the target dialogue skills.

[0005] According to the second aspect of the present disclosure, an intelligent question and answer device is proposed, including: an acquisition module, used to acquire historical question and answer information, and generate multiple candidate reply information according to the historical question and answer information through a generative large model; a prediction module, used to predict the conversion rate of the multiple candidate reply information, and acquire the conversion rate information of the candidate reply information; an output module, used to determine the target reply information for output according to the conversion rate information of the candidate reply information.

[0006] According to the third aspect of the present disclosure, an electronic device is proposed, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the intelligent question and answer method proposed in the first aspect above.

[0007] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is proposed, wherein the computer instructions are used to enable the computer to execute the intelligent question-answering method proposed in the first aspect above.

[0008] According to a fifth aspect of the present disclosure, a computer program product is proposed, including a computer program, which implements the intelligent question-answering method proposed in the first aspect when executed by a processor.

[0009] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0011] Figure 1 A flowchart of an intelligent question-answering method according to an embodiment of the present disclosure;

[0012] Figure 2 A flowchart of an intelligent question-answering method according to an embodiment of the present disclosure;

[0013] Figure 3 A flowchart of a process for constructing a dialogue skill library according to an embodiment of the present disclosure;

[0014] Figure 4 A flowchart of a process for constructing a question list knowledge base according to an embodiment of the present disclosure;

[0015] FIG5( a ) is a schematic diagram of original corpus data according to an embodiment of the present disclosure;

[0016] FIG5( b ) is a schematic diagram of corpus data according to an embodiment of the present disclosure;

[0017] Figure 6 It is a structural diagram of an intelligent question-answering device according to an embodiment of the present disclosure;

[0018] Figure 7 A schematic block diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0019] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0020] Artificial Intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence.

[0021] A large model refers to a machine learning model with a large parameter scale and complexity. It requires a large amount of computing resources and storage space to train and store, and often requires distributed computing and special hardware acceleration technology. Large models have stronger generalization and expression capabilities.

[0022] Intelligent question answering refers to an automatic question answering system implemented using artificial intelligence technology. It enables computers to understand and answer questions raised by users by analyzing, understanding and mining large amounts of text data. Intelligent question answering systems usually include key technologies such as natural language processing, knowledge representation, reasoning and answer generation.

[0023] Figure 1 FIG. 1 is a flow chart of an intelligent question-answering method according to an embodiment of the present disclosure. Figure 1 As shown, the method includes:

[0024] S101, obtaining historical question and answer information.

[0025] It should be noted that the execution subject of the intelligent question-answering method of the embodiment of the present disclosure may be a hardware device with data intelligent question-answering capabilities and / or the necessary software required to drive the hardware device to work. Optionally, the execution subject may include a workstation, a server, a computer, a user terminal and other intelligent devices. Among them, the user terminal includes but is not limited to a mobile phone, a computer, an intelligent voice interaction device, a smart home appliance, a car terminal, etc.

[0026] Among them, the historical question and answer information includes the inquiry content and answer content corresponding to different question and answer roles.

[0027] For example, the reply content of the merchant's customer service and the inquiry content of the target object can be obtained as historical question and answer information.

[0028] S102, matching target dialogue skills from a pre-built dialogue skill library based on historical question and answer information.

[0029] In an embodiment of the present disclosure, corpus data may be acquired, semantic extraction may be performed on the corpus data, key intent information of the user may be extracted from the corpus data, user key information of the user may be acquired, and a first vector of the key intent information and the user key information may be acquired, customer service speech features may be extracted from the corpus data, and a second vector may be generated based on the extracted customer service speech features, the first vector and the second vector may be associated to generate an associated vector, and a conversation skills library may be generated based on the associated vector.

[0030] It should be noted that the present disclosure does not limit the specific method of matching the target dialogue skills from a pre-built dialogue skill library based on historical question and answer information, and it can be selected according to actual conditions.

[0031] Optionally, the target dialogue skill may be matched from a pre-built dialogue skill library by using Retrieval Augmented Generation (RAG) technology.

[0032] S103, in response to matching the target dialogue skill, generating target reply information according to the target dialogue skill.

[0033] In the disclosed embodiment, after matching the target dialogue skills, the target dialogue skills can be combined with the inquiry information input by the user to generate more accurate and professional target reply information. By generating the target reply information, the interaction satisfaction and trust between the user and the customer service can be effectively improved.

[0034] For example, user A searches for the merchant keyword "AA Decoration" on a mobile phone and finds the merchant page "AA Decoration" with the "agent" label in the search results. He clicks to enter the landing page of the merchant's agent and accesses the chat box to achieve real-time interaction with the agent. Based on historical question and answer information, for example: User A asks "What decoration packages do you recommend?" In response to matching the target dialogue skills, based on the target dialogue skills, the agent can generate target reply information containing sales logic, "The economical package we provide for new users includes design, construction and materials. Do you need me to explain it to you in detail?" Products or services can be introduced to users based on more professional language, thereby enhancing user trust and satisfaction.

[0035] The intelligent question and answer method proposed in the present disclosure obtains historical question and answer information, matches the target dialogue skill from a pre-built dialogue skill library based on the historical question and answer information, and generates target reply information based on the target dialogue skill in response to matching the target dialogue skill. Therefore, the present disclosure generates the target reply information based on the target dialogue skill, thereby improving the professionalism and persuasiveness of the target reply information, better meeting the needs of users, improving the accuracy of the target reply information in the intelligent question and answer scenario, and improving the user experience.

[0036] Figure 2 It is a flowchart of the intelligent question-answering method according to the second embodiment of the present disclosure.

[0037] like Figure 2 As shown, in Figure 2 Based on the illustrated embodiment, the intelligent question-answering method of the embodiment of the present disclosure may specifically include the following steps:

[0038] S201, obtaining historical question and answer information.

[0039] For the relevant contents of step S201, please refer to the above embodiment, which will not be described again here.

[0040] Optionally, step S102 in the above embodiment, “matching target dialogue skills from a pre-built dialogue skill library according to historical question and answer information” may specifically include the following steps S202-S204.

[0041] S202: Generate a third vector based on historical question and answer information.

[0042] In an embodiment of the present disclosure, after the historical question and answer information is acquired, the historical question and answer information may be vectorized to generate a third vector.

[0043] S203: Based on the third vector, perform an approximate nearest neighbor ANN vector search in the dialogue skill library to obtain a candidate association vector that is closest to the third vector.

[0044] Among them, Approximate Nearest Neighbor (ANN) vector retrieval is an efficient similar vector search method that can quickly locate similar objects of the target vector in the global vector space. Compared with the traditional exact nearest neighbor search (NNS), Approximate Nearest Neighbor ANN vector retrieval reduces the computational cost while ensuring good retrieval results in large-scale vector retrieval scenarios.

[0045] For example, the third vector may be used as a query to perform an approximate nearest neighbor ANN vector search in the dialogue skill library to obtain a candidate association vector that is closest to the third vector.

[0046] As a possible way to achieve this, Figure 3 As shown, the process of building a dialogue skill library may include the following steps:

[0047] S301, performing semantic extraction on the corpus data to extract the user's key intention information from the corpus data.

[0048] In an embodiment of the present disclosure, user corpus data in the corpus data is determined, and first interference information is removed from the user corpus data to obtain user target corpus data, and key information and semantics are extracted from the user target corpus data to obtain key intention information of the user.

[0049] It should be noted that in order to ensure that key information in the user corpus data can be effectively extracted and stored in a structured manner and to provide data support for the subsequent process, the first interference information can be removed from the user corpus data to obtain the user target corpus data.

[0050] For example, the first interference information includes, but is not limited to, the forwarding consultation information in the user corpus data, the conversation information of the sales confirming the contact information, the service information outside the conversation provided by the sales, the ambiguous information or the unclear information, etc.

[0051] In the disclosed embodiment, after obtaining the user target corpus data, key information and semantics can be extracted from the user target corpus data to obtain the user's key intention information.

[0052] For example, in the education industry, the user's key intent information includes the subject to be tutored, tutoring schedule, tutoring experience, and the conditions for attending school.

[0053] S302, obtaining key user information of the user, and obtaining key intention information and a first vector of the key user information.

[0054] For example, in the education industry, key user information includes the user's name, grade, region, school, elective subjects, etc.

[0055] In an embodiment of the present disclosure, after acquiring key intent information and user key information, the key intent information and user key information may be vectorized to obtain a first vector.

[0056] S303, extracting customer service speech features from the corpus data, and generating a second vector according to the extracted customer service speech features.

[0057] In an embodiment of the present disclosure, customer service corpus data in the corpus data is determined, and the second interference information in the customer service corpus data is masked to obtain customer service target corpus data, and customer service speech features are extracted from the customer service target corpus data.

[0058] It should be noted that the second interference information includes but is not limited to information such as time, price and specific location. The second interference information may introduce noise and affect the coherence and accuracy of the target reply information of the dialogue. Therefore, the second interference information in the customer service corpus data can be masked, and the second interference information can be masked to obtain the customer service target corpus data. The customer service speech features are extracted from the customer service target corpus data, which can ensure the purity of the customer service target corpus data while demonstrating the skills in customer service communication to a greater extent, providing more reliable support for the subsequent generation of target reply information.

[0059] S304, associating the first vector and the second vector to generate an associated vector, and generating a dialogue skill library based on the associated vector.

[0060] Optionally, the associated vectors can be stored in a vector database (VDB) to generate a dialogue skills library that supports ANN vector retrieval. The vector database adopts a daily update mechanism to ensure that the generated question and answer information responses are consistent with the current business status, ensuring the real-time and practical nature of the dialogue skills library, laying a high-quality data foundation for the subsequent generation of target reply information based on the target dialogue skills, and achieving efficient and orderly advancement of business processes.

[0061] S204, determining a target dialogue skill based on the candidate association vectors.

[0062] In the disclosed embodiment, the confidence of the candidate association vector may be obtained, the candidate association vector may be filtered according to the confidence to obtain the target association vector, and the dialogue skill corresponding to the target association vector may be obtained as the target dialogue skill.

[0063] Optionally, the similarity between each candidate association vector and the third vector may be used as the confidence of the candidate association vector.

[0064] It should be noted that when the confidence level is low, it often leads to greater uncertainty and deviation. In order to ensure that the target response information finally generated is more reliable and accurate, after obtaining the confidence level of the candidate association vector, the confidence level of the candidate association vector can be sorted in descending order, and the candidate association vectors in the first K positions are selected from the sorting results as the target association vector, where K is a positive integer, for example: K is 5.

[0065] In the disclosed embodiment, after the target association vector is obtained, the dialogue skill corresponding to the target association vector may be used as the target dialogue skill.

[0066] S205, in response to matching the target dialogue skill, generating target reply information according to the target dialogue skill.

[0067] For the relevant contents of step S205, please refer to the above embodiment, which will not be repeated here.

[0068] S206, in response to failure to match the target dialogue skill, determining auxiliary response information determined by the large model based on historical question and answer information.

[0069] In the disclosed embodiment, three-segment information, namely, response information, answer information, and guidance information, may be acquired, and auxiliary response information may be generated based on the three-segment information.

[0070] It should be noted that auxiliary response information should include polite responses and professional answers to users, and should also continue to guide and inspire users to speak, deeply explore user needs by asking questions, and apply conversation skills in a targeted manner to continue the conversation process and achieve the ultimate goal.

[0071] S207, matching the target question from a pre-built question list knowledge base based on the historical question and answer information.

[0072] In the disclosed embodiment, a target business tag of historical question and answer information is obtained, a target question list matching the target business tag is obtained from a question list knowledge base, and a target question is determined from the target question list.

[0073] It should be noted that the present disclosure does not limit the specific method for obtaining the target business label of the historical question and answer information, and it can be selected according to the actual situation.

[0074] Optionally, the last round of question and answer information in the historical question and answer information may be determined, key information of the last round of question and answer information may be extracted based on the KG model, and a target business label may be obtained based on the key information.

[0075] Optionally, in response to the target service tag not being determined based on the critical information, a historical service tag of the historical question and answer information is acquired as the target service tag.

[0076] It should be noted that since the user's key intent information may change at any time, the KG model will maintain the historical business label variables of the historical question and answer information. If a new business label is identified in this round of question and answer information, the historical business label variable can be modified to the latest business label. If no new business label is identified in this round of question and answer information, it is considered that the user's key intent information has not changed, and the historical business label variable remains unchanged.

[0077] In the disclosed embodiment, a second prompt for question screening can be generated according to question screening rules, and the second prompt can be input into the big model. The big model can then screen the questions in the target question list based on the second prompt to obtain the target question.

[0078] It should be noted that the second prompt for question screening is generated according to the question screening rules, that is, by tuning the prompt, the large model screens the questions in the target question list to obtain suitable target questions that are in line with the current dialogue process.

[0079] For example, in order to avoid asking repeated questions, ensure a good user experience, and explore user needs from the shallower to the deeper, the question screening rules can be: (1) Information that the user has already mentioned cannot be asked repeatedly; (2) Content that the customer service has already asked cannot be asked repeatedly; (3) On the basis of meeting the conditions of (1) and (2), questions are asked in the order of the target questions.

[0080] As a possible way to achieve this, Figure 4As shown, the process of constructing the question list knowledge base may include the following steps:

[0081] S401, acquiring corpus data and determining a business tag corresponding to the corpus data.

[0082] The corpus data may be sales corpus data.

[0083] Optionally, if the original form of the corpus data is audio, the corpus data in audio form may be converted into corpus data in text form.

[0084] For example, for a piece of audio corpus data A, a piece of text corpus data A' can be obtained by converting the audio corpus data A.

[0085] Optionally, you can obtain public courses and books, which often include specific language, personnel communication skills, industry insights, and professional suggestions for user demands, which can effectively improve users' interactive satisfaction and trust. The content in courses and books can be converted to obtain text corpus data.

[0086] It should be noted that during the conversion process, there are often various problems such as typos, irrelevant modal particles, and incoherent dialogue forms. Therefore, the converted text corpus data needs to be cleaned to obtain high-quality corpus data.

[0087] Optionally, the big model's powerful semantic understanding and content generation capabilities can be used to generate prompt words for cleaning the corpus data, and the prompt words for cleaning the corpus data can be input into the big model. The big model can then clean the corpus data based on the prompt words for cleaning the corpus data to obtain high-quality corpus data.

[0088] For example, the cleaning objectives include at least: (1) merging multiple consecutive speeches of the same dialogue character into a single one, and maintaining the independence and integrity of the speeches between the characters to avoid confusing the content of the dialogue between the two parties; (2) eliminating irrelevant and redundant modal particles, and correcting errors and ambiguities that occur during audio dumping based on context information; (3) retaining dialogue content related to business consultation, and removing irrelevant content such as messy information, procedural content, repeated confirmations, and meaningless modal particles; (4) standardizing the arrangement format of the dialogue text, and displaying the speech content of each character alternately line by line to form a complete dialogue.

[0089] For example, if the original corpus data in the converted text form is as shown in Figure 5(a), the corpus data is cleaned by the large model based on the prompt word prompt for corpus data cleaning, and high-quality corpus data as shown in Figure 5(b) can be obtained.

[0090] It should be noted that since the KG model can effectively identify entities, categories, intents, attributes and regions of short input texts in various industries, and has good classification capabilities for similar inputs, in the disclosed embodiment, the business label corresponding to the corpus data can be determined through the Knowledge Graph (KG) model. Since the input of the KG model is in the form of a short query word query, the corpus data can be compressed, that is, the corpus data is compressed into a short query form to obtain a user intent text, wherein the user intent text can be in the form of: user subject + purpose + consulting content, and auxiliary information for identification is extracted from the corpus data, wherein the auxiliary information can be merchant information, and the overall intent of the corpus data is extracted based on the big model, and the merchant information is used to help the big model better understand the corpus data, so as to realize the recognition of the user intent text based on the auxiliary information, obtain the intent key field (intent_list) and the category key field (category_list), and generate the business label corresponding to the corpus data according to the intent key field and the category key field.

[0091] Optionally, if the user intent text fails to show a clear intent, resulting in the intent key fields being unclear and inaccurate, the business tag corresponding to the corpus data is empty, and the corresponding corpus data can be discarded.

[0092] For example, if the user intent text is not "the user wants to consult about tutoring for freshman mathematics", and the auxiliary information is "the user entity is freshman mathematics, the purpose type is course, the consultation content (attribute) is online teaching (training method), course (entity label and freshman data (standard name)", the user intent text is identified based on the auxiliary information, and the intent key field intent_list is obtained as "K12-training period-ask the training institution-ask the institution" and the category key field category_list is obtained as "Education and Training>K12 Education>High School Tutoring>High School Mathematics>Fourth Grade Mathematics". The intent key field and the category key field are processed to generate the business label corresponding to the corpus data as "Education and Training-K12 Education-High School Tutoring-High School Mathematics-Fourth Grade Mathematics / K12-Training Period-Ask the Training Institution-Ask the Institution".

[0093] Optionally, if the intent key field and the category key field contain multiple results, the intent key field and the category key field may be processed according to pre-set rules to generate business tags corresponding to the corpus data.

[0094] In the embodiment of the present disclosure, by acquiring the business tag corresponding to the corpus data, the corpus data can be mapped to a limited business tag set, and each piece of corpus data has a corresponding business tag.

[0095] S402, obtaining a list of sub-questions of the corpus data.

[0096] In the disclosed embodiment, in order to enable the big model to focus on extracting questions asked by customer service without being disturbed by user speeches, the user corpus data in the corpus data can be masked to obtain masked corpus data, and the first prompt word prompt for question extraction is obtained. The first prompt and the masked corpus data are input into the big model, and the big model extracts questions from the masked corpus data under the constraint of the first prompt to generate a list of sub-questions of the corpus data.

[0097] Optionally, the large model may perform question extraction on the masked corpus data under the constraints of the relevance and quantity of the questions in the first prompt, and generate a list of sub-questions of the corpus data.

[0098] S403: Merge the sub-question lists of the corpus data associated with the same business tag to generate a question list corresponding to the business tag.

[0099] In the disclosed embodiment, a list of sub-questions of corpus data associated with the same business tag may be obtained, correlation extraction may be performed on the list of sub-questions, and logic optimization may be performed on the list of sub-questions according to the correlation to generate a list of questions corresponding to the business tag.

[0100] Optionally, a list of sub-questions of corpus data associated with the same business tag can be obtained, and the large model can extract the relevance of the sub-question list through Prompt tuning to obtain the relevance between the sub-question lists, and logically optimize the sub-question list based on the relevance, that is, it can reflect the gradual process of gradual development of the business to generate a list of questions corresponding to the business tag.

[0101] S404: Generate a question list knowledge base based on the question list corresponding to the business tag.

[0102] Each question list corresponds to a business tag, and the question lists corresponding to the business tags can be summarized to generate a question list knowledge base.

[0103] S208, in response to matching the target question, generating target reply information according to the target question and the auxiliary response information.

[0104] In the disclosed embodiment, after matching the target question, a third prompt for question optimization is generated, and the third prompt is input into the big model, and the big model polishes the target question and auxiliary response information based on the third prompt. Since the big model only needs to focus on the requirements of language style, the big model polishes the target question and auxiliary response information based on the third prompt, so as to generate a target response information that is more content-rich, friendly and close to user needs.

[0105] For example, user A searches for the merchant keyword "AA Decoration" on a mobile phone and finds the merchant page "AA Decoration" with the "agent" label in the search results. He clicks to enter the landing page of the merchant's agent and accesses the chat box to achieve real-time interaction with the agent. Based on historical question and answer information, for example: User A asks, "How is the decoration cost calculated?" The target question is matched from the pre-built question list knowledge base, for example: "Do you want to know the cost of decorating the living room or the budget for decorating the whole house?" Precisely guide questions based on the matched questions to stimulate users to further express their needs and reduce the occurrence of awkward or ineffective conversations.

[0106] S209, in response to not matching the target question, generating target reply information based on the auxiliary response information.

[0107] It should be noted that in order to maintain a good user experience and business continuity, when the target dialogue skills and the target questions are not matched, the target reply information is generated based on the auxiliary response information. This can be understood as a fallback solution for the intelligent question and answer method proposed in the embodiment of the present disclosure.

[0108] The intelligent question-answering method proposed in the present disclosure obtains historical question-answering information, generates a third vector according to the historical question-answering information, performs an approximate nearest neighbor ANN vector search in a dialogue skill library according to the third vector, obtains a candidate associated vector closest to the third vector, determines a target dialogue skill according to the candidate associated vector, generates target reply information according to the target dialogue skill in response to matching the target dialogue skill, determines auxiliary reply information determined by a large model based on the historical question-answering information in response to not matching the target dialogue skill, matches a target question from a pre-constructed question list knowledge base according to the historical question-answering information, generates target reply information according to the target question and the auxiliary reply information in response to matching the target question, and generates a target reply information according to the auxiliary reply information in response to not matching the target question. Information, therefore, the present disclosure realizes efficient retrieval of target dialogue skills and target questions through RAG technology, generates target reply information according to the target dialogue skills, improves the professionalism and persuasiveness of the target reply information, and is more in line with user needs. In the intelligent question-and-answer scenario, the accuracy of the target reply information is improved, the user experience is improved, and mechanical replies are reduced, which is conducive to stimulating users to participate in active dialogues. The target reply information is generated according to the target question and auxiliary reply information, and the target question for user demands can be accurately generated, which improves the dialogue ability in complex business scenarios and optimizes the user experience. The target reply information is generated according to the auxiliary reply information, which ensures the continuity of the target reply information. The target reply information is collaboratively generated through the above three methods to promote the efficient development of business processes.

[0109] It should be noted that the information (including but not limited to question and answer information, user device information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in this disclosure are all authorized by the user or fully authorized by all parties, and the source, use and processing of the relevant data comply with relevant laws, regulations and standards, and do not violate public order and good morals.

[0110] According to an embodiment of the present disclosure, the present disclosure also provides an intelligent question-answering device for implementing the above-mentioned intelligent question-answering method.

[0111] Figure 6 is a block diagram of an intelligent question-answering device according to an embodiment of the present disclosure.

[0112] like Figure 6 As shown, the intelligent question-answering device 600 includes: an acquisition module 601, a matching module 602 and a reply module 603.

[0113] Acquisition module 601, used to acquire historical question and answer information;

[0114] A matching module 602, configured to match a target dialogue skill from a pre-built dialogue skill library according to the historical question and answer information;

[0115] The reply module 603 is used to generate the target reply information according to the target dialogue skill in response to matching the target dialogue skill.

[0116] In one embodiment of the present disclosure, the device 600 is also used to: in response to not matching the target dialogue skill, determine the auxiliary response information determined by the large model based on the historical question and answer information; match the target question from a pre-built question list knowledge base based on the historical question and answer information; in response to matching the target question, generate target reply information based on the target question and the auxiliary response information; in response to not matching the target question, generate target reply information based on the auxiliary response information.

[0117] In one embodiment of the present disclosure, the device 600 is also used for: the process of constructing the question list knowledge base includes: obtaining corpus data and determining the business tag corresponding to the corpus data; obtaining a sub-question list of the corpus data; merging the sub-question lists of the corpus data associated with the same business tag to generate a question list corresponding to the business tag; and generating the question list knowledge base according to the question list corresponding to the business tag.

[0118] In one embodiment of the present disclosure, the device 600 is also used to: perform masking on the user corpus data in the corpus data to obtain masked corpus data; obtain a first prompt word prompt for question extraction, and input the first prompt and the masked corpus data into a large model; and have the large model perform question extraction on the masked corpus data under the constraint of the first prompt to generate a sub-question list of the corpus data.

[0119] In one embodiment of the present disclosure, the device 600 is further used to: obtain a list of sub-questions of the corpus data associated with the same business tag; extract relevance from the list of sub-questions, and logically optimize the list of sub-questions based on the relevance to generate a list of questions corresponding to the business tag.

[0120] In one embodiment of the present disclosure, the device 600 is also used to: compress the corpus data to obtain user intention text; extract auxiliary information for intent identification from the corpus data; identify the user intention text based on the auxiliary information to obtain intent key fields and category key fields; and generate business tags corresponding to the corpus data based on the intent key fields and category key fields.

[0121] In one embodiment of the present disclosure, the device 600 is also used for: the process of constructing the dialogue skills library includes: performing semantic extraction on the corpus data, extracting the user's key intention information from the corpus data; obtaining the user's key information of the user, and obtaining a first vector of the key intention information and the user's key information; extracting customer service speech features from the corpus data, and generating a second vector based on the extracted customer service speech features; associating the first vector and the second vector to generate an associated vector, and generating the dialogue skills library based on the associated vector.

[0122] In one embodiment of the present disclosure, the device 600 is also used to: determine user corpus data in the corpus data, and remove first interference information from the user corpus data to obtain user target corpus data; perform key information and semantic extraction on the user target corpus data to obtain key intention information of the user.

[0123] In one embodiment of the present disclosure, the device 600 is also used to: determine customer service corpus data in the corpus data, and mask the second interference information in the customer service corpus data to obtain customer service target corpus data; and extract customer service speech features from the customer service target corpus data.

[0124] In one embodiment of the present disclosure, the device 600 is further used to: obtain a target business tag of the historical question and answer information; obtain a target question list matching the target business tag from the question list knowledge base, and determine the target question from the target question list.

[0125] In one embodiment of the present disclosure, the device 600 is further used to: determine the last round of question and answer information in the historical question and answer information; extract key information from the last round of question and answer information, and obtain the target business tag based on the key information.

[0126] In one embodiment of the present disclosure, the device 600 is further used to: determine the last round of question and answer information in the historical question and answer information; extract key information from the last round of question and answer information, and obtain the target business tag based on the key information.

[0127] In one embodiment of the present disclosure, the device 600 is also used to: generate a second prompt for question screening according to question screening rules; input the second prompt into the big model, and the big model screens the questions in the target question list based on the second prompt to obtain the target question.

[0128] In one embodiment of the present disclosure, the device 600 is also used to: generate a third prompt for question tuning, and input the third prompt into the big model, and the big model polishes the target question and the auxiliary response information based on the third prompt to generate the target answer information.

[0129] In one embodiment of the present disclosure, the matching module 602 is further used to: generate a third vector based on the historical question and answer information; perform an approximate nearest neighbor ANN vector search in the dialogue skill library based on the third vector to obtain a candidate association vector that is closest to the third vector; and determine the target dialogue skill based on the candidate association vector.

[0130] In one embodiment of the present disclosure, the matching module 602 is further used to: obtain the confidence of the candidate association vector; filter the candidate association vector according to the confidence to obtain the target association vector; obtain the dialogue skill corresponding to the target association vector as the target dialogue skill.

[0131] The intelligent question and answer device proposed in the present disclosure obtains historical question and answer information, matches the target dialogue skill from a pre-built dialogue skill library according to the historical question and answer information, and generates target reply information according to the target dialogue skill in response to matching the target dialogue skill. Therefore, the present disclosure generates the target reply information according to the target dialogue skill, thereby improving the professionalism and persuasiveness of the target reply information, better meeting the needs of users, improving the accuracy of the target reply information in the intelligent question and answer scenario, and improving the user experience.

[0132] According to an embodiment of the present disclosure, the present disclosure also proposes an electronic device, a readable storage medium, and a computer program product.

[0133] Figure 7 A schematic block diagram of an example electronic device 700 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0134] like Figure 7As shown, the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0135] A number of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0136] The computing unit 701 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 701 performs the various methods and processes described above, such as the intelligent question-answering method. For example, in some embodiments, the intelligent question-answering method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the intelligent question-answering method described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the intelligent question-answering method in any other appropriate manner (e.g., by means of firmware).

[0137] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0138] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be presented to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, implements the functions / operations specified in the flow chart and / or block diagram. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0139] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, 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), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0140] To propose interactions with a user account, the systems and techniques described herein may be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user account; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user account can propose input to the computer. Other types of devices may also be used to propose interactions with the user account; for example, feedback proposed to the user account may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user account may be received in any form (including acoustic input, voice input, or tactile input).

[0141] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user account computer with a graphical user account interface or a web browser through which a user account can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communications network). Examples of communications networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0142] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0143] According to an embodiment of the present disclosure, the present disclosure further provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, the steps of the intelligent question-answering method described in the above embodiment of the present disclosure are implemented.

[0144] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0145] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. An intelligent question-answering method, wherein: The method comprises: Get historical question and answer information; According to the historical question and answer information, matching target dialogue skills from a pre-built dialogue skill library; In response to matching the target dialogue skill, the target response information is generated according to the target dialogue skill.

2. The method according to claim 1, wherein: The method further comprises: In response to failure to match the target dialogue skill, determining auxiliary response information determined by the large model based on the historical question and answer information; According to the historical question and answer information, matching the target question from a pre-built question list knowledge base; In response to matching the target question, generating target answer information according to the target question and the auxiliary answer information; In response to not matching the target question, target answer information is generated based on the auxiliary response information.

3. The method according to claim 2, wherein: The process of constructing the question list knowledge base includes: Acquire corpus data and determine the business tags corresponding to the corpus data; Obtain a list of sub-questions of the corpus data; Merging sub-question lists of corpus data associated with the same business tag to generate a question list corresponding to the business tag; The question list knowledge base is generated according to the question list corresponding to the business tag.

4. The method according to claim 3, wherein: The obtaining of the sub-question list of the corpus data includes: Performing masking processing on the user corpus data in the corpus data to obtain masked corpus data; Obtaining a first prompt word prompt for question extraction, and inputting the first prompt and the masked corpus data into a large model; The large model performs question extraction on the masked corpus data under the constraint of the first prompt to generate a sub-question list of the corpus data.

5. The method according to claim 3, wherein: The merging of sub-question lists of corpus data associated with the same business tag to generate a question list corresponding to the business tag includes: Obtaining a list of sub-questions of the corpus data associated with the same business tag; The sub-question list is extracted for correlation, and the sub-question list is logically optimized according to the correlation to generate a question list corresponding to the business tag.

6. The method according to claim 3, wherein: The determining of the business label corresponding to the corpus data includes: Compressing the corpus data to obtain a user intention text; Extracting auxiliary information for intent recognition from the corpus data; Identify the user intention text based on the auxiliary information to obtain intention key fields and category key fields; A business tag corresponding to the corpus data is generated according to the intent key field and the category key field.

7. The method according to any one of claims 1 to 6, wherein: The process of constructing the dialogue skill library includes: Performing semantic extraction on the corpus data to extract key intention information of the user from the corpus data; Acquire key user information of the user, and acquire the key intent information and a first vector of the key user information; Extracting customer service speech features from the corpus data, and generating a second vector according to the extracted customer service speech features; The first vector and the second vector are associated to generate an associated vector, and the dialogue skill library is generated based on the associated vector.

8. The method according to claim 7, wherein: The performing semantic extraction on the corpus data to extract key intention information of the user from the corpus data includes: Determining user corpus data in the corpus data, and removing first interference information from the user corpus data to obtain user target corpus data; Key information and semantics are extracted from the user target corpus data to obtain key intention information of the user.

9. The method according to claim 7, wherein: The extracting of customer service speech features from the corpus data includes: Determining customer service corpus data in the corpus data, and performing masking processing on second interference information in the customer service corpus data to obtain customer service target corpus data; The customer service speech features are extracted from the customer service target corpus data.

10. The method according to any one of claims 2 to 6, wherein: Determining the target question from a pre-built question list knowledge base based on the historical question and answer information includes: Obtain a target business tag of the historical question and answer information; A target question list matching the target business tag is obtained from the question list knowledge base, and the target question is determined from the target question list.

11. The method according to claim 10, wherein: The step of obtaining the target business tag of the historical question and answer information includes: Determining the last round of question and answer information in the historical question and answer information; Key information is extracted from the last round of question and answer information, and the target business tag is obtained based on the key information.

12. The method according to claim 11, wherein: The method further comprises: In response to the target service tag not being determined based on the key information, a historical service tag of the historical question and answer information is acquired as the target service tag.

13. The method according to claim 10, wherein: Determining the target problem from the target problem list includes: Generate the second prompt for question screening according to the question screening rules; The second prompt is input into the big model, and the big model screens the questions in the target question list based on the second prompt to obtain the target question.

14. The method according to claim 13, wherein: The step of generating target answer information according to the target question and the auxiliary answer information includes: Generate a third prompt for question optimization, and input the third prompt into the big model, so that the big model can polish the target question and the auxiliary response information based on the third prompt to generate the target response information.

15. The method according to claim 7, wherein: The step of matching the target dialogue skill from a pre-built dialogue skill library according to the historical question and answer information includes: Generate a third vector according to the historical question and answer information; According to the third vector, performing an approximate nearest neighbor ANN vector search in the dialogue skill library to obtain a candidate association vector that is closest to the third vector; The target dialogue skill is determined according to the candidate association vector.

16. The method according to claim 15, wherein: The step of determining the target dialogue skill according to the candidate association vector comprises: Obtaining the confidence of the candidate association vector; Filtering candidate association vectors according to the confidence level to obtain a target association vector; The dialogue skill corresponding to the target association vector is obtained as the target dialogue skill.

17. An intelligent question-answering device, wherein: The device comprises: Acquisition module, used to obtain historical question and answer information; A matching module, configured to match a target dialogue skill from a pre-built dialogue skill library according to the historical question and answer information; The reply module is used to generate the target reply information according to the target dialogue skill in response to matching the target dialogue skill.

18. An electronic device, characterized in that: including a processor and a memory; The processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the method according to any one of claims 1 to 16.

19. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 16 is implemented.

20. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 16.

Images