Methods and apparatus for information recognition and processing based on business scenarios

Abstract

This invention relates to the field of information recognition and processing technology, and discloses a method and apparatus for information recognition and processing based on business scenarios. The method includes: performing dual-path parallel processing on the information to be recognized based on a first analysis model and a second analysis model to obtain preliminary processing results; extracting target key information corresponding to the preliminary processing results; generating target completion information based on acquired historical interaction information and target key information; determining target intent information and target entity information corresponding to the information to be recognized based on the target completion information to generate target association information; determining target operation instructions based on the target association information and a predefined business process template, and executing the corresponding instructions to obtain control feedback results, thereby generating the information recognition result corresponding to the information to be recognized. Therefore, this invention can perform information recognition and processing, which is beneficial to improving the accuracy and reliability of information recognition and processing, as well as enhancing the intelligence and efficiency of information recognition and processing.

Description

Technical Field

[0001] This invention relates to the field of information recognition and processing technology, and in particular to a method and apparatus for information recognition and processing based on business scenarios. Background Technology

[0002] With the continuous development of artificial intelligence, intelligent voice interaction and text recognition have been widely applied in various scenarios of daily life. Traditional intent recognition and information extraction usually rely on fixed extraction logic, which is a relatively simple approach. In addition, for business processes that need to be executed automatically, such as work order submission, CRM updates, and appointment scheduling, traditional systems lack deep integration with backend systems and cannot achieve closed-loop processing from user input to automatic execution of the business process, resulting in low efficiency and accuracy in information recognition and processing.

[0003] Therefore, it is particularly important to provide a new information identification and processing method to improve the efficiency and accuracy of information identification and processing. Summary of the Invention

[0004] This invention provides a method and apparatus for information recognition and processing based on business scenarios, which can recognize and process information in various business scenarios, thereby improving the accuracy and reliability of information recognition and processing, as well as the intelligence and efficiency of information recognition and processing.

[0005] The first aspect of this invention discloses a method for information identification and processing based on a business scenario, the method comprising:

[0006] The information to be identified is acquired, and a dual-path parallel processing operation is performed on the information to be identified based on a pre-determined first analysis model and a pre-determined second analysis model to obtain preliminary processing results.

[0007] Extract the target key information corresponding to the preliminary processing result, and generate target completion information based on the obtained historical interaction information and the target key information;

[0008] Based on the target completion information, the target intent information and the target entity information corresponding to the information to be identified are determined. Target association information is generated based on the target intent information and the target entity information. Target operation instructions are determined based on the target association information and a predefined business process template. Control operations matching the target operation instructions are executed to obtain control feedback results.

[0009] Based on the control feedback result, an information recognition result corresponding to the information to be recognized is generated.

[0010] As an optional implementation, in a first aspect of the present invention, the method further includes:

[0011] Based on the information recognition results, structured text information is generated, and a speech synthesis operation is performed on the structured text information through a pre-determined speech conversion module to obtain synthesized output speech;

[0012] The system obtains user information corresponding to the voice receiving user, determines voice receiving parameters based on the user information, performs voice adjustment operations on the synthesized output voice based on the voice receiving parameters, obtains target output voice, and feeds back the target output voice to the user terminal corresponding to the voice receiving user.

[0013] As an optional implementation, in a first aspect of the present invention, the step of performing dual-path parallel processing on the information to be identified based on a predetermined first analysis model and a predetermined second analysis model to obtain preliminary processing results includes:

[0014] The information to be identified is input into a predetermined first analysis model to perform a first analysis operation on the information to be identified through the predetermined first analysis model to obtain a first output result, wherein the first output result includes the intent output information corresponding to the information to be identified and the confidence parameter corresponding to each intent output information;

[0015] The information to be identified is input into a pre-determined second analysis model to perform a second analysis operation on the information to be identified through the pre-determined second analysis model, and a second output result is obtained. The second output result includes the information intent category corresponding to the information to be identified and the probability parameter corresponding to each information intent category.

[0016] Normalization is performed on the first output result and the second output result to obtain the normalization result, and a preliminary processing result corresponding to the information to be identified is generated based on the normalization result.

[0017] As an optional implementation, in a first aspect of the present invention, the extraction of the target key information corresponding to the preliminary processing result includes:

[0018] Perform preprocessing operations on the preliminary processing results to obtain preprocessed results, wherein the preprocessing operations include one or more of the following: word segmentation, encoding, punctuation correction, and text cleaning.

[0019] Historical semantic information is acquired, and a target extraction model is constructed based on the historical semantic information, wherein the historical semantic information includes one or more of historical background context information, historical intent information, and historical context information;

[0020] Based on the target extraction model, an extraction operation is performed on the preliminary processing result to obtain a structured extraction result, and based on the structured extraction result, the target key information corresponding to the preliminary processing result is generated.

[0021] As an optional implementation, in a first aspect of the present invention, generating target completion information based on the acquired historical interaction information and the target key information includes:

[0022] Obtain historical interaction information, and based on the obtained historical interaction information, determine the target missing information corresponding to the target key information;

[0023] Based on the missing target information, completion guidance parameters are generated, and the completion guidance parameters are transmitted to the target user terminal corresponding to the information to be identified, and completion feedback information corresponding to the target user terminal is obtained.

[0024] Based on the completion feedback information, an information completion operation is performed on the target key information to generate target completion information.

[0025] As an optional implementation, in a first aspect of the present invention, determining the target operation instruction based on the target association information and a predefined business process template includes:

[0026] Based on the target association information and the predefined business process template, a mapping relationship is generated between the target association information and the predefined business process template.

[0027] Based on the mapping relationship, the backend call instructions are assembled, and the target operation instructions are generated based on all the backend call instructions.

[0028] As an optional implementation, in a first aspect of the present invention, the step of performing normalization processing on the first output result and the second output result to obtain a normalization processing result, and generating a preliminary processing result corresponding to the information to be identified based on the normalization processing result, includes:

[0029] Normalization is performed on the first output result and the second output result to obtain a normalized result, wherein the normalized result includes the confidence scores corresponding to the first output result and the second output result;

[0030] Based on the normalization processing result, determine the result weight parameters corresponding to the first output result and the second output result, determine the weighted average confidence level of the first output result and the second output result based on the result weight parameters, and determine whether the weighted average confidence level is greater than or equal to a preset confidence level threshold.

[0031] When it is determined that the weighted average confidence level is greater than or equal to the preset confidence level threshold, a preliminary processing result corresponding to the information to be identified is generated based on the normalization processing result.

[0032] When it is determined that the weighted average confidence level is less than the preset confidence level threshold, the first semantic information corresponding to the first output result and the second semantic information corresponding to the second output result are determined, and it is determined whether the first semantic information and the second semantic information match.

[0033] When it is determined that the first semantic information matches the second semantic information, a preliminary processing result corresponding to the information to be identified is generated based on the first semantic information and the second semantic information.

[0034] When it is determined that the first semantic information does not match the second semantic information, supplementary description information corresponding to the information to be identified is obtained, and a preliminary processing result corresponding to the information to be identified is generated based on the supplementary description information and the information to be identified.

[0035] A second aspect of the present invention discloses an apparatus for information recognition and processing based on a business scenario, the apparatus comprising:

[0036] The acquisition module is used to acquire the information to be identified.

[0037] The processing module is used to perform dual-path parallel processing operations on the information to be identified based on a pre-determined first analysis model and a pre-determined second analysis model to obtain preliminary processing results;

[0038] The extraction module is used to extract the target key information corresponding to the preliminary processing results;

[0039] The generation module is used to generate target completion information based on the acquired historical interaction information and the target key information;

[0040] The determination module is used to determine the target intent information corresponding to the information to be identified and the target entity information corresponding to the information to be identified based on the target completion information;

[0041] The generation module is further configured to generate target association information based on the target intent information and the target entity information;

[0042] The determining module is further configured to determine the target operation instruction based on the target association information and the predefined business process template;

[0043] The execution module is used to execute control operations that match the target operation instructions and obtain control feedback results;

[0044] The generation module is further configured to generate an information recognition result corresponding to the information to be recognized based on the control feedback result.

[0045] As an optional implementation, in a second aspect of the present invention, the generation module is further configured to generate structured text information based on the information recognition result;

[0046] The device further includes:

[0047] The synthesis module is used to perform speech synthesis operations on the structured text information through a pre-determined speech conversion module to obtain synthesized output speech;

[0048] The acquisition module is also used to acquire user information corresponding to the voice receiving user;

[0049] The determining module is further configured to determine voice reception parameters based on the user information;

[0050] The adjustment module is used to perform a speech adjustment operation on the synthesized output speech based on the speech receiving parameters to obtain the target output speech;

[0051] The feedback module is used to feed back the target output voice to the user terminal corresponding to the voice receiving user.

[0052] As an optional implementation, in a second aspect of the present invention, the processing module performs dual-path parallel processing on the information to be identified based on a predetermined first analysis model and a predetermined second analysis model to obtain preliminary processing results. The specific methods include:

[0053] The information to be identified is input into a predetermined first analysis model to perform a first analysis operation on the information to be identified through the predetermined first analysis model to obtain a first output result, wherein the first output result includes the intent output information corresponding to the information to be identified and the confidence parameter corresponding to each intent output information;

[0054] The information to be identified is input into a pre-determined second analysis model to perform a second analysis operation on the information to be identified through the pre-determined second analysis model, and a second output result is obtained. The second output result includes the information intent category corresponding to the information to be identified and the probability parameter corresponding to each information intent category.

[0055] Normalization is performed on the first output result and the second output result to obtain the normalization result, and a preliminary processing result corresponding to the information to be identified is generated based on the normalization result.

[0056] As an optional implementation, in a second aspect of the present invention, the extraction module extracts the target key information corresponding to the preliminary processing result in the following specific manner:

[0057] Perform preprocessing operations on the preliminary processing results to obtain preprocessed results, wherein the preprocessing operations include one or more of the following: word segmentation, encoding, punctuation correction, and text cleaning.

[0058] Historical semantic information is acquired, and a target extraction model is constructed based on the historical semantic information, wherein the historical semantic information includes one or more of historical background context information, historical intent information, and historical context information;

[0059] Based on the target extraction model, an extraction operation is performed on the preliminary processing result to obtain a structured extraction result, and based on the structured extraction result, the target key information corresponding to the preliminary processing result is generated.

[0060] As an optional implementation, in a second aspect of the present invention, the specific method by which the generation module generates target completion information based on the acquired historical interaction information and the target key information includes:

[0061] Obtain historical interaction information, and based on the obtained historical interaction information, determine the target missing information corresponding to the target key information;

[0062] Based on the missing target information, completion guidance parameters are generated, and the completion guidance parameters are transmitted to the target user terminal corresponding to the information to be identified, and completion feedback information corresponding to the target user terminal is obtained.

[0063] Based on the completion feedback information, an information completion operation is performed on the target key information to generate target completion information.

[0064] As an optional implementation, in a second aspect of the present invention, the specific method by which the determining module determines the target operation instruction based on the target association information and the predefined business process template includes:

[0065] Based on the target association information and the predefined business process template, a mapping relationship is generated between the target association information and the predefined business process template.

[0066] Based on the mapping relationship, the backend call instructions are assembled, and the target operation instructions are generated based on all the backend call instructions.

[0067] As an optional implementation, in a second aspect of the present invention, the specific method by which the processing module performs normalization processing on the first output result and the second output result to obtain a normalized processing result, and generates a preliminary processing result corresponding to the information to be identified based on the normalized processing result, includes:

[0068] Normalization is performed on the first output result and the second output result to obtain a normalized result, wherein the normalized result includes the confidence scores corresponding to the first output result and the second output result;

[0069] Based on the normalization processing result, determine the result weight parameters corresponding to the first output result and the second output result, determine the weighted average confidence level of the first output result and the second output result based on the result weight parameters, and determine whether the weighted average confidence level is greater than or equal to a preset confidence level threshold.

[0070] When it is determined that the weighted average confidence level is greater than or equal to the preset confidence level threshold, a preliminary processing result corresponding to the information to be identified is generated based on the normalization processing result.

[0071] When it is determined that the weighted average confidence level is less than the preset confidence level threshold, the first semantic information corresponding to the first output result and the second semantic information corresponding to the second output result are determined, and it is determined whether the first semantic information and the second semantic information match.

[0072] When it is determined that the first semantic information matches the second semantic information, a preliminary processing result corresponding to the information to be identified is generated based on the first semantic information and the second semantic information.

[0073] When it is determined that the first semantic information does not match the second semantic information, supplementary description information corresponding to the information to be identified is obtained, and a preliminary processing result corresponding to the information to be identified is generated based on the supplementary description information and the information to be identified.

[0074] A third aspect of the present invention discloses another apparatus for information recognition and processing based on a business scenario, the apparatus comprising:

[0075] Memory containing executable program code;

[0076] A processor coupled to the memory;

[0077] The processor calls the executable program code stored in the memory to execute some or all of the steps in the information identification and processing method based on business scenarios according to any of the first aspects of the present invention.

[0078] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the information identification and processing method based on any of the business scenarios described in the first aspect of the present invention.

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] In this embodiment of the invention, a preliminary processing result is obtained by performing dual-path parallel processing on the information to be identified based on a first analysis model and a second analysis model; target key information corresponding to the preliminary processing result is extracted, and target completion information is generated based on the acquired historical interaction information and target key information; based on the target completion information, target intent information and target entity information corresponding to the information to be identified are determined to generate target association information; target operation instructions are determined based on the target association information and a predefined business process template, and the corresponding instructions are executed to obtain control feedback results, thereby generating the information identification result corresponding to the information to be identified. It is evident that implementing this invention enables information identification and processing in various business scenarios, which is beneficial for improving the accuracy and reliability of information identification and processing, as well as enhancing the intelligence and efficiency of information identification and processing. Attached Figure Description

[0081] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0082] Figure 1 This is a flowchart illustrating a method for information identification and processing based on a business scenario, as disclosed in an embodiment of the present invention.

[0083] Figure 2 This is a flowchart illustrating another method for information identification and processing based on a business scenario disclosed in an embodiment of the present invention;

[0084] Figure 3 This is a schematic diagram of the structure of an information recognition and processing device based on a business scenario disclosed in an embodiment of the present invention;

[0085] Figure 4 This is a schematic diagram of another device for information recognition and processing based on a business scenario disclosed in an embodiment of the present invention;

[0086] Figure 5 This is a schematic diagram of the structure of another information recognition and processing device based on a business scenario disclosed in an embodiment of the present invention. Detailed Implementation

[0087] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0088] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0089] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0090] This invention discloses a method and apparatus for information recognition and processing based on business scenarios. It can recognize and process information in various business scenarios, which helps improve the accuracy and reliability of information recognition and processing, as well as its intelligence and efficiency. These will be described in detail below.

[0091] Example 1

[0092] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for information identification and processing based on a business scenario, as disclosed in an embodiment of the present invention. Figure 1 The described method for information recognition and processing based on business scenarios can be applied to a device for information recognition and processing based on business scenarios. This device can be integrated into a cloud server or a local server; this embodiment of the invention does not impose any limitations. Figure 1As shown, the information recognition and processing method based on business scenarios may include the following operations.

[0093] 101. Obtain the information to be identified, and perform dual-path parallel processing on the information to be identified based on the pre-determined first analysis model and the pre-determined second analysis model to obtain preliminary processing results.

[0094] In this embodiment of the invention, optionally, the information to be recognized may include directly received text input information and received user voice information, which is then converted into text by ASR. ASR technology, or Automatic Speech Recognition, is a technology that converts human speech signals into text or text information. It belongs to the speech processing branch of the field of artificial intelligence, realizes the automatic conversion between speech and text, and provides a convenient entry point for human-computer interaction.

[0095] In this embodiment of the invention, optionally, the pre-determined first analysis model may include a Large Language Model (LLM model), wherein a Large Language Model (LLM) is a deep learning model trained on massive amounts of text data that is capable of understanding, generating and processing human language.

[0096] In this embodiment of the invention, optionally, the pre-determined second analysis model may include a traditional lightweight classification model, wherein the traditional lightweight classification model may include one of the TextCNN model, BiLSTM model, or BERT model, used to quickly classify the current input and output a main intent category and its probability distribution. Specifically, the TextCNN model is a text convolutional neural network, which is a model that transfers convolutional neural networks (CNNs) to text processing; its core is to extract local features from the text using convolution operations. BiLSTM (Bidirectional Long Short-Term Memory Network) is an improved version of recurrent neural networks (RNNs), solving the gradient vanishing problem of RNNs through "bidirectional" + "gating mechanism," and better capturing the temporal dependencies of the text. The BERT model is a pre-trained model based on Transformer, which achieves a deep understanding of the global semantics of text through a "bidirectional pre-training + fine-tuning" mode, and is a milestone model in the field of NLP.

[0097] In this embodiment of the invention, optionally, the dual-path parallel processing can be that the first analysis model and the second analysis model perform synchronous processing operations on the information to be identified, thereby obtaining preliminary processing results.

[0098] In this embodiment of the invention, optionally, the preliminary processing results may include the intent candidates and their confidence scores output by the first analysis model LLM model, and the main intent categories and their probability distributions output by the second analysis model.

[0099] 102. Extract the target key information corresponding to the preliminary processing results, and generate target completion information based on the obtained historical interaction information and target key information.

[0100] In this embodiment of the invention, the target key information may optionally include key information extracted from the user statement obtained from the preliminary processing results. For example, the key information may include one or more of the following: date information, account information, amount information, and address information.

[0101] In this embodiment of the invention, the historical interaction information can optionally be obtained through Dialog State Tracking (DST) technology. Dialog State Tracking (DST) is one of the core technologies of dialogue systems (such as intelligent customer service and voice assistants), which is responsible for capturing and updating key information in the dialogue process in real time, and providing a basis for system decision-making (such as generating responses and guiding dialogue).

[0102] In this embodiment of the invention, optionally, the target completion information is obtained by performing a completion operation on the key information of the target based on historical interaction information.

[0103] 103. Based on the target completion information, determine the target intent information and the target entity information corresponding to the information to be identified. Generate target association information based on the target intent information and the target entity information. Based on the target association information and the predefined business process template, determine the target operation instructions and execute the control operations that match the target operation instructions to obtain control feedback results.

[0104] In this embodiment of the invention, optionally, the above-mentioned determination of the target intent information corresponding to the information to be identified and the target entity information corresponding to the information to be identified based on the target completion information may include:

[0105] Based on the target completion information, the intent information corresponding to the target completion information is determined, and based on the target completion information, the target entity information corresponding to the information to be identified is obtained by performing information extraction operation on the target completion information through a pre-determined entity extraction module.

[0106] Among them, the pre-determined entity extraction module, also known as named entity recognition (NER), is one of the core modules in natural language processing (NLP). It is responsible for automatically identifying and extracting entities with specific meanings (such as people's names, place names, organization names, time, events, product names, etc.) from unstructured text. Its goal is to "structure" the key information in the text, providing a foundation for subsequent tasks such as information retrieval, knowledge graph construction, and question answering systems.

[0107] In this embodiment of the invention, optionally, the generation of target association information based on target intent information and target entity information may include:

[0108] Perform information fusion operations on target intent information and target entity information to generate target-related information.

[0109] In this embodiment of the invention, optionally, the target operation instruction may include one or more of the following operations: CRM update, appointment submission, and work order creation. CRM update can be defined as follows: CRM (Customer Relationship Management) is a core tool used by enterprises to manage customer information, interaction records, and transaction data. CRM update refers to the operation of adding, modifying, supplementing, or marking customer-related data in the CRM system, with the aim of maintaining the accuracy and completeness of customer information. Appointment submission is a request initiated by a customer or enterprise employee to agree on a specific business scenario (such as service, meeting, or experience). A work order is a standardized document used to record, track, and process customer issues or internal tasks. Work order creation refers to initiating the recording of a pending task.

[0110] In this embodiment of the invention, optionally, the control feedback result can be the operation feedback result corresponding to the target operation instruction, obtained in real time. Further, the control feedback result can include the result returned by the backend business system after executing the target operation instruction, such as work order number, appointment number, confirmation information, etc.

[0111] 104. Based on the control feedback results, generate the information recognition results corresponding to the information to be identified.

[0112] In this embodiment of the invention, optionally, the information recognition result can be the final feedback information generated based on the control feedback result. Further, the information recognition result can be synthesized into speech feedback to the user through a text-to-speech (TTS) module, or directly displayed to the user in text form.

[0113] It is evident that implementation Figure 1The described information recognition and processing method based on business scenarios can acquire the information to be identified and perform parallel processing based on a first analysis model and a second analysis model to obtain preliminary processing results and extract corresponding target key information. Based on historical interaction information and target key information, target completion information is generated to determine the corresponding target intent information and target entity information, thereby generating target association information. Based on target association information and predefined business process templates, target operation instructions are determined and corresponding control feedback results are obtained to generate information recognition results. Through the parallel processing and fusion strategy of the first and second analysis models, the method can fully utilize the semantic reasoning ability of large models and the rapid classification ability of traditional models, significantly improving the accuracy of intent recognition and entity extraction. Based on intent information and entity information, the accuracy of key information extraction is further improved. By combining target completion information with historical interaction information, logical coherence is maintained, ensuring the completeness and accuracy of the information required in the automatic execution process. Thus, information can be identified and processed in various business scenarios, which is conducive to improving the accuracy and reliability of information recognition and processing, as well as the intelligence and efficiency of information recognition and processing.

[0114] Example 2

[0115] Please see Figure 2 , Figure 2 This is a flowchart illustrating another method for information identification and processing based on a business scenario disclosed in an embodiment of the present invention. Figure 2 The described method for information recognition and processing based on business scenarios can be applied to a device for information recognition and processing based on business scenarios. This device can be integrated into a cloud server or a local server; this embodiment of the invention does not impose any limitations. Figure 2 As shown, this information recognition and processing method based on business scenarios may include the following operations:

[0116] 201. Obtain the information to be identified. Based on the pre-determined first analysis model and the pre-determined second analysis model, perform dual-path parallel processing on the information to be identified to obtain preliminary processing results.

[0117] 202. Extract the target key information corresponding to the preliminary processing results, and generate target completion information based on the obtained historical interaction information and target key information.

[0118] 203. Based on the target completion information, determine the target intent information and the target entity information corresponding to the information to be identified. Generate target association information based on the target intent information and the target entity information. Based on the target association information and the predefined business process template, determine the target operation instructions and execute the control operations that match the target operation instructions to obtain control feedback results.

[0119] 204. Based on the control feedback results, generate the information recognition results corresponding to the information to be identified.

[0120] In this embodiment of the invention, for a detailed description of steps 201-204, please refer to the other descriptions of steps 101-104 in Embodiment 1. This embodiment of the invention will not repeat them.

[0121] 205. Based on the information recognition results, generate structured text information, and perform speech synthesis operation on the structured text information through a pre-determined speech conversion module to obtain the synthesized output speech.

[0122] In this embodiment of the invention, optionally, the generation of structured text information based on the information recognition result may include:

[0123] Based on the information recognition results, the corresponding recognition text information is determined and converted into a structured format to generate structured text information; the structured format may include JSON format.

[0124] In this embodiment of the invention, optionally, the predetermined speech conversion module may include a TTS module, wherein the TTS module, or Text-to-Speech module, is a technical component that converts text information into natural and fluent speech signals.

[0125] In this embodiment of the invention, optionally, the above-mentioned process of performing speech synthesis on structured text information through a pre-determined speech conversion module to obtain synthesized output speech may include:

[0126] The structured text information is input into a pre-determined speech conversion module, which performs speech synthesis on the structured text information to convert it into speech information and obtain synthesized output speech.

[0127] 206. Obtain user information corresponding to the voice receiving user, determine voice receiving parameters based on the user information, and perform voice adjustment operation on the synthesized output voice based on the voice receiving parameters to obtain the target output voice, and feed back the target output voice to the user terminal corresponding to the voice receiving user.

[0128] In this embodiment of the invention, optionally, the user information corresponding to the voice receiving user may include user preference settings, historical interaction records, device information, etc., wherein the user preference settings may include voice style (such as formal, friendly, lively), speech rate (fast, medium, slow), and tone (high, medium, low); historical interaction records may include the user's historical interaction methods; device information may include one or more of the following: the user terminal model, operating system, and audio output device.

[0129] In this embodiment of the invention, optionally, the above-mentioned determination of voice reception parameters based on user information may include: determining user voice preference information based on user information, and determining voice reception parameters based on voice preference information. Further, the voice reception parameters may include one or more of voice style parameters, voice rate parameters, and voice volume parameters.

[0130] In this embodiment of the invention, optionally, the above-mentioned process of performing voice adjustment operations on the synthesized output voice based on voice reception parameters to obtain the target output voice may include: performing adjustment operations on the synthesized output voice that match the voice reception parameters to obtain the target output voice. For example, if the user prefers a lively voice, the intonation and rhythm of the voice are adjusted to make it more lively.

[0131] In this embodiment of the invention, optionally, feeding back the target output voice to the user terminal corresponding to the voice receiving user can be done by sending the target output voice to the user terminal, so that the user can hear the voice information fed back by the system through the device's audio output device.

[0132] It is evident that implementation Figure 2The described information recognition and processing method based on business scenarios can generate structured text information based on the information recognition results, and perform speech synthesis through a speech conversion module to obtain synthesized output speech. It acquires user information corresponding to the voice receiving user and determines the voice receiving parameters. Based on the voice receiving parameters, it performs voice adjustment operations on the synthesized output speech to obtain the target output speech, and feeds the target output speech back to the corresponding user terminal. By acquiring user preference settings and device information, it can generate voice feedback that meets the user's personalized needs, which is beneficial to improving the user's interactive experience. Furthermore, by adjusting the voice receiving parameters, it can further optimize the naturalness and comprehensibility of the speech, enhancing user trust and satisfaction. It can quickly generate and provide feedback on voice information, ensuring that users can receive timely feedback from the system, improving interaction efficiency. It can also generate suitable voice feedback based on different user information, adapting to the needs of users of different ages, hearing abilities, and device environments. Furthermore, it can analyze user preferences and device information, dynamically adjusting voice parameters, which helps improve the intelligence and adaptability of information recognition. Through the process of voice synthesis and adjustment, it can not only provide accurate information feedback but also optimize user experience and improve the overall level of intelligence. In turn, it can recognize and process information in various business scenarios, which helps improve the accuracy and reliability of information recognition and processing, as well as the intelligence and efficiency of information recognition and processing.

[0133] In an optional embodiment, based on a pre-determined first analysis model and a pre-determined second analysis model, a dual-path parallel processing operation is performed on the information to be identified to obtain preliminary processing results, including:

[0134] The information to be identified is input into a pre-determined first analysis model, and a first analysis operation is performed on the information to be identified through the pre-determined first analysis model to obtain a first output result. The first output result includes the intent output information corresponding to the information to be identified and the confidence parameter corresponding to each intent output information.

[0135] The information to be identified is input into a pre-determined second analysis model, and a second analysis operation is performed on the information to be identified through the pre-determined second analysis model to obtain a second output result. The second output result includes the information intent category corresponding to the information to be identified and the probability parameter corresponding to each information intent category.

[0136] Normalization is performed on the first and second output results to obtain the normalized result. Based on the normalized result, the preliminary processing result corresponding to the information to be identified is generated.

[0137] In this optional embodiment, the pre-determined first analysis model may include a Large Language Model (LLM model), wherein a Large Language Model (LLM) is a deep learning model trained on massive amounts of text data that can understand, generate and process human language; the pre-determined second analysis model may include a traditional lightweight classification model, wherein a traditional lightweight classification model may include one of the TextCNN model, BiLSTM model or BERT model.

[0138] In this optional embodiment, optionally, for example, the intent output information corresponding to the information to be identified may include a list of intent candidates output by the LLM model, such as querying account balance, transferring funds, or querying transaction records, or one or more of these. The probability parameter corresponding to each intent category may include a confidence score corresponding to each intent candidate, for example, [0.85, 0.10, 0.05]. The confidence score represents the model's confidence in the intent and is typically a value between 0 and 1.

[0139] In this optional embodiment, optionally, for example, the information intent category corresponding to the information to be identified may include one or more of query category, search category, and analysis category; the probability parameter corresponding to each information intent category may include the probability distribution of that intent category, for example, [0.90]. The probability parameter represents the model's predicted probability for that intent category.

[0140] In this optional embodiment, the normalization process may optionally include normalizing the confidence and probability parameters in the first and second outputs to ensure they are on the same scale. For example, the confidence score of the LLM and the probability value of the traditional model may both be mapped to the [0, 1] interval.

[0141] In this optional embodiment, the input of the information to be identified into a predetermined first analysis model and the input of the information to be identified into a predetermined second analysis model can be parallel processing operations performed synchronously.

[0142] In this optional embodiment, the preliminary processing result may optionally include a comprehensive processing result generated by combining the normalized first output result and the second output result.

[0143] As can be seen, implementing this optional embodiment allows the information to be identified to be input into the first analysis model and the second analysis model to perform the first analysis operation and the second analysis operation respectively, thereby obtaining the first output result and the second output result. Normalization processing is then performed on the first output result and the second output result to obtain the normalized result. Based on the normalized result, a preliminary processing result corresponding to the information to be identified is generated. This allows for parallel preliminary operations through the first analysis model and the second analysis model, enabling rapid and efficient output of intent classification results, thus improving the accuracy and reliability of user intent identification. Furthermore, the normalization processing unifies the output results of different models to the same scale, avoiding discrepancies caused by different models. The deviation caused by different model output ranges further improves the accuracy of the fusion results. The parallel execution of analysis operations by the first and second analysis models helps to improve response speed and efficiency, which can meet the application scenarios with high real-time requirements. Furthermore, through dual-path parallel processing and normalized fusion, the user's intent can be identified more accurately, which helps to improve user satisfaction and convenience. The parallel processing architecture ensures that the user input can be responded to quickly, improving the interactive experience and efficiency. In turn, it can identify and process information in various business scenarios, which helps to improve the accuracy and reliability of information identification and processing, as well as the intelligence and efficiency of information identification and processing.

[0144] In another optional embodiment, the target key information corresponding to the preliminary processing result is extracted, including:

[0145] Preprocessing operations are performed on the preliminary processing results to obtain preprocessed results. The preprocessing operations include one or more of the following: word segmentation, encoding, punctuation correction, and text cleaning.

[0146] Historical semantic information is acquired, and a target extraction model is constructed based on the historical semantic information. The historical semantic information includes one or more of the following: historical background context information, historical intent information, and historical context information.

[0147] Based on the target extraction model, an extraction operation is performed on the preliminary processing results to obtain structured extraction results, and based on the structured extraction results, the target key information corresponding to the preliminary processing results is generated.

[0148] In this optional embodiment, the preliminary processing results may include the results obtained by performing dual-path parallel processing operations through the first analysis model and the second analysis model, including intent candidates and their confidence (from the first analysis model) and intent categories and their probabilities (from the second analysis model).

[0149] In this optional embodiment, the word segmentation operation may include dividing the text into words, phrases, or clauses for easier subsequent processing; the encoding operation may include converting the text into an encoding format suitable for model processing, such as Unicode or word embedding vectors; the punctuation correction operation may include correcting punctuation errors in the text; and the text cleaning operation may include removing noise from the text, such as extra spaces, special characters, etc.

[0150] In this optional embodiment, the historical context information may include background information of the user's interaction with the system, such as the user's scene and time; the historical intent information may include the user's intention information in the interaction with the system; and the historical context information may include the historical dialogue content of the user's interaction with the system.

[0151] In this optional embodiment, the above-mentioned construction of the target extraction model based on historical semantic information may include: determining historical semantic keywords based on historical semantic information, and constructing the target extraction model based on historical semantic keywords.

[0152] In this optional embodiment, the target extraction model can be a rule-based extraction model or a machine learning or deep learning model, such as fine-tuning BERT or other pre-trained language models.

[0153] In this optional embodiment, optionally, for example, a Prompt containing background context, user intent, and historical context is constructed and input into the LLM model. The LLM model performs zero-shot or few-shot inference through natural language prompts (such as "Please extract the time, amount, and address information involved in the above utterance") and outputs structured results, such as a list of fields in JSON format. Here, the Prompt refers to the text instructions or information that the user inputs to the AI ​​model (such as the Large Language Model LLM, image generation model, etc.) to guide the model to generate the expected output.

[0154] In this optional embodiment, the structured extraction result may optionally include key information extracted by the target extraction model from the preprocessing result, presented in a structured format. Further optionally, based on the structured extraction result, the target key information corresponding to the preliminary processing result may include final key information generated based on the structured extraction result, for use in subsequent business process processing.

[0155] As can be seen, implementing this optional embodiment can perform preprocessing operations on the preliminary processing results to obtain preprocessed results, acquire historical semantic information to construct a target extraction model, and perform extraction operations on the preliminary processing results based on the target extraction model to obtain structured extraction results, thereby generating corresponding target key information. Preprocessing operations can standardize text format, providing clearer input data for subsequent information extraction operations, which is beneficial to improving the accuracy and reliability of information extraction. Furthermore, by combining historical background, intent, and contextual information, the target extraction model can better understand the semantics of the current input, thus facilitating more accurate extraction of key information and improving the accuracy and reliability of extracted information. Introducing historical semantic information allows for a better understanding of user intent and needs, improving user experience and convenience, and enabling more precise extraction of the key information required by the user, reducing the amount of additional information the user needs to provide, improving interaction efficiency and convenience. Ultimately, this allows for information recognition and processing in various business scenarios, improving the accuracy and reliability of information recognition and processing, as well as enhancing the intelligence and efficiency of information recognition and processing.

[0156] In another optional embodiment, target completion information is generated based on the acquired historical interaction information and target key information, including:

[0157] Obtain historical interaction information, and based on the obtained historical interaction information, determine the missing target information corresponding to the target key information;

[0158] Based on the missing target information, completeness guidance parameters are generated and transmitted to the target user terminal corresponding to the information to be identified, and completeness feedback information corresponding to the target user terminal is obtained.

[0159] Based on the feedback information, perform information completion operations on the key information of the target to generate target completion information.

[0160] In this optional embodiment, the historical interaction information may include the user's previous dialogue history with the system, including user input, system feedback, intent recognition results, entity extraction results, etc.

[0161] In this optional embodiment, the process of determining the target missing information corresponding to the target key information based on the acquired historical interaction information may include: performing a missing information detection operation on the target key information based on the acquired historical interaction information and a pre-determined business information template to obtain the target missing information corresponding to the target key information. For example, by checking for missing slots in the target key information using a predefined business process template, and by comparing the target key information with the template, the missing slots are identified to determine the target missing information.

[0162] In this optional embodiment, the completion guidance parameters may optionally include parameters generated based on the target missing information to guide the user in completing the missing information. For example, the completion guidance text may include a system-generated natural language prompt to guide the user in completing the missing information. For example: "To check your account balance, please provide your password"; the completion guidance voice may include: if the system supports voice interaction, the completion guidance text can be converted into voice and sent to the user terminal to obtain feedback from the user terminal and obtain the corresponding supplementary information.

[0163] In this optional embodiment, the acquisition of the completion feedback information corresponding to the target user terminal can be achieved by the system receiving the user's completion feedback information through the user terminal, for example, by converting the user's voice input into text through the speech recognition (ASR) module, or by directly receiving the text input by the user, and thus determining the completion feedback information based on the acquired information.

[0164] In this optional embodiment, further optionally, for example, the missing information detection (Slot FillingCheck) can compare the information extracted from the user's current input by the entity extraction module with the template requirements; and mark unfilled or low-confidence slots as "to be supplemented"; supporting multi-round cumulative judgment—if the user has provided information in the previous round, the corresponding slot will be automatically filled. Further, completion prompt generation is performed through completion guidance parameters, and for slots to be supplemented, the system dynamically generates natural language completion prompts. This method can include template-based completion prompts, including automatically generating standard prompts based on slot definitions; and context-enhanced completion prompts (LLM-assisted), which can include using a large language model (LLM) combined with context to generate more natural, gentler-toned, and personalized questions to obtain completion feedback information. Even further, the user's supplementary response is reprocessed through the intent recognition and entity extraction modules; the extracted entity information is filled into the corresponding slots to generate the target completion information.

[0165] In this optional embodiment, if the user fails to effectively supplement the required information (such as giving vague answers or skipping multiple times) after being guided by the completion guidance parameters, multiple rounds of supplementary questions or alternative ways of expression are provided. If the user still fails, the system prompts the user to transfer to human assistance or enter the error-tolerant process. The system records the user's current progress status and supports continuing later.

[0166] As can be seen, implementing this optional embodiment can obtain historical interaction information to determine the missing target information corresponding to the target key information, generate completion guidance parameters based on the missing target information and transmit them to the target user terminal, obtain the completion feedback information corresponding to the target user terminal, and perform information completion operation on the target key information according to the completion feedback information to generate target completed information. It can automatically detect the missing parts in the target key information and generate completion guidance parameters, ensuring the integrity and reliability of the information provided by the user, and can gradually guide the user to supplement the missing information, improving the integrity and accuracy of the information. The automatic completion mechanism reduces the need for manual intervention, improves the automation and intelligence of information completion, enhances the user experience and convenience, can quickly detect missing information and generate completion guidance parameters, which is conducive to improving business processing efficiency and processing intelligence. It can also dynamically generate completion guidance parameters according to different business process templates, adapting to various business scenarios and user needs, thereby improving interaction efficiency and interaction convenience, and enabling information recognition and processing in various business scenarios, which is conducive to improving the accuracy and reliability of information recognition and processing, as well as improving the intelligence and efficiency of information recognition and processing.

[0167] In another optional embodiment, the target operation instruction is determined based on the target association information and a predefined business process template, including:

[0168] Based on the target association information and the predefined business process template, a mapping relationship between the target association information and the predefined business process template is generated.

[0169] Based on the mapping relationship, assemble the backend call instructions, and generate the target operation instructions based on all the backend call instructions.

[0170] In this optional embodiment, the target association information may include structured information formed by combining user intent and entity information, which is used to indicate the specific operation that the system needs to perform.

[0171] In this optional embodiment, the predefined business process template may optionally include a predefined business operation blueprint, which describes how to map user intent and entity information to specific backend operations. Further, the predefined business process template may include one or more of the following: appointment process template, fault reporting process template, and lost / found process template.

[0172] In this optional embodiment, the mapping relationship between the target association information and the predefined business process template may include matching the intent and entity information in the target association information with the fields in the business process template to generate a mapping relationship.

[0173] In this optional embodiment, the process of assembling backend call instructions according to the mapping relationship and generating target operation instructions based on all backend call instructions may include:

[0174] Based on the mapping relationship, at least one target call instruction matching the mapping relationship is identified from a pre-determined set of call instructions. An assembly operation is then performed on all target call instructions to obtain backend call instructions. Based on all backend call instructions, a target operation instruction is generated. For example, based on the mapping relationship, the system assembles specific backend call instructions, including the API endpoint, method, parameters, etc.; and integrates all backend call instructions into a single target operation instruction for submission to the backend system for execution. Furthermore, the target operation instruction may include one or more of the following: CRM update instructions, appointment submission instructions, and work order creation instructions.

[0175] As can be seen, implementing this optional embodiment can generate a mapping relationship between target association information and predefined business process templates based on target association information and predefined business process templates. It assembles backend call instructions and generates target operation instructions based on this mapping relationship. This automatically maps user intent and entity information to predefined business process templates, which is beneficial for improving the automation and intelligence of business processing, as well as its efficiency and convenience. Furthermore, the predefined business process templates can dynamically adapt to different business needs, improving the adaptability and matching degree of business scenarios. The structured design based on target association information and business process templates can more accurately generate backend call instructions, reducing business processing failures caused by human error and improving the generation of target operation instructions. The accuracy and reliability of commands, as well as the intelligence and efficiency of generating target operation commands, are improved, thereby enhancing the reliability of business processing. Through automated mapping and command assembly processes, user satisfaction and ease of use are improved. The generation of mapping relationships, the assembly of backend call commands, and the generation of target operation commands can be independently optimized and adjusted. It also facilitates the expansion and maintenance of business scenario applications and can dynamically generate backend call commands according to different business process templates, adapting to various business scenarios and user needs, thereby improving interaction efficiency and convenience. Furthermore, it enables information recognition and processing in various business scenarios, improving the accuracy and reliability of information recognition and processing, as well as enhancing the intelligence and efficiency of information recognition and processing.

[0176] In another optional embodiment, normalization processing is performed on the first output result and the second output result to obtain a normalized result, and based on the normalized result, a preliminary processing result corresponding to the information to be identified is generated, including:

[0177] Normalization is performed on the first output and the second output to obtain the normalized result, which includes the confidence scores corresponding to the first output and the second output.

[0178] Based on the normalization processing result, determine the result weight parameters corresponding to the first output result and the second output result, determine the weighted average confidence of the first output result and the second output result based on the result weight parameters, and determine whether the weighted average confidence is greater than or equal to the preset confidence threshold.

[0179] When it is determined that the weighted average confidence level is greater than or equal to the preset confidence level threshold, the preliminary processing result corresponding to the information to be identified is generated based on the normalization processing result.

[0180] When it is determined that the weighted average confidence level is less than the preset confidence level threshold, the first semantic information corresponding to the first output result and the second semantic information corresponding to the second output result are determined, and it is determined whether the first semantic information and the second semantic information match.

[0181] When it is determined that the first semantic information matches the second semantic information, a preliminary processing result corresponding to the information to be identified is generated based on the first semantic information and the second semantic information.

[0182] When it is determined that the first semantic information does not match the second semantic information, supplementary description information corresponding to the information to be identified is obtained, and preliminary processing results corresponding to the information to be identified are generated based on the supplementary description information and the information to be identified.

[0183] In this optional embodiment, the normalization process may optionally include normalizing the confidence and probability parameters in the first and second outputs to ensure they are on the same scale. For example, the confidence score of the LLM and the probability value of the traditional model may both be mapped to the [0, 1] interval.

[0184] In this optional embodiment, the above-mentioned determination of the result weight parameters corresponding to the first output result and the second output result based on the normalization processing result, and the determination of the weighted average confidence of the first output result and the second output result based on the result weight parameters, may include:

[0185] Based on the normalization processing result, determine the weight values ​​corresponding to the first output result and the second output result, so as to determine the result weight parameters corresponding to the first output result and the second output result, wherein the result weight parameters include weight values;

[0186] Based on the result weight parameters, calculate the weighted average confidence level of each intent corresponding to the first output result and each intent corresponding to the second output result.

[0187] In this optional embodiment, the process of generating the preliminary processing result corresponding to the information to be identified based on the normalization processing result may include: determining the highest confidence level from the weighted average confidence level corresponding to the first output result and the weighted average confidence level corresponding to the second output result based on the normalization processing result, and determining the intent corresponding to the highest confidence level as the preliminary processing result corresponding to the information to be identified. That is, if the intent corresponding to the first output result is inconsistent with the intent corresponding to the second output result, the result with a confidence level higher than a set threshold is taken as the final intent; or, the result with a higher confidence level is taken as the final intent.

[0188] In this optional embodiment, further optionally, after determining the weighted average confidence level of the first output result and the second output result based on the result weight parameter, the method may further include:

[0189] Determine whether the output intent corresponding to the first output result is the same as the output intent corresponding to the second output result. If the output intent corresponding to the first output result is the same as the output intent corresponding to the second output result, determine the output intent corresponding to the first output result as the preliminary processing result corresponding to the information to be identified, or determine the output intent corresponding to the second output result as the preliminary processing result corresponding to the information to be identified. If the output intent corresponding to the first output result is different from the output intent corresponding to the second output result, trigger the step of determining whether the weighted average confidence is greater than or equal to a preset confidence threshold.

[0190] In this optional embodiment, the aforementioned preliminary processing result for generating the information to be identified based on the first semantic information and the second semantic information may include: performing a semantic information fusion operation based on the first semantic information and the second semantic information to obtain a semantic fusion result, and generating the preliminary processing result for the information to be identified based on the semantic fusion result. That is, if the semantics of the first semantic information and the second semantic information are similar, they can be fused and regarded as the same intent category.

[0191] In this optional embodiment, the process of obtaining supplementary descriptive information corresponding to the information to be identified and generating a preliminary processing result based on the supplementary descriptive information and the information to be identified may include: obtaining supplementary descriptive information corresponding to the information to be identified, and performing semantic supplementation processing on the information to be identified based on the supplementary descriptive information to generate a preliminary processing result. For example, the supplementary descriptive information may include prompts for the user to provide more information to improve the accuracy of identification; for example, the system prompts: "To better understand your needs, please provide more contextual information"; and obtains corresponding feedback information, such as the user's feedback "I want to check my account balance to confirm recent transactions," and combines the supplementary descriptive information and the original information to be identified to re-perform intent identification and confidence calculation to generate the final preliminary processing result.

[0192] In this optional embodiment, the analysis results corresponding to the first analysis model and the analysis results corresponding to the second analysis model can be recorded, that is, the outputs and fusion process of LLM and traditional model can be recorded for subsequent model optimization and anomaly identification.

[0193] As can be seen, implementing this optional embodiment can perform normalization processing on the first output result and the second output result to obtain a normalized processing result, determine the result weight parameters corresponding to the first output result and the second output result, and determine the corresponding weighted average confidence level. If it is determined that the weighted average confidence level is greater than or equal to the preset confidence level threshold, a preliminary processing result corresponding to the information to be identified is generated based on the normalization processing result. If it is less than the preset confidence level threshold, it is determined whether the first semantic information and the second semantic information match. If they match, a preliminary processing result corresponding to the information to be identified is generated. If they do not match, supplementary description information is obtained, and then combined with the supplementary description information and the information to be identified to generate the corresponding preliminary processing result. By setting a confidence level threshold, the accuracy and reliability of intent recognition can be improved. When the weighted average confidence level is lower than the threshold, the linguistic performance of the two models is compared. Whether semantic information matches can further verify the reliability of the results. When the outputs of two models are inconsistent, the intent can be re-evaluated by obtaining supplementary descriptive information from the user, which helps improve the accuracy and reliability of intent evaluation. It can also dynamically adjust the interaction strategy based on confidence and semantic matching, which helps improve the intelligence and convenience of interaction with users and improve the user experience. Furthermore, it can dynamically adjust weight parameters and confidence thresholds according to different business scenarios and user needs, adapting to various application scenarios. It also helps to expand and maintain business scenario applications, adapting to various business scenarios and user needs, thereby improving interaction efficiency and convenience. In turn, it can identify and process information in various business scenarios, which helps improve the accuracy and reliability of information identification and processing, as well as the intelligence and efficiency of information identification and processing.

[0194] Example 3

[0195] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of an information recognition and processing device based on a business scenario, as disclosed in an embodiment of the present invention. Figure 3 As shown, the device for information recognition and processing based on business scenarios may include:

[0196] The acquisition module 301 is used to acquire the information to be identified;

[0197] The processing module 302 is used to perform dual-path parallel processing operations on the information to be identified based on a predetermined first analysis model and a predetermined second analysis model to obtain preliminary processing results;

[0198] Extraction module 303 is used to extract key target information corresponding to the preliminary processing results;

[0199] The generation module 304 is used to generate target completion information based on the acquired historical interaction information and target key information;

[0200] The determination module 305 is used to determine the target intent information and the target entity information corresponding to the information to be identified based on the target completion information;

[0201] The generation module 304 is also used to generate target association information based on the target intent information and the target entity information;

[0202] The determination module 305 is also used to determine the target operation instruction based on the target association information and the predefined business process template;

[0203] Execution module 306 is used to execute control operations that match the target operation instructions and obtain control feedback results;

[0204] The generation module 304 is also used to generate information recognition results corresponding to the information to be recognized based on the control feedback results.

[0205] It is evident that implementation Figure 3 The described device can acquire information to be identified and perform parallel processing based on a first analysis model and a second analysis model to obtain preliminary processing results and extract corresponding target key information. Based on historical interaction information and target key information, it generates target completion information to determine the corresponding target intent information and target entity information, thereby generating target association information. Based on target association information and a predefined business process template, it determines the target operation instruction and obtains the corresponding control feedback results to generate information recognition results. Through the parallel processing and fusion strategy of the first and second analysis models, it can fully utilize the semantic reasoning ability of the large model and the fast classification ability of the traditional model, significantly improving the accuracy of intent recognition and entity extraction. Based on intent information and entity information, it further improves the accuracy of key information extraction. Through target completion information, it can maintain logical coherence by combining historical interaction information, ensuring the completeness and accuracy of the information required in the automatic execution process. Thus, it can identify and process information in various business scenarios, which is conducive to improving the accuracy and reliability of information recognition and processing, as well as improving the intelligence and efficiency of information recognition and processing.

[0206] In an optional embodiment, such as Figure 4 As shown, the generation module 304 is also used to generate structured text information based on the information recognition results;

[0207] The device also includes:

[0208] The synthesis module 307 is used to perform speech synthesis operations on structured text information through a pre-determined speech conversion module to obtain synthesized output speech;

[0209] The acquisition module 301 is also used to acquire user information corresponding to the voice receiving user;

[0210] The determination module 305 is also used to determine voice reception parameters based on user information;

[0211] The adjustment module 308 is used to perform speech adjustment operations on the synthesized output speech based on the speech reception parameters to obtain the target output speech;

[0212] Feedback module 309 is used to feed back the target output voice to the user terminal corresponding to the voice receiving user.

[0213] It is evident that implementation Figure 4 The described device can generate structured text information based on information recognition results and perform speech synthesis through a speech conversion module to obtain synthesized output speech. It acquires user information corresponding to the voice receiving user and determines voice receiving parameters. Based on these parameters, it performs voice adjustment operations on the synthesized output speech to obtain the target output speech and feeds it back to the corresponding user terminal. By acquiring user preference settings and device information, it can generate voice feedback that meets the user's personalized needs, improving the user's interactive experience. Furthermore, by adjusting the voice receiving parameters, it further optimizes the naturalness and comprehensibility of the speech, enhancing user trust and satisfaction. It can also quickly generate and reflect... The system provides voice feedback to ensure users receive timely feedback, improving interaction efficiency. It can also generate suitable voice feedback based on different user information, adapting to the needs of users of different ages, hearing abilities, and device environments. Furthermore, it can analyze user preferences and device information, dynamically adjusting voice parameters to enhance the intelligence and adaptability of information recognition. Through the voice synthesis and adjustment process, it not only provides accurate information feedback but also optimizes the user experience, improving the overall level of intelligence. This enables information recognition and processing in various business scenarios, improving the accuracy and reliability of information recognition and processing, as well as enhancing its intelligence and efficiency.

[0214] In another alternative embodiment, such as Figure 4 As shown, the processing module 302 performs dual-path parallel processing on the information to be identified based on a pre-determined first analysis model and a pre-determined second analysis model to obtain preliminary processing results. The specific methods include:

[0215] The information to be identified is input into a pre-determined first analysis model, and a first analysis operation is performed on the information to be identified through the pre-determined first analysis model to obtain a first output result. The first output result includes the intent output information corresponding to the information to be identified and the confidence parameter corresponding to each intent output information.

[0216] The information to be identified is input into a pre-determined second analysis model, and a second analysis operation is performed on the information to be identified through the pre-determined second analysis model to obtain a second output result. The second output result includes the information intent category corresponding to the information to be identified and the probability parameter corresponding to each information intent category.

[0217] Normalization is performed on the first and second output results to obtain the normalized result. Based on the normalized result, the preliminary processing result corresponding to the information to be identified is generated.

[0218] It is evident that implementation Figure 4 The described apparatus can input information to be identified into a first analysis model and a second analysis model to perform a first analysis operation and a second analysis operation respectively, thereby obtaining a first output result and a second output result. Normalization processing is performed on the first output result and the second output result to obtain a normalized result. Based on the normalized result, a preliminary processing result corresponding to the information to be identified is generated. It can perform parallel preliminary operations through the first analysis model and the second analysis model, enabling rapid and efficient output of intent classification results, thus improving the accuracy and reliability of user intent identification. Furthermore, through normalization processing, the output results of different models are unified to the same scale, avoiding discrepancies due to differences in model output. The deviation caused by different output ranges further improves the accuracy of the fusion results. The parallel execution of analysis operations by the first and second analysis models helps to improve response speed and efficiency, which can meet the application scenarios with high real-time requirements. Furthermore, through dual-path parallel processing and normalized fusion, the user's intent can be identified more accurately, which helps to improve user satisfaction and convenience. The parallel processing architecture ensures that the user input can be responded to quickly, improving the interactive experience and efficiency. In turn, it can identify and process information in various business scenarios, which helps to improve the accuracy and reliability of information identification and processing, as well as the intelligence and efficiency of information identification and processing.

[0219] In yet another alternative embodiment, such as Figure 4 As shown, the specific methods by which the extraction module 303 extracts the target key information corresponding to the preliminary processing results include:

[0220] Preprocessing operations are performed on the preliminary processing results to obtain preprocessed results. The preprocessing operations include one or more of the following: word segmentation, encoding, punctuation correction, and text cleaning.

[0221] Historical semantic information is acquired, and a target extraction model is constructed based on the historical semantic information. The historical semantic information includes one or more of the following: historical background context information, historical intent information, and historical context information.

[0222] Based on the target extraction model, an extraction operation is performed on the preliminary processing results to obtain structured extraction results, and based on the structured extraction results, the target key information corresponding to the preliminary processing results is generated.

[0223] It is evident that implementation Figure 4 The described device can perform preprocessing operations on the preliminary processing results to obtain preprocessed results, acquire historical semantic information to construct a target extraction model, and perform extraction operations on the preliminary processing results based on the target extraction model to obtain structured extraction results, thereby generating corresponding target key information. Preprocessing operations can standardize text format, providing clearer input data for subsequent information extraction operations, which is beneficial to improving the accuracy and reliability of information extraction. Furthermore, by combining historical background, intent, and contextual information, the target extraction model can better understand the semantics of the current input, thus facilitating more accurate extraction of key information and improving the accuracy and reliability of extracted information. Introducing historical semantic information allows for a better understanding of user intent and needs, improving user experience and convenience, and enabling more precise extraction of the key information required by the user, reducing the amount of additional information the user needs to provide, improving interaction efficiency and convenience. Ultimately, this allows for information recognition and processing in various business scenarios, improving the accuracy and reliability of information recognition and processing, as well as enhancing the intelligence and efficiency of information recognition and processing.

[0224] In yet another alternative embodiment, such as Figure 4 As shown, the specific methods by which the generation module 304 generates target completion information based on the acquired historical interaction information and target key information include:

[0225] Obtain historical interaction information, and based on the obtained historical interaction information, determine the missing target information corresponding to the target key information;

[0226] Based on the missing target information, completeness guidance parameters are generated and transmitted to the target user terminal corresponding to the information to be identified, and completeness feedback information corresponding to the target user terminal is obtained.

[0227] Based on the feedback information, perform information completion operations on the key information of the target to generate target completion information.

[0228] It is evident that implementation Figure 4The described device can acquire historical interaction information to determine the missing target information corresponding to key target information, generate completion guidance parameters based on the missing target information and transmit them to the target user terminal, obtain completion feedback information corresponding to the target user terminal, and perform information completion operations on the key target information according to the completion feedback information to generate target completed information. It can automatically detect missing parts in the key target information and generate completion guidance parameters to ensure the integrity and reliability of the information provided by the user. It can also gradually guide the user to fill in the missing information, improving the integrity and accuracy of the information. The automatic completion mechanism reduces the need for manual intervention, improves the automation and intelligence of information completion, enhances the user experience and convenience, and can quickly detect missing information and generate completion guidance parameters, which is conducive to improving business processing efficiency and intelligence. It can also dynamically generate completion guidance parameters according to different business process templates, adapting to various business scenarios and user needs, thereby improving interaction efficiency and convenience. In turn, it can identify and process information in various business scenarios, which is conducive to improving the accuracy and reliability of information identification and processing, as well as the intelligence and efficiency of information identification and processing.

[0229] In yet another alternative embodiment, such as Figure 4 As shown, the specific methods by which the determining module 305 determines the target operation instruction based on the target association information and the predefined business process template include:

[0230] Based on the target association information and the predefined business process template, a mapping relationship between the target association information and the predefined business process template is generated.

[0231] Based on the mapping relationship, assemble the backend call instructions, and generate the target operation instructions based on all the backend call instructions.

[0232] It is evident that implementation Figure 4The described device can generate a mapping relationship between target association information and predefined business process templates based on target association information and predefined business process templates. It then assembles backend call instructions and generates target operation instructions based on this mapping relationship. This automatically maps user intent and entity information to predefined business process templates, improving the automation and intelligence of business processing, as well as its efficiency and convenience. Furthermore, the predefined business process templates allow for dynamic adaptation to different business needs, enhancing adaptability and matching to business scenarios. The structured design based on target association information and business process templates enables more accurate generation of backend call instructions, reducing business processing failures caused by human error and improving the accuracy of generated target operation instructions. This improves the accuracy and reliability of information processing, enhances the intelligence and efficiency of generating target operation instructions, and thus improves the reliability of business processing. The automated mapping and instruction assembly process also enhances user satisfaction and ease of use. The generation of mapping relationships, the assembly of backend call instructions, and the generation of target operation instructions can be independently optimized and adjusted. Furthermore, it facilitates the expansion and maintenance of business application scenarios and allows for the dynamic generation of backend call instructions based on different business process templates, adapting to various business scenarios and user needs. This improves interaction efficiency and convenience, enabling information recognition and processing in various business scenarios, thereby enhancing the accuracy and reliability of information recognition and processing, as well as its intelligence and efficiency.

[0233] In yet another alternative embodiment, such as Figure 4 As shown, the processing module 302 performs normalization processing on the first output result and the second output result to obtain the normalized processing result, and generates the preliminary processing result corresponding to the information to be identified based on the normalized processing result in the following specific ways:

[0234] Normalization is performed on the first output and the second output to obtain the normalized result, which includes the confidence scores corresponding to the first output and the second output.

[0235] Based on the normalization processing result, determine the result weight parameters corresponding to the first output result and the second output result, determine the weighted average confidence of the first output result and the second output result based on the result weight parameters, and determine whether the weighted average confidence is greater than or equal to the preset confidence threshold.

[0236] When it is determined that the weighted average confidence level is greater than or equal to the preset confidence level threshold, the preliminary processing result corresponding to the information to be identified is generated based on the normalization processing result.

[0237] When it is determined that the weighted average confidence level is less than the preset confidence level threshold, the first semantic information corresponding to the first output result and the second semantic information corresponding to the second output result are determined, and it is determined whether the first semantic information and the second semantic information match.

[0238] When it is determined that the first semantic information matches the second semantic information, a preliminary processing result corresponding to the information to be identified is generated based on the first semantic information and the second semantic information.

[0239] When it is determined that the first semantic information does not match the second semantic information, supplementary description information corresponding to the information to be identified is obtained, and preliminary processing results corresponding to the information to be identified are generated based on the supplementary description information and the information to be identified.

[0240] It is evident that implementation Figure 4 The described device can perform normalization processing on a first output result and a second output result to obtain a normalized processing result, determine the result weight parameters corresponding to the first output result and the second output result, and determine the corresponding weighted average confidence level. If the weighted average confidence level is greater than or equal to a preset confidence threshold, a preliminary processing result corresponding to the information to be identified is generated based on the normalized processing result. If it is less than the preset confidence threshold, it determines whether the first semantic information and the second semantic information match. If they match, a preliminary processing result corresponding to the information to be identified is generated. If they do not match, supplementary descriptive information is obtained, and then the supplementary descriptive information and the information to be identified are combined to generate the corresponding preliminary processing result. By setting a confidence threshold, the accuracy and reliability of intent recognition can be improved. When the weighted average confidence level is lower than the threshold, the semantic information of the two models is compared. Whether a match is found can further verify the reliability of the results. When the outputs of two models are inconsistent, the intent can be reassessed by obtaining supplementary descriptive information from the user, which helps improve the accuracy and reliability of intent assessment. It can also dynamically adjust the interaction strategy based on confidence and semantic matching, which helps improve the intelligence and convenience of interaction with users and enhances the user experience. Furthermore, it can dynamically adjust weight parameters and confidence thresholds according to different business scenarios and user needs, adapting to various application scenarios. It also facilitates the expansion and maintenance of business scenario applications, adapting to various business scenarios and user needs, thereby improving interaction efficiency and convenience. In turn, it can identify and process information in various business scenarios, which helps improve the accuracy and reliability of information identification and processing, as well as the intelligence and efficiency of information identification and processing.

[0241] Example 4

[0242] Please see Figure 5 , Figure 5This is a schematic diagram of the structure of another information recognition and processing device based on a business scenario disclosed in an embodiment of the present invention. For example... Figure 5 As shown, the device for information recognition and processing based on business scenarios may include:

[0243] Memory 401 storing executable program code;

[0244] Processor 402 coupled to memory 401;

[0245] The processor 402 calls the executable program code stored in the memory 401 to execute some or all of the steps in any of the information identification and processing methods based on business scenarios in Embodiment 1 of the present invention.

[0246] Example 5

[0247] This invention discloses a computer storage medium storing computer instructions. When these computer instructions are invoked, they are used to execute some or all of the steps in any of the information identification and processing methods based on business scenarios disclosed in Embodiment 1 of this invention.

[0248] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0249] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0250] Finally, it should be noted that the above embodiments are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for information identification and processing based on business scenarios, characterized in that, The method includes: The information to be identified is acquired, and a dual-path parallel processing operation is performed on the information to be identified based on a pre-determined first analysis model and a pre-determined second analysis model to obtain preliminary processing results. Extract the target key information corresponding to the preliminary processing result, and generate target completion information based on the obtained historical interaction information and the target key information; Based on the target completion information, the target intent information and the target entity information corresponding to the information to be identified are determined. Target association information is generated based on the target intent information and the target entity information. Target operation instructions are determined based on the target association information and a predefined business process template. Control operations matching the target operation instructions are executed to obtain control feedback results. Based on the control feedback result, an information recognition result corresponding to the information to be recognized is generated; the control feedback result includes the feedback result corresponding to the target operation instruction obtained in real time, wherein the control feedback result includes the result returned by the backend business system after executing the target operation instruction; the information recognition result is the final feedback information generated according to the control feedback result, and the information recognition result is fed back to the user through text-to-speech module synthesized speech, or through the text form corresponding to the information recognition result; The information to be identified is processed in parallel using a pre-determined first analysis model and a pre-determined second analysis model to obtain preliminary processing results, including: The information to be identified is input into a pre-determined first analysis model to perform a first analysis operation on the information to be identified through the pre-determined first analysis model to obtain a first output result, wherein the first output result includes the intent output information corresponding to the information to be identified and the confidence parameter corresponding to each intent output information; the pre-determined first analysis model includes an LLM large language model; The information to be identified is input into a pre-determined second analysis model to perform a second analysis operation on the information to be identified, thereby obtaining a second output result. The second output result includes the information intent category corresponding to the information to be identified and the probability parameter corresponding to each information intent category. The pre-determined second analysis model includes a traditional lightweight classification model, which includes one of the TextCNN model, BiLSTM model, or BERT model. Normalization processing is performed on the first output result and the second output result to obtain a normalized processing result, and a preliminary processing result corresponding to the information to be identified is generated based on the normalized processing result; wherein, the dual-path parallel processing includes performing synchronous processing operations on the information to be identified through the first analysis model and the second analysis model to obtain a preliminary processing result; The step of determining the target intent information corresponding to the information to be identified and the target entity information corresponding to the information to be identified based on the target completion information includes: Based on the target completion information, the intent information corresponding to the target completion information is determined, and based on the target completion information, the target completion information is extracted by a pre-determined entity extraction module to obtain the target entity information corresponding to the information to be identified. The entity extraction module, which is predetermined, includes a Named Entity Recognition module, which is a named entity recognition module used to automatically identify and extract entity information with specific meaning from the target completion information. And, the step of generating target association information based on the target intent information and the target entity information includes: An information fusion operation is performed on the target intent information and the target entity information to generate target association information.

2. The method for information identification and processing based on business scenarios according to claim 1, characterized in that, The method further includes: Based on the information recognition results, structured text information is generated, and a speech synthesis operation is performed on the structured text information through a pre-determined speech conversion module to obtain synthesized output speech; The system obtains user information corresponding to the voice receiving user, determines voice receiving parameters based on the user information, performs voice adjustment operations on the synthesized output voice based on the voice receiving parameters, obtains target output voice, and feeds back the target output voice to the user terminal corresponding to the voice receiving user.

3. The method for information identification and processing based on business scenarios according to claim 1 or 2, characterized in that, The extraction of the target key information corresponding to the preliminary processing result includes: Perform preprocessing operations on the preliminary processing results to obtain preprocessed results, wherein the preprocessing operations include one or more of the following: word segmentation, encoding, punctuation correction, and text cleaning. Historical semantic information is acquired, and a target extraction model is constructed based on the historical semantic information, wherein the historical semantic information includes one or more of historical background context information, historical intent information, and historical context information; Based on the target extraction model, an extraction operation is performed on the preliminary processing result to obtain a structured extraction result, and based on the structured extraction result, the target key information corresponding to the preliminary processing result is generated.

4. The method for information identification and processing based on business scenarios according to claim 1 or 2, characterized in that, The generation of target completion information based on the acquired historical interaction information and the target key information includes: Obtain historical interaction information, and based on the obtained historical interaction information, determine the target missing information corresponding to the target key information; Based on the missing target information, completion guidance parameters are generated, and the completion guidance parameters are transmitted to the target user terminal corresponding to the information to be identified, and completion feedback information corresponding to the target user terminal is obtained. Based on the completion feedback information, an information completion operation is performed on the target key information to generate target completion information.

5. The method for information identification and processing based on business scenarios according to claim 1 or 2, characterized in that, The step of determining the target operation instruction based on the target association information and the predefined business process template includes: Based on the target association information and the predefined business process template, a mapping relationship is generated between the target association information and the predefined business process template. Based on the mapping relationship, the backend call instructions are assembled, and the target operation instructions are generated based on all the backend call instructions.

6. The method for information identification and processing based on business scenarios according to claim 3, characterized in that, The step of performing normalization processing on the first output result and the second output result to obtain a normalized processing result, and generating a preliminary processing result corresponding to the information to be identified based on the normalized processing result, includes: Normalization is performed on the first output result and the second output result to obtain a normalized result, wherein the normalized result includes the confidence scores corresponding to the first output result and the second output result; Based on the normalization processing result, determine the result weight parameters corresponding to the first output result and the second output result, determine the weighted average confidence level of the first output result and the second output result based on the result weight parameters, and determine whether the weighted average confidence level is greater than or equal to a preset confidence level threshold. When it is determined that the weighted average confidence level is greater than or equal to the preset confidence level threshold, a preliminary processing result corresponding to the information to be identified is generated based on the normalization processing result. When it is determined that the weighted average confidence level is less than the preset confidence level threshold, the first semantic information corresponding to the first output result and the second semantic information corresponding to the second output result are determined, and it is determined whether the first semantic information and the second semantic information match. When it is determined that the first semantic information matches the second semantic information, a preliminary processing result corresponding to the information to be identified is generated based on the first semantic information and the second semantic information. When it is determined that the first semantic information does not match the second semantic information, supplementary description information corresponding to the information to be identified is obtained, and a preliminary processing result corresponding to the information to be identified is generated based on the supplementary description information and the information to be identified.

7. An apparatus for information recognition and processing based on business scenarios, characterized in that, The device includes: The acquisition module is used to acquire the information to be identified. The processing module is used to perform dual-path parallel processing operations on the information to be identified based on a pre-determined first analysis model and a pre-determined second analysis model to obtain preliminary processing results; The extraction module is used to extract the target key information corresponding to the preliminary processing results; The generation module is used to generate target completion information based on the acquired historical interaction information and the target key information; The determination module is used to determine the target intent information corresponding to the information to be identified and the target entity information corresponding to the information to be identified based on the target completion information; The generation module is further configured to generate target association information based on the target intent information and the target entity information; The determining module is further configured to determine the target operation instruction based on the target association information and the predefined business process template; The execution module is used to execute control operations that match the target operation instructions and obtain control feedback results; The generation module is further configured to generate an information recognition result corresponding to the information to be recognized based on the control feedback result; the control feedback result includes the feedback result corresponding to the target operation instruction obtained in real time, wherein the control feedback result includes the result returned by the backend business system after executing the target operation instruction; the information recognition result is the final feedback information generated based on the control feedback result, and the information recognition result is fed back to the user through text-to-speech synthesis by the text-to-speech module, or through the text form corresponding to the information recognition result; The processing module performs dual-path parallel processing on the information to be identified based on a pre-determined first analysis model and a pre-determined second analysis model to obtain preliminary processing results. Specific methods for obtaining preliminary processing results include: The information to be identified is input into a predetermined first analysis model to perform a first analysis operation on the information to be identified through the predetermined first analysis model to obtain a first output result, wherein the first output result includes the intent output information corresponding to the information to be identified and the confidence parameter corresponding to each intent output information; The information to be identified is input into a pre-determined second analysis model to perform a second analysis operation on the information to be identified through the pre-determined second analysis model, and a second output result is obtained. The second output result includes the information intent category corresponding to the information to be identified and the probability parameter corresponding to each information intent category. Normalization processing is performed on the first output result and the second output result to obtain a normalized processing result, and a preliminary processing result corresponding to the information to be identified is generated based on the normalized processing result; wherein, the dual-path parallel processing includes performing synchronous processing operations on the information to be identified through the first analysis model and the second analysis model to obtain a preliminary processing result; The specific methods by which the determining module determines the target intent information corresponding to the information to be identified and the target entity information corresponding to the information to be identified based on the target completion information include: Based on the target completion information, the intent information corresponding to the target completion information is determined, and based on the target completion information, the target completion information is extracted by a pre-determined entity extraction module to obtain the target entity information corresponding to the information to be identified. The entity extraction module, which is predetermined, includes a Named Entity Recognition module, which is a named entity recognition module used to automatically identify and extract entity information with specific meaning from the target completion information. Furthermore, the specific methods by which the generation module generates target association information based on the target intent information and the target entity information include: An information fusion operation is performed on the target intent information and the target entity information to generate target association information.

8. An apparatus for information recognition and processing based on business scenarios, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the information identification and processing method based on the business scenario as described in any one of claims 1-6.

9. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked, are used to execute the information identification and processing method based on any one of claims 1-6.

Images