Service information generation method and device, equipment, medium and product

Abstract

The embodiment of the invention discloses a service information generation method and device, equipment and a medium, and the method comprises the steps: obtaining to-be-replied service dialogue information, and determining corresponding service interaction state information according to the to-be-replied service dialogue information and corresponding historical service dialogue information; performing knowledge content matching in a preset business knowledge base according to the business interaction state information, the to-be-replied business dialogue information and the historical business dialogue information to obtain business interaction prompt information; and inputting the to-be-replied service dialogue information and the service interaction prompt information into a service interaction information generation model to obtain service reply information corresponding to the to-be-replied service dialogue information. According to the technical scheme provided by the embodiment of the invention, the problem of low dialogue quality of an artificial intelligence customer service in a business interaction process is solved, the dialogue fluency and semantic accuracy of an artificial intelligence robot in the business interaction process can be improved, the illusion probability of an output result of an artificial intelligence model is reduced, and the business progress is smoother.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of artificial intelligence technology, and in particular, to a method, device, equipment, medium and product for generating business information. Background Art

[0002] In the application process of artificial intelligence dialogue robots in corporate communication accounts, they automatically reply according to the information content entered by users, realize dialogue with users, answer users' questions, and promote business progress and cooperation.

[0003] However, in the process of implementing the present invention, it is found that there are at least the following technical problems in the prior art:

[0004] The responses of artificial intelligence conversational robots lack vertical knowledge of specific business scenarios, the output conversation content is hallucinatory, the conversation quality is low, and it is not conducive to business development. Summary of the invention

[0005] The embodiments of the present invention provide a business information generation method, device, equipment, medium and product, which can improve the fluency and semantic accuracy of the conversation of the artificial intelligence robot during the business interaction process, reduce the probability of the output results of the artificial intelligence model presenting hallucinations, and make business progress smoother.

[0006] In a first aspect, an embodiment of the present invention provides a method for generating business information, the method comprising:

[0007] Obtaining the business conversation information to be replied, and determining the corresponding business interaction status information according to the business conversation information to be replied and the corresponding historical business conversation information;

[0008] According to the business interaction status information, the business conversation information to be replied and the historical business conversation information, knowledge content matching is performed in the preset business knowledge base to obtain business interaction prompt information;

[0009] The business conversation information to be replied and the business interaction prompt information are input into the business interaction information generation model to obtain the business reply information corresponding to the business conversation information to be replied.

[0010] In a second aspect, an embodiment of the present invention provides a device for generating business information, the device comprising:

[0011] The business interaction status confirmation module is used to obtain the business conversation information to be replied, and determine the corresponding business interaction status information according to the business conversation information to be replied and the corresponding historical business conversation information;

[0012] A prompt information determination module is used to match knowledge content in a preset business knowledge base according to business interaction status information, business conversation information to be replied and historical business conversation information to obtain business interaction prompt information;

[0013] The information generation module is used to input the business dialogue information to be replied and the business interaction prompt information into the business interaction information generation model to obtain the business reply information corresponding to the business dialogue information to be replied.

[0014] In a third aspect, an embodiment of the present invention further provides a computer device, the computer device comprising:

[0015] one or more processors;

[0016] A memory for storing one or more programs;

[0017] When one or more programs are executed by one or more processors, the one or more processors implement the business information generating method provided by any embodiment of the present invention.

[0018] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for generating business information as provided in any embodiment of the present invention.

[0019] In a fifth aspect, an embodiment of the present disclosure further provides a computer program product, including a computer program, which, when executed by a processor, implements a method for generating business information as provided in any embodiment of the present invention.

[0020] The embodiments of the above invention have the following advantages or beneficial effects:

[0021] The embodiment of the present invention obtains the business conversation information to be replied, and determines the corresponding business interaction status information according to the business conversation information to be replied and the corresponding historical business conversation information; performs knowledge content matching in a preset business knowledge base according to the business interaction status information, the business conversation information to be replied and the historical business conversation information to obtain business interaction prompt information; inputs the business conversation information to be replied and the business interaction prompt information into the business interaction information generation model to obtain the business reply information corresponding to the business conversation information to be replied. The technical solution of the embodiment of the present invention solves the problem of low conversation quality of artificial intelligence customer service during business interaction, can improve the fluency and semantic accuracy of conversation of artificial intelligence robots during business interaction, reduce the probability of illusion in the output results of artificial intelligence models, and make business progress smoother. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of a method for generating business information provided by an embodiment of the present invention;

[0023] Figure 2 is a flow chart of a method for generating business information provided by an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of an information processing process in constructing a business knowledge base provided by an embodiment of the present invention;

[0025] Figure 4 is a flow chart of a method for generating business information provided by an embodiment of the present invention;

[0026] Figure 5 It is a schematic diagram of a vectorization model training process provided by an embodiment of the present invention;

[0027] Figure 6 is a flow chart of a method for generating business information provided by an embodiment of the present invention;

[0028] Figure 7 This is a schematic diagram of a service information quality analysis provided by an embodiment of the present invention;

[0029] Figure 8 This is a schematic diagram of a standard process for generating business information provided by an embodiment of the present invention;

[0030] Fig. 9 This is a schematic diagram of a business information generation process in an application scenario provided by an embodiment of the present invention;

[0031] Fig.10 It is a structural schematic diagram of a business information generating device provided by an embodiment of the present invention;

[0032] Fig.11 It is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0034] Figure 1 This is a flowchart of a method for generating business information provided by an embodiment of the present invention. This embodiment can be applied to scenarios where the content of a conversation is automatically replied to, especially in some business scenarios where an artificial intelligence customer service agent talks to a user. The method can be executed by a business information generating device, which can be implemented by software and / or hardware and integrated into a computer device with application development capabilities.

[0035] like Figure 1As shown, the business information generation method of this embodiment includes the following steps:

[0036] S110: Acquire the business conversation information to be replied, and determine the corresponding business interaction state information according to the business conversation information to be replied and the corresponding historical business conversation information.

[0037] The business conversation information to be replied may be information input by a user in a business-related conversation, and the information may be information in one or more forms of voice, text, image or video. The business information generating device may generate corresponding business reply information for the content of the business conversation information to be replied, and complete a business information generating process. The business conversation information to be replied may also be prompt information for a user to enter a business-related conversation, and the business information generating device may generate corresponding response information according to the prompt information, actively start the business-related conversation content, and promote the business interaction.

[0038] Business-related conversations can be conversations in business interaction scenarios with automatic reply functions. By communicating with users through artificial intelligence, providing users with corresponding business information based on the information provided by users, and answering users' business-related questions, the business progress can be promoted on the premise of improving the speed of responding to user needs and reducing labor costs.

[0039] In this step, after the business conversation information to be replied is obtained, the corresponding business interaction state information is determined according to the business conversation information to be replied and the corresponding historical business conversation information.

[0040] The historical business conversation information corresponding to the business conversation information to be replied may be the historical conversation information in the conversation period where the business conversation information to be replied is located, or the historical business conversation information in the historical business conversation period for the business information generating device that issued the business conversation information to be replied.

[0041] The business interaction status information may be information used to indicate the progress of the business interaction. For example, the percentage of information that can be provided in the business dialogue information to be replied and the corresponding historical business dialogue information among all the information required or provided to complete the corresponding business content in any business scenario may be used to indicate the progress of the business interaction. Alternatively, in any business scenario, multiple stages of information interaction are required to complete the corresponding business content, and the key content of the information interaction in each stage is different. The information interaction stage corresponding to the business dialogue information to be replied and the corresponding historical business dialogue information may also indicate the progress of the business interaction.

[0042] When determining the corresponding business interaction status information based on the business conversation information to be replied and the corresponding historical business conversation information, the corresponding business interaction status information can be determined by matching the business conversation information to be replied and the corresponding historical business conversation information with the business progress content of different business interaction states.

[0043] The service interaction status information can enable the service information generating device to understand the progress of the service session. When subsequently generating service reply information, the service interaction status information is combined to generate service reply information matching the service progress, thereby improving the quality of the service reply information.

[0044] S120: performing knowledge content matching in a preset business knowledge base according to the business interaction state information, the business conversation information to be replied and the historical business conversation information to obtain business interaction prompt information.

[0045] The preset business knowledge base may be a knowledge information base of the business field corresponding to the business dialogue information to be replied, including dialogue content, question and answer information and related business knowledge in the business field.

[0046] The business interaction prompt information can be used as a prompt in the subsequent business reply information generation process, which can make the business reply information generated by the business interaction information generation model more in line with the context and semantics of the business interaction, and improve the accuracy of the end-to-end robot reply.

[0047] Knowledge content matching is performed in a preset business knowledge base based on business interaction status information, business conversation information to be replied to, and historical business conversation information. The vectors corresponding to the business interaction status information, business conversation information to be replied to, and historical business conversation information are matched with the knowledge content vectors in the preset business knowledge base, and the corresponding knowledge content is determined as the business interaction prompt information through methods such as vector distance or a preset similarity threshold of the vector.

[0048] S130: Input the business conversation information to be replied and the business interaction prompt information into a business interaction information generation model to obtain business reply information corresponding to the business conversation information to be replied.

[0049] The business interaction prompt information is a prompt word determined based on the business interaction status information, the business conversation information to be replied and the historical business conversation information. It can be understood as using the vertical knowledge information and business interaction status of the corresponding business field as prompts to generate business reply information corresponding to the business conversation information to be replied, which can improve the accuracy of the business reply information and improve the smoothness of the business process.

[0050] Among them, the business interaction information generation model can be a pre-trained large-scale language model (LargeLanguage model), which is a natural language model with large-scale parameters and complex computing structure, and is mainly constructed by a deep neural network.

[0051] Guided by the knowledge content in the preset business knowledge base and assisted by the business interaction prompt information, the business interaction information generation model can well realize its business response information generation function.

[0052] The technical solution of this embodiment obtains the business conversation information to be replied, and determines the corresponding business interaction status information according to the business conversation information to be replied and the corresponding historical business conversation information; performs knowledge content matching in a preset business knowledge base according to the business interaction status information, the business conversation information to be replied and the historical business conversation information to obtain business interaction prompt information; inputs the business conversation information to be replied and the business interaction prompt information into the business interaction information generation model to obtain the business reply information corresponding to the business conversation information to be replied. The technical solution of the embodiment of the present invention solves the problem of low conversation quality of artificial intelligence customer service during business interaction, and can improve the fluency and semantic accuracy of the conversation of artificial intelligence robots during business interaction, reduce the probability of illusion in the output results of artificial intelligence models, and make business progress smoother.

[0053] Figure 2 This is a flowchart of a method for generating business information provided by an embodiment of the present invention. This embodiment and the method for generating business information in the above embodiment belong to the same inventive concept, and further describes the process of determining business interaction status information. The method can be executed by a business information generating device, which can be implemented by software and / or hardware and integrated into a computer device with application development function.

[0054] like Figure 2 As shown, the business information generation method of this embodiment includes the following steps:

[0055] S210: Acquire the business conversation information to be replied, and extract the target business key information from the business conversation information to be replied and the historical business conversation information.

[0056] The target business key information may be key element information in the business content in the business scenario corresponding to the business dialogue information to be replied, which may be understood as key information in the communication between the user and the business information generating device during a business interaction process.

[0057] The business key information dictionary can be pre-set according to the business interaction target in the corresponding business scenario, and the target business key information in the business dialogue information to be replied and the historical business dialogue information can be determined and extracted by dictionary matching.

[0058] S220: Match the target business key information with the preset business interaction progress information, determine the business interaction progress of the session period corresponding to the business dialogue information to be replied, and use the business interaction progress as the business interaction state information.

[0059] The preset business interaction progress information may be reference information for indicating the process of business interaction, may include all business key information items required in business interaction under the corresponding business scenario, and may also include identification information of different business interaction progress during the business interaction process.

[0060] By matching the target business key information with the preset business interaction progress information, the business interaction progress of the session period corresponding to the business dialogue information to be replied can be determined, that is, the corresponding business interaction state information can be determined.

[0061] S230: performing knowledge content matching in a preset business knowledge base according to the business interaction state information, the business conversation information to be replied and the historical business conversation information to obtain business interaction prompt information.

[0062] Specifically, in the process of text recognition or text generation based on the model, text vectors are used as information objects for processing. Business interaction status information, pending business conversation information, and historical business conversation information can be converted into corresponding text vectors, that is, text vectorization representation. Text vectorization representation can be achieved through a vectorization model.

[0063] Vectorization refers to encoding words or text information into vectors that can represent semantic information. Common word vectors include bag-of-words model and skip-gram. Text vectors mainly rely on model encoding, such as Bert (Bidirectional Encoder Representation from Transformers) and other models with Transformer as the skeleton.

[0064] After obtaining the vector representation of business interaction status information, business conversation information to be replied and historical business conversation information, the corresponding vector representation can be matched with the vector representation of knowledge content in the preset business knowledge base to determine semantically similar or related knowledge content as business interaction prompt information.

[0065] The preset business knowledge base is constructed based on historical online business interaction information, historical offline business interaction information and business knowledge information in the corresponding business field, and can provide information reference for the business interaction information generation model. In an optional implementation, the business interaction information generation model can also be trained through the preset business knowledge base to achieve fine-tuning of the general model and obtain an expert model in the corresponding business field.

[0066] Specifically, the construction process of the preset business knowledge base may include the following process:

[0067] First, historical online business interaction information, historical offline business interaction information, and business knowledge information in the target business domain are obtained. Then, the historical online business interaction information, historical offline business interaction information, and business knowledge information are respectively vectorized to obtain corresponding information representation vectors; further, cluster analysis is performed on the information representation vectors to obtain information clustering results, and information is filtered on the information clustering results to obtain a preset business knowledge base. The preset business knowledge base includes knowledge question and answer pairs associated with the target business domain.

[0068] In a specific business knowledge base construction example, such as the construction of an artificial intelligence customer service knowledge base in business marketing, you can refer to Figure 3 The process shown.

[0069] The historical online business interaction information, historical offline business interaction information and business knowledge information in the target business field can be a large amount of conversation data accumulated by manual customer service and operations in the corresponding business scenarios, including enterprise communication account (online customer service) conversation data, customer service (manual or offline) conversation data, operation manual collation data and internal operation document data of relevant business departments, etc. All data needs to be cleaned and organized into question-answer pairs.

[0070] like Figure 3 As shown in the figure, for manually sorted data that does not need to extract question-answer pairs, it can be used directly; for pure document knowledge, the document can be segmented into text blocks, and then the generation ability of the large model can be used to extract question-answer pairs; for conversation data in enterprise micro-service scenarios, the question recognition method can be used to extract customer questions, and the manual replies can be used as reference answers. Among them, question recognition can use the StructBert classification model to mark questions, and the coverage of this method is better than regular rule recognition. After all question-answer pairs are aggregated, symbols and sensitive words are first filtered, and then the minhash algorithm is used for preliminary deduplication. The deduplicated data is sent to the BGE (BAAI General Embedding) model for semantic vectorization. Then, the k-means algorithm is used to cluster similar questions. These similar questions are ranked by weighted scores according to the number of classes and the average distance within the class, and the central questions with scores above the 0.25 percentile are obtained as the final selection. The final answer to the question will be manually verified to ensure the accuracy of the question-answer pair.

[0071]

[0072] Among them, score i is the weighted score of the i-th category, is the normalized value of the number of samples in this class, d i is the average center distance of the class. This score will be used as the basis for the final category sorting. Through observation, the number of cluster samples with lower scores tends to be smaller and the cluster distance is larger. Finally, the intra-class center problem with a score above 0.25 is selected as the final choice.

[0073] S240: Input the business conversation information to be replied and the business interaction prompt information into a business interaction information generation model to obtain business reply information corresponding to the business conversation information to be replied.

[0074] The technical solution of this embodiment obtains the business dialogue information to be replied, and extracts the target business key information from the business dialogue information to be replied and the historical business dialogue information; matches the target business key information with the preset business interaction progress information, determines the business interaction progress of the session period corresponding to the business dialogue information to be replied, and uses the business interaction progress as the business interaction status information; performs knowledge content matching in the preset business knowledge base according to the business interaction status information, the business dialogue information to be replied and the historical business dialogue information, and obtains the business interaction prompt information; inputs the business dialogue information to be replied and the business interaction prompt information into the business interaction information generation model, and obtains the business reply information corresponding to the business dialogue information to be replied. The technical solution of the embodiment of the present invention solves the problem of low dialogue quality of artificial intelligence customer service in the process of business interaction. The prompt information of the information generation model can be determined by the business interaction progress to generate the final business reply information, improve the fluency and semantic accuracy of the dialogue of the artificial intelligence robot in the process of business interaction, reduce the probability of hallucination in the output results of the artificial intelligence model, and make the business progress smoother.

[0075] Figure 4 This is a flowchart of a method for generating business information provided by an embodiment of the present invention. This embodiment and the method for generating business information in the above embodiment belong to the same inventive concept, and further illustrates the process of vectorizing the acquired information. The method can be executed by a business information generating device, which can be implemented by software and / or hardware and integrated into a computer device with application development function.

[0076] like Figure 4 As shown, the business information generation method of this embodiment includes the following steps:

[0077] S310: Acquire the business conversation information to be replied, and determine the corresponding business interaction state information according to the business conversation information to be replied and the corresponding historical business conversation information.

[0078] S320: Input text contents corresponding to the business interaction status information, the business conversation information to be replied, and the historical business conversation information into a preset text vectorization model respectively to obtain corresponding information text vectors.

[0079] Among them, the preset text vectorization model is a model determined by training the business knowledge texts and corresponding business knowledge text vectors in the preset business knowledge base as model training samples.

[0080] Optionally, the preset text vectorization model can be a BGE (BAAI General Embedding) model. The data used in BGE training is general text in various fields. In order to ensure the recall effect of its vectorization during retrieval, in this embodiment, the vector data in the preset business knowledge base of the corresponding business field is used to fine-tune the model in a targeted manner.

[0081] Specifically, the training process of the preset text vectorization model includes the following steps:

[0082] First, business knowledge texts and corresponding business knowledge text vectors are selected as anchor samples in the preset business knowledge base. All questions in the preset business knowledge base can be extracted as anchor samples.

[0083] Then, according to the vector distance between each model training sample and the anchor sample in the preset business knowledge base, determine the positive and negative samples corresponding to the anchor sample. For positive samples, it is necessary to generate samples similar to the anchor sample, and similar samples generated during clustering can be used. For negative samples, random sampling can be performed in the vector of the entire preset business knowledge base. The following formula As shown, x is the anchor point vector, x+ is the positive sample vector, x- is the negative sample vector, s(x,y) is the distance function, and the dot product similarity of vectors x and y is taken here.

[0084] Through sample vectorization, the core idea of ​​Info Noise Contrastive Estimation (InfoNCE) is to make the distance between the positive sample vector and the anchor vector closer, and the distance between the negative sample vector and the anchor vector farther. To ensure the vector effect, samples with a similarity of more than 0.90 to the anchor are taken as positive samples, and samples with a similarity of less than 0.85 to the anchor are taken as negative samples. At the same time, the similarity of positive samples is taken as low as possible, and the similarity of negative samples is taken as high as possible. In this way, 18,000 positive and negative sample triplets of the questions are finally obtained, and BGE is fine-tuned by contrastive learning. The top5 recall rate of the offline test set can be improved by about 8% after fine-tuning.

[0085] Finally, the initial text vectorization model is trained through the triplets consisting of anchor samples, positive samples, and negative samples to obtain the preset text vectorization model. The training process can be as follows Figure 5 . Among them, the encoder f(.) represents the corresponding model to be trained.

[0086] S330, performing vector analysis based on the information text vector and the business knowledge text vector in the preset business knowledge base, determining a target business knowledge text vector that semantically matches the information text vector, and using the target business knowledge text vector as business interaction prompt information.

[0087] When performing vector analysis based on the information text vector and the business knowledge text vector in the preset business knowledge base, the corresponding knowledge content can be determined as business interaction prompt information by calculating the vector distance between the information text vector and the business knowledge text vector or a preset similarity threshold of the vectors.

[0088] S340: Input the business conversation information to be replied and the business interaction prompt information into a business interaction information generation model to obtain business reply information corresponding to the business conversation information to be replied.

[0089] The technical solution of this embodiment is to obtain the business conversation information to be replied, and determine the corresponding business interaction status information according to the business conversation information to be replied and the corresponding historical business conversation information; input the text contents corresponding to the business interaction status information, the business conversation information to be replied and the historical business conversation information into a preset text vectorization model respectively to obtain the corresponding information text vector; perform vector analysis based on the information text vector and the business knowledge text vector in the preset business knowledge base to determine the target business knowledge text vector that semantically matches the information text vector, and use the target business knowledge text vector as the business interaction prompt information; input the business conversation information to be replied and the business interaction prompt information into the business interaction information generation model to obtain the business reply information corresponding to the business conversation information to be replied. The technical solution of the embodiment of the present invention solves the problem of low conversation quality of artificial intelligence customer service during business interaction, and further solves the problem of low vectorization recall rate of business interaction information. Various information can be vectorized through a fine-tuned preset text vectorization model, which improves the recall effect of vectorization results during retrieval. At the same time, the prompt information of the information generation model is determined according to the progress of business interaction, and the final business response information is generated, thereby improving the fluency and semantic accuracy of the artificial intelligence robot's conversation during business interaction, reducing the probability of illusions in the output results of the artificial intelligence model, and making business progress smoother.

[0090] Figure 6 This is a flow chart of a method for generating business information provided by an embodiment of the present invention. This embodiment and the method for generating business information in the above embodiment belong to the same inventive concept, and further illustrates the process of analyzing the conversation quality of business reply information. The method can be executed by a business information generating device, which can be implemented by software and / or hardware and integrated into a computer device with application development function.

[0091] like Figure 6As shown, the business information generation method of this embodiment includes the following steps:

[0092] S410: Acquire the business conversation information to be replied, and determine the corresponding business interaction state information according to the business conversation information to be replied and the corresponding historical business conversation information.

[0093] S420: performing knowledge content matching in a preset business knowledge base according to the business interaction state information, the business conversation information to be replied and the historical business conversation information to obtain business interaction prompt information.

[0094] S430: Input the business conversation information to be replied and the business interaction prompt information into a business interaction information generation model to obtain business reply information corresponding to the business conversation information to be replied.

[0095] S440: Perform a dialogue quality analysis on the business response information, and output the business response information if the business response information passes the dialogue quality analysis.

[0096] Specifically, the dialogue quality analysis of the business response information may include repetition rate analysis, semantic matching confidence analysis, and abnormal information filtering analysis such as sensitive information, so as to ensure the quality of the business response information.

[0097] The business reply information passes the dialogue quality analysis, indicating that the quality of the business reply information meets the corresponding quality standards, and the business reply information can be output to reply to the business dialogue information to be replied. If the business reply information does not pass the dialogue quality analysis, the business reply information generated by the model will not be directly output, and the relevant business personnel can be notified to manually intervene to reply.

[0098] In an optional implementation, the process of analyzing the conversation quality of the service reply information can refer to Figure 7The process shown. On the one hand, it is necessary to determine whether the business reply information is a duplicate reply message in the validity period of the corresponding business session, that is, to filter and control the duplicate messages to prevent the duplicate messages from reaching the customer multiple times. On the other hand, it is necessary to determine the semantic matching confidence between the business reply information and the business conversation information to be replied, that is, to manage the accuracy of the large model conversation matching. The matching level can be defined according to the value of the semantic matching confidence, including four levels from 0 to 3. 0 means divergent answer, indicating that the confidence of the business reply information is low and may not answer the question. 1 means vague correctness, which can answer the question but cannot provide more comprehensive information. 2 means absolutely correct, with high semantic matching confidence, and can be directly output as the reply content. 3 Others may be some custom situations that require further information processing. On the other hand, it will also determine whether the business reply information contains abnormal information or abnormal status, that is, to monitor the health status of the conversation. In the application scenario of the automatic reply robot of the enterprise communication account, the health status of the conversation can be monitored by the online status check monitoring of the enterprise communication account or the post-holiday monitoring status check monitoring of the large model.

[0099] Through the technical solution of this embodiment, the quality of business response information generated by the business interaction information generation model can be monitored, and the content of the speech generated by the large model can be finely managed and controlled, which can effectively prevent the anomalies that may occur in the parsing process of the large model, and avoid sending vague or irrelevant messages to users, thereby maintaining the user experience and ensuring that every communication can truly reach the user's core needs and improve communication efficiency and quality.

[0100] Exemplarily, the application of the business information generation method in a large model robot in a marketing scenario is used as an example to illustrate the process of implementing the business information generation method. In this example, Langchain is used to build a series of SOPs (Standard Operation Procedures) in the business information generation process, build a business question and answer knowledge base, and retrieve vectors stored in the business question and answer knowledge base (vector library) based on user questions and / or non-question speeches. The accuracy of the large model robot's responses is improved through large model polishing, reducing error cases.

[0101] Specifically, the SOP for the business information generation function in this scenario is as follows: Figure 8As shown. Among them, the interface can be an interface for users to interact with the large model robot, such as an interface for automatically replying to a robot dialogue with an enterprise communication account. The interface input can be the information content input from the above interface, corresponding to the business dialogue information to be replied. Then, the dialogue state can be analyzed to determine the business interaction state information. The summary information (the business dialogue information to be replied, that is, the corresponding historical business interaction information), the question and answer pair extraction (Query extraction) in the conversation, and the corresponding fixed speech content can be used to search the question and answer knowledge base for knowledge content matching. Then, business reply information can be generated based on the question and answer knowledge base retrieval results and summary information, the extracted question and answer pair information and the corresponding fixed speech content, and the business reply information is subjected to confidence judgment and information parsing analysis (quality analysis) before information output. Among them, RAG (Retrieval Augmented Generation) is a way to realize vector search to retrieve relevant information from the question and answer knowledge base, and generate prompts in combination with questions, so that the model can provide accurate answers.

[0102] Furthermore, the process of obtaining user input information from the interface can be achieved through event subscription. Users can actively or passively add enterprise communication account friends to establish a dialogue channel with the enterprise communication account robot. In turn, the enterprise communication account robot can obtain the business interaction information to be replied to by subscribing to the friend's event or the user's message event. Fig. 9 As shown in the figure, when a friend sends an event and / or a customer sends a message, the relevant interface of the AI ​​big model (business interaction information generation model) will be called, the event type will be input, and the big model will return the response words (answer) after parsing, and send it to the user (customer) through the corresponding output interface. This realizes efficient and high-quality business interaction between the business side and the user.

[0103] Fig.10 A structural schematic diagram of a business information generating device provided in an embodiment of the present invention. This embodiment can be applied to scenarios where automatic replies to conversation content occur, especially in situations where artificial intelligence customer service staff have conversations with users in some business scenarios. The business information generating device can be implemented by software and / or hardware and integrated into a computer terminal device with application development capabilities.

[0104] like Fig.10 As shown, the service information generating device includes: a service interaction state confirming module 510 , a prompt information determining module 520 and an information generating module 530 .

[0105] Among them, the business interaction status confirmation module 510 is used to obtain the business dialogue information to be replied, and determine the corresponding business interaction status information according to the business dialogue information to be replied and the corresponding historical business dialogue information; the prompt information determination module 520 is used to match the knowledge content in the preset business knowledge base according to the business interaction status information, the business dialogue information to be replied and the historical business dialogue information to obtain the business interaction prompt information; the information generation module 530 is used to input the business dialogue information to be replied and the business interaction prompt information into the business interaction information generation model to obtain the business reply information corresponding to the business dialogue information to be replied.

[0106] The technical solution of this embodiment obtains the business conversation information to be replied, and determines the corresponding business interaction status information according to the business conversation information to be replied and the corresponding historical business conversation information; performs knowledge content matching in a preset business knowledge base according to the business interaction status information, the business conversation information to be replied and the historical business conversation information to obtain business interaction prompt information; inputs the business conversation information to be replied and the business interaction prompt information into the business interaction information generation model to obtain the business reply information corresponding to the business conversation information to be replied. The technical solution of the embodiment of the present invention solves the problem of low conversation quality of artificial intelligence customer service during business interaction, and can improve the fluency and semantic accuracy of the conversation of artificial intelligence robots during business interaction, reduce the probability of illusion in the output results of artificial intelligence models, and make business progress smoother.

[0107] In an optional implementation, the service interaction status confirmation module 510 is specifically used to:

[0108] Extract key information of the target business from the business conversation information to be replied and the historical business conversation information;

[0109] The target business key information is matched with the preset business interaction progress information, the business interaction progress of the session cycle corresponding to the business dialogue information to be replied is determined, and the business interaction progress is used as the business interaction state information.

[0110] In an optional implementation, the prompt information determination module 520 is specifically configured to:

[0111] Input the text contents corresponding to the business interaction status information, the business conversation information to be replied and the historical business conversation information into the preset text vectorization model respectively to obtain the corresponding information text vector; wherein the preset text vectorization model is a model determined by training the business knowledge text and the corresponding business knowledge text vector in the preset business knowledge base as model training samples;

[0112] Based on the information text vector and the business knowledge text vector in the preset business knowledge base, vector analysis is performed to determine the target business knowledge text vector that semantically matches the information text vector, and use the target business knowledge text vector as business interaction prompt information.

[0113] In an optional implementation, the business information generating device further includes a model training module for training a preset text vectorization model, wherein the training process of the preset text vectorization model includes:

[0114] Select business knowledge text and corresponding business knowledge text vector as anchor point samples in the preset business knowledge base;

[0115] According to the vector distance between each model training sample and the anchor sample in the preset business knowledge base, the positive sample and the negative sample corresponding to the anchor sample are determined;

[0116] The initial text vectorization model is trained through a triplet consisting of an anchor sample, a positive sample, and a negative sample to obtain a preset text vectorization model.

[0117] In an optional implementation, the business information generating device further includes a business reply information quality inspection module, which is used to:

[0118] A dialogue quality analysis is performed on the business response information, and when the business response information passes the dialogue quality analysis, the business response information is output.

[0119] In an optional implementation, the business response information quality inspection module is specifically used to:

[0120] Determine whether the service reply information is a duplicate reply message within the validity period of the corresponding service session;

[0121] and / or, determining a semantic matching confidence level between the service reply information and the service conversation information to be replied;

[0122] And / or, determine whether the business response information contains abnormal information.

[0123] In an optional implementation, the business information generating device further includes a knowledge base building module, which is used to:

[0124] Obtain historical online business interaction information, historical offline business interaction information and business knowledge information within the target business field;

[0125] The historical online business interaction information, the historical offline business interaction information and the business knowledge information are respectively vectorized to obtain the corresponding information representation vectors;

[0126] Perform cluster analysis on the information representation vector to obtain information clustering results, and perform information screening on the information clustering results to obtain a preset business knowledge base;

[0127] The preset business knowledge base includes knowledge question and answer pairs associated with the target business field.

[0128] The service information generating device provided in the embodiment of the present invention can execute the service information generating method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0129] Fig.11 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Fig.11 A block diagram of an exemplary computer device 12 suitable for use in implementing embodiments of the present invention is shown. Fig.11 The computer device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention. The computer device 12 can be any terminal device with computing capabilities, such as an intelligent controller and server, a mobile phone and other terminal devices.

[0130] like Fig.11 As shown, the computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0131] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0132] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0133] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Fig.11not shown, usually called a "hard drive"). Although Fig.11 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The system memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0134] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0135] The computer device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), one or more devices that enable a user to interact with the computer device 12, and / or any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the computer device 12 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with the other modules of the computer device 12 via the bus 18. It should be understood that although Fig.11 Not shown, other hardware and / or software modules may be used in conjunction with computer device 12, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0136] The processing unit 16 executes various functional applications and data processing by running the program stored in the system memory 28, for example, implementing the service information generation method provided in the embodiment of the present invention, which includes:

[0137] Obtaining the business conversation information to be replied, and determining the corresponding business interaction status information according to the business conversation information to be replied and the corresponding historical business conversation information;

[0138] According to the business interaction status information, the business conversation information to be replied and the historical business conversation information, knowledge content matching is performed in the preset business knowledge base to obtain business interaction prompt information;

[0139] The business conversation information to be replied and the business interaction prompt information are input into the business interaction information generation model to obtain the business reply information corresponding to the business conversation information to be replied.

[0140] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for generating business information provided by any embodiment of the present invention is implemented. The method includes:

[0141] Obtaining the business conversation information to be replied, and determining the corresponding business interaction status information according to the business conversation information to be replied and the corresponding historical business conversation information;

[0142] According to the business interaction status information, the business conversation information to be replied and the historical business conversation information, knowledge content matching is performed in the preset business knowledge base to obtain business interaction prompt information;

[0143] The business conversation information to be replied and the business interaction prompt information are input into the business interaction information generation model to obtain the business reply information corresponding to the business conversation information to be replied.

[0144] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, 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 above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0145] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0146] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0147] Computer program code for performing the operation of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0148] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0149] The embodiments of the present disclosure also provide a computer program product, including a computer program, which, when executed by a processor, implements the business information generating method provided in any one of the embodiments of the present disclosure.

[0150] In the process of implementation, the computer program product can be written in one or more programming languages ​​or a combination thereof to perform the computer program code for the disclosed operation, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect through the Internet).

[0151] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for generating business information, characterized in that: include: Acquire the business conversation information to be replied, and determine the corresponding business interaction state information according to the business conversation information to be replied and the corresponding historical business conversation information; According to the business interaction state information, the business conversation information to be replied and the historical business conversation information, knowledge content matching is performed in a preset business knowledge base to obtain business interaction prompt information; The business conversation information to be replied and the business interaction prompt information are input into a business interaction information generation model to obtain business reply information corresponding to the business conversation information to be replied.

2. The method according to claim 1, characterized in that The determining corresponding service interaction status information according to the service conversation information to be replied and the corresponding historical service conversation information includes: Extracting target business key information from the to-be-replied business conversation information and the historical business conversation information; The target business key information is matched with preset business interaction progress information to determine the business interaction progress of the session period corresponding to the business dialogue information to be replied, and the business interaction progress is used as the business interaction state information.

3. The method according to claim 1, characterized in that: The performing knowledge content matching in a preset business knowledge base according to the business interaction state information, the to-be-replied business dialogue information and the historical business dialogue information to obtain the business interaction prompt information includes: Input the text contents corresponding to the business interaction status information, the business conversation information to be replied and the historical business conversation information into a preset text vectorization model respectively to obtain corresponding information text vectors; wherein the preset text vectorization model is a model determined by training the business knowledge texts and corresponding business knowledge text vectors in the preset business knowledge base as model training samples; Based on the information text vector and the business knowledge text vector in the preset business knowledge base, a target business knowledge text vector that semantically matches the information text vector is determined, and the target business knowledge text vector is used as the business interaction prompt information.

4. The method according to claim 3, characterized in that The training process of the preset text vectorization model includes: Selecting business knowledge text and corresponding business knowledge text vector as anchor point samples in the preset business knowledge base; Determine the positive samples and negative samples corresponding to the anchor point samples according to the vector distance between each model training sample in the preset business knowledge base and the anchor point sample; The initial text vectorization model is trained by using a triple consisting of the anchor point sample, the positive sample and the negative sample to obtain the preset text vectorization model.

5. The method according to any one of claims 1 to 4, characterized in that: Also includes: Performing a dialogue quality analysis on the service reply information, and outputting the service reply information if the service reply information passes the dialogue quality analysis.

6. The method according to claim 5, characterized in that The performing dialogue quality analysis on the business response information includes: Determine whether the service reply information is a duplicate reply message within the validity period of the corresponding service session; and / or, determining a semantic matching confidence level between the service reply information and the service conversation information to be replied; And / or, determining whether the business response information contains abnormal information.

7. The method according to claim 1, characterized in that The process of constructing the preset business knowledge base includes: Obtain historical online business interaction information, historical offline business interaction information and business knowledge information within the target business field; Vectorizing the historical online business interaction information, the historical offline business interaction information, and the business knowledge information to obtain corresponding information representation vectors; Performing cluster analysis on the information representation vector to obtain information clustering results, and performing information screening on the information clustering results to obtain the preset business knowledge base; The preset business knowledge base includes knowledge question and answer pairs associated with the target business field.

8. A business information generating device, characterized in that: include: A business interaction status confirmation module is used to obtain the business conversation information to be replied, and determine the corresponding business interaction status information according to the business conversation information to be replied and the corresponding historical business conversation information; A prompt information determination module, configured to perform knowledge content matching in a preset business knowledge base according to the business interaction state information, the business conversation information to be replied and the historical business conversation information, to obtain business interaction prompt information; The information generation module is used to input the business dialogue information to be replied and the business interaction prompt information into the business interaction information generation model to obtain the business reply information corresponding to the business dialogue information to be replied.

9. A computer device, characterized in that: The computer device comprises: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the business information generating method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the business information generating method as described in any one of claims 1 to 7 is implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the computer program implements the service information generating method according to any one of claims 1 to 7.

Images