Demand information identification method and device based on voice information and large model

Abstract

The embodiment of the invention discloses a demand information identification method and device based on voice information and a large model. A specific embodiment of the method comprises the steps that dialogue voice text information for a consultation scene is acquired from a recording storage database, and the dialogue voice text information is text information obtained after dialogue voice between different users is converted into texts; inputting the dialogue voice text information and a preset prompt information set into a prompt word embedding model to obtain an embedded structured text; generating a to-be-followed task information set according to the embedded structured text; and performing task demand identification on the follow-up task scoring report to generate follow-up task demand information, and controlling execution equipment to perform task execution on the follow-up task demand information. According to the embodiment, the completeness of key information capture is improved, and the response period of the demand information of the user is shortened.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and specifically to a method and apparatus for identifying demand information based on voice information and large models. Background Technology

[0002] Voice information and large-scale model-based requirement information recognition is a technology for identifying and displaying requirement information under the prompt information engineering framework. Currently, the common method for identifying and displaying requirement information is for manual follow-up on user requirements. For example, when multiple users submit task requests, the service advisor records the requirement information for each task and tracks the progress of task completion.

[0003] However, when using the above methods to identify and display demand information, the following technical problems often arise: Because the voice conversations between users are quite complex, manual operation often results in incomplete capture of key information and an inability to respond to task execution in a timely manner. This leads to longer cycles for responding to user needs, causing task backlog and reducing task timeliness. Summary of the Invention

[0004] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] Some embodiments of this disclosure propose a method and apparatus for identifying demand information based on voice information and large models to solve the technical problems mentioned in the background section above.

[0006] In a first aspect, some embodiments of this disclosure provide a method for identifying demand information based on voice information and a large model. The method includes: obtaining dialogue voice-text information for a consultation scenario from a recording storage database, wherein the dialogue voice-text information is text information converted from dialogue voice between different users; inputting the dialogue voice-text information and a preset prompt information set into a prompt word embedding model to obtain embedded structured text; generating a set of tasks to be followed up based on the embedded structured text; identifying task weights in the set of tasks to be followed up to generate a set of task weight information; generating a follow-up task scoring report based on the set of tasks to be followed up and the set of task weight information; identifying task requirements in the follow-up task scoring report to generate follow-up task requirement information; and controlling an execution device to execute tasks based on the follow-up task requirement information.

[0007] Secondly, some embodiments of this disclosure provide a demand information recognition device based on voice information and a large model. The device includes: an acquisition unit configured to acquire dialogue voice-text information for a consultation scenario from a recording storage database, wherein the dialogue voice-text information is text information after the dialogue voice between different users is converted into text; an input unit configured to input the dialogue voice-text information and a preset prompt information set into a prompt word embedding model to obtain embedded structured text; a first generation unit configured to generate a set of tasks to be followed up based on the embedded structured text; a scoring unit configured to perform task weight recognition on the set of tasks to be followed up to generate a set of task weight information; a second generation unit configured to generate a follow-up task scoring report based on the set of tasks to be followed up and the set of task weight information; and a recognition unit configured to perform task demand recognition on the follow-up task scoring report to generate follow-up task demand information, and to control an execution device to perform task execution on the follow-up task demand information.

[0008] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0009] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0010] The above-described embodiments of this disclosure have the following beneficial effects: The demand information recognition method based on voice information and a large model, as described in some embodiments of this disclosure, improves the completeness of key information capture and shortens the response cycle for user demand information. Specifically, the reason for incomplete key information capture and a longer response cycle for user demand information is that, due to the complexity of user dialogue voices, manual operation often leads to incomplete key information capture and an inability to respond to task execution in a timely manner, resulting in a longer response cycle for user demand information, easily causing task backlog and reducing task timeliness. Based on this, some embodiments of this disclosure obtain dialogue voice text information for consultation scenarios from a recording storage database, wherein the dialogue voice text information is the text information after converting dialogue voices between different users into text. This solves the problem of how to quickly extract useful information from a large amount of recording data. The above dialogue voice text information and a preset prompt information set are input into a prompt word embedding model to obtain embedded structured text. This converts unstructured voice text into processable structured data, avoiding manual capture of key information and improving the completeness of key information capture. Then, based on the above embedded structured text, a task information set to be followed up is generated. The aforementioned set of tasks to be followed up is weighted to generate a set of task weight information. This allows for the execution of user requests based on different priorities, shortening the response cycle for user requests. The task weight information set enables timely processing of high-priority tasks, preventing task backlog and improving timeliness. Based on the set of tasks to be followed up and their weight information, a task rating report is generated. This provides a clearer view of the task ratings. The task rating report is then used to identify task requirements, generating task requirement information and controlling the execution device to execute tasks based on these requirements. Through a prompt word embedding model and weight recognition, high-priority tasks can be accurately identified, ensuring timely processing of important tasks. This improves the completeness of key information capture, shortens the response cycle for user requests, and enhances task timeliness. Attached Figure Description

[0011] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0012] Figure 1 This is a schematic diagram illustrating an application scenario of the demand information recognition method based on voice information and large model according to some embodiments of this disclosure; Figure 2 This is a flowchart of some embodiments of the demand information recognition method based on speech information and large model according to the present disclosure; Figure 3 This is a schematic diagram of the structure of some embodiments of the demand information recognition device based on voice information and large model according to the present disclosure; Figure 4 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure; Figure 5 It is a graph data structure suitable for implementing some embodiments of the present disclosure, which is a set of associated requirement information. Detailed Implementation

[0013] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0014] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0015] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0016] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0017] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0018] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] Figure 1 This is a schematic diagram illustrating an application scenario of the demand information recognition method based on voice information and large model according to some embodiments of this disclosure.

[0020] exist Figure 1In the application scenario, firstly, the recording storage database 101 can store dialogue voice-text information for consultation scenarios, wherein the dialogue voice-text information is the text information after the dialogue voice between different users is converted into text. Then, the dialogue voice-text information and the preset prompt information set are input into the prompt word embedding model 102 to obtain embedded structured text. Then, based on the embedded structured text, a task information set 103 to be followed up is generated. Next, the task information set to be followed up is subjected to task weight identification to generate a task weight information set 104 to be followed up. Based on the task information set to be followed up and the task weight information set to be followed up, a follow-up task scoring report 105 is generated. The follow-up task scoring report is subjected to task requirement identification to generate follow-up task requirement information 106, and the control execution device 107 executes the task according to the follow-up task requirement information.

[0021] It should be noted that computing devices can be either hardware or software. When a computing device is hardware, it can be implemented as a distributed cluster of multiple servers or terminal devices, or as a single server or terminal device. When a computing device is software, it can be installed on the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here. It should be understood that... Figure 1 The number of execution devices in the system can be arbitrary, depending on the implementation requirements.

[0022] Continue to refer to Figure 2 The flowchart 200 illustrates some embodiments of the demand information recognition method based on speech information and large model according to the present disclosure. The demand information recognition method based on speech information and large model includes the following steps: Step 201: Retrieve the dialogue voice and text information for the consultation scenario from the audio recording storage database.

[0023] In some embodiments, the execution entity (e.g., a computing device) of the demand information recognition method based on voice information and large model can obtain dialogue voice text information for consultation scenarios from a recording storage database via wired or wireless connection, wherein the aforementioned dialogue voice text information is text information after the dialogue voice between different users is converted into text.

[0024] Here, the aforementioned audio recording database is used to store conversations recorded by the voice recorder. The aforementioned consultation scenario can refer to a conversation between a service advisor and a customer. For example, the consultation scenario could be: "Service Advisor: What do you need? Customer: I'd like to fix my tire." The voice recorder can be worn by the service advisor.

[0025] It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other currently known or future wireless connection methods.

[0026] Optionally, the aforementioned implementing entity can retrieve the dialogue voice and text information for the consultation scenario from the audio recording storage database through the following steps: The first step is to collect the voice dialogue information between different users in the above consultation scenarios.

[0027] Here, the different users mentioned above can include, but are not limited to, at least one of the following: service advisor, customer, and customer service representative. Specifically, customers can consult service advisors, and service advisors can consult customer service representatives.

[0028] As an example, the aforementioned implementing entity can use a voice recorder to collect voice conversations between different users in the above consultation scenario. For example, the voice conversation information may include, but is not limited to: whether the receptionist's name is mentioned, whether the service advisor proactively inquires about the customer's channel of entry, whether the service advisor proactively invites the customer for a vehicle pre-inspection, whether the customer's license plate number is mentioned, whether the service advisor mentions that the customer is a first-time visitor, whether the service advisor inquires whether the customer will renew their insurance, whether the service advisor recommends products for first-time visitors, whether the service advisor adds the customer on WeChat (WeChat for Business), whether the service advisor proactively informs the customer of the subsequent procedures, and whether the service advisor invites the customer to have a used car appraisal when the customer indicates that they want to sell the car.

[0029] The second step is to input the above dialogue voice information into the voiceprint recognition model to obtain user voice information, wherein the above user voice information is voice information containing voice user tags.

[0030] Here, the input to the aforementioned voiceprint recognition model is dialogue speech information, and the output is user speech information. The aforementioned voiceprint recognition model is used to identify the timbre of different users in the aforementioned dialogue speech information and to associate speech with corresponding users. The aforementioned voiceprint recognition model can be used to represent the correspondence between dialogue speech information and user speech information. The aforementioned voiceprint recognition model can sequentially compare dialogue speech information with multiple sets of preset dialogue speech information in a preset voiceprint relationship table. The aforementioned preset voiceprint relationship table can be created based on the analysis of a large amount of preset dialogue speech information. Each set of preset dialogue speech information corresponds to preset user speech information. The aforementioned preset user speech information can be pre-defined user speech information. For example, the aforementioned voiceprint recognition model can refer to a Deep Neural Speaker Embedding System (DeepSpeaker) model.

[0031] The third step is to transcribe the above-mentioned character voice information into text to generate dialogue voice text information.

[0032] As an example, the aforementioned executing entity can use Automatic Speech Recognition (ASR) technology to transcribe the voice information of the aforementioned characters into text to generate dialogue voice text information.

[0033] Step 202: Input the above dialogue voice text information and the preset prompt information set into the prompt word embedding model to obtain embedded structured text.

[0034] In some embodiments, the execution entity may input the dialogue voice text information and the preset prompt information set into the prompt word embedding model to obtain embedded structured text.

[0035] Here, the aforementioned preset prompt information set can refer to a collection of pre-defined prompt information. For example, the aforementioned preset prompt information set may include, but is not limited to, at least one of the following: service advisor confirming customer tasks, service advisor conducting pre-inspection with customer, and customer signing confirmation. Service advisor confirming customer tasks could mean that when a customer reports a vibration when braking, the service advisor checks the customer's tire tread depth, presence of cracks, brake fluid level, and whether the front and rear brake pads (or discs) show significant wear, surface cracks, or obvious grooves. The aforementioned prompt word embedding model can be used to embed the aforementioned preset prompt information set into the dialogue voice text information. The aforementioned prompt word embedding model may include: an input layer, a text embedding layer, an encoder layer, a prompt word guidance layer, a semantic analysis layer, and an output layer. The aforementioned text embedding layer can map each word or phrase to a fixed-dimensional vector. The aforementioned text embedding layer includes an embedding matrix. For example, segmenting the dialogue voice text information into words, such as "customer complains about vehicle brake failure and requests immediate handling," can be segmented into "customer, complaint, vehicle, brake, failure, request, immediately, handling." Then, Word2Vec is used to map "customer, complaint, vehicle, brake, malfunction, request, immediately, handle" to a high-dimensional vector space. The encoder layer described above encodes the processed speech-text vector sequence to extract semantic features and contextual information from the text. This encoder layer includes Recurrent Neural Network (RNN) units, such as Long Short-Term Memory (LSTM) networks. The prompt word guidance layer performs weighted summation, concatenation, or other operations on the prompt word vectors and text vectors, integrating the prompt word information into the semantic representation of the text. This prompt word guidance layer includes an embedding matrix. For example, the prompt word could refer to "urgency" or "severity of the problem." The semantic analysis layer transforms the semantic representation of the text into specific structured information. For example, for standard actions of a service consultant, the model, guided by prompt words, identifies from the text whether the consultant has completed the corresponding action and outputs structured results, such as "Service consultant introduces themselves: yes / no" or "Confirm customer's project: yes / no." The input layer is used to input the dialogue voice text information and a preset prompt information set. The semantic analysis layer includes a feed-forward neural network (FFNN). For example, the dialogue voice text information input to the input layer could be the text information "A customer complains of vehicle brake failure and requests immediate handling." The preset prompt information set may include, but is not limited to, at least one of the following: problem severity, urgency, and request type. The output layer is used to output embedded structured text.The above output layer includes a fully connected neural network (FCNN). Here, the input of the above prompt word embedding model can refer to the dialogue speech text information and a preset prompt information set. The output can refer to an embedded structured text. For example, the embedded structured text can refer to the urgency level: [0.234, 0.567,..., 0.890]. Among them, 0.234 can represent the semantic features corresponding to the prompt word. The above prompt word embedding model can be used to represent the correspondence between the dialogue speech text information and the preset prompt information set and the embedded structured text. The above prompt word embedding model can compare the dialogue speech text information with multiple groups of preset dialogue speech text information in the preset prompt word embedding relationship table in sequence. The above preset prompt word embedding relationship table can be created based on the analysis of a large number of preset dialogue speech text information. Each group of preset dialogue speech text information corresponds to a preset embedded structured text. The above preset embedded structured text can be a preset embedded structured text.

[0036] Optionally, the above execution entity can input the above dialogue speech text information and the preset prompt information set into the prompt word embedding model through the following steps to obtain an embedded structured text: In the first step, perform word segmentation on the above dialogue speech text information to generate a word-segmented speech text sequence.

[0037] As an example, the above execution entity can use a Chinese word segmentation tool to perform word segmentation on the above dialogue speech text information to generate a word-segmented speech text sequence. For example, the Chinese word segmentation tool can refer to Jieba. For example, the dialogue speech text information is "There is a problem with my car recently. The power is insufficient when stepping on the accelerator." The word-segmented speech text sequence is "My, car, recently, have, problem, step on, accelerator, when, power, insufficient." In the second step, remove stop words from the above word-segmented speech text sequence to generate a speech text sequence after removal.

[0038] Here, the above stop words can refer to words that frequently appear in the text but contribute little to the semantics. For example, the above stop words can refer to "de", "shi", "he".

[0039] As an example, the above execution entity can剔除 the stop words in the above word-segmented speech text sequence to generate a speech text sequence after removal. For example, the speech text sequence after removal is "car, recently, problem, step on, accelerator, when, power, insufficient." In the third step, convert the above speech text sequence after removal into a vector to obtain a speech text vector sequence.

[0040] It should be noted that in the translation of the sentence in ID=16, the Chinese word "剔除" is directly used as there is no exact equivalent in English and it is more understandable in the context. If a more detailed translation is required, it can be "remove the stop words by weeding out".As an example, the aforementioned execution entity can input the removed speech-text sequence into the text embedding layer of the cue word embedding model to obtain a speech-text vector sequence. The text embedding layer can map each word or phrase to a fixed-dimensional vector.

[0041] The fourth step is to normalize the above speech-text vector sequence to generate a processed speech-text vector sequence.

[0042] As an example, the aforementioned execution entity can normalize each speech text vector in the aforementioned speech text vector sequence to generate a processed speech text vector, thus obtaining a processed speech text vector sequence. Here, the aforementioned normalization can refer to normalization.

[0043] The fifth step is to encode the processed speech-text vector sequence to generate encoded text semantic information.

[0044] As an example, the aforementioned execution entity can input the processed speech-text vector sequence into the encoder layer of the cue word embedding model to obtain encoded text semantic information. The encoder layer is used to encode the processed speech-text vector sequence to extract semantic features and contextual information from the text. The aforementioned encoder layer can refer to a multi-head attention mechanism.

[0045] The sixth step is to convert the above preset prompt information set into a vector to obtain a prompt information vector set.

[0046] As an example, the aforementioned execution entity can input each preset prompt information in the preset prompt information set into the text embedding layer of the prompt word embedding model to generate a prompt information vector and obtain a prompt information vector set.

[0047] Step 7: Generate a sequence of prompt information vectors based on the above set of prompt information vectors.

[0048] As an example, the aforementioned execution entity can use a Chinese word segmentation tool to segment the aforementioned prompt information vector set to generate a segmented prompt information vector set. Then, stop words in the segmented prompt information vector set are removed to obtain a removed prompt information vector sequence, which serves as the prompt information vector sequence.

[0049] Step 8: Fuse the above prompt information vector sequence with the above speech text vector sequence to obtain the fused text semantic information.

[0050] As an example, the aforementioned execution entity can concatenate the aforementioned prompt information vector sequence with the aforementioned speech-text vector sequence to obtain the concatenated text semantic information, which can then be used as the fused text semantic information. For example, the aforementioned prompt information vector sequence could be [1, 3, 100], representing the embedding vectors of 3 words in the input text. The aforementioned speech-text vector sequence could be [1, 10, 100], representing the embedding vectors of 10 prompt words. The aforementioned concatenated text semantic information could be [1, 13, 100].

[0051] The ninth step is to perform structured information transformation on the above-mentioned fused text semantic information and the above-mentioned encoded text semantic information to obtain the transformed structured text, which is used as the embedded structured text.

[0052] As an example, the aforementioned executing entity can input the fused text semantic information and the encoded text semantic information into the semantic analysis layer of the prompt word embedding model to obtain embedded structured text.

[0053] As another example, the aforementioned execution entity can map the fused text semantic information and the encoded text semantic information into embedded structured text based on a schema template (e.g., Apache Avro). For example, the fused text semantic information could mean "I want a refund of the price difference." The encoded text semantic information could mean "The customer requests a refund of the price difference." Then the embedded structured text would be {"Type": "After-sales", "Subclass": "Price Compensation", "Urgency": "3"}.

[0054] Step 203: Generate a set of task information to be followed up based on the above embedded structured text.

[0055] In some embodiments, the aforementioned execution entity may generate a set of task information to be followed up based on the aforementioned embedded structured text.

[0056] Here, the "tasks to be followed up" information in the set of tasks to be followed up can refer to task information that service advisors need to follow up with on for customers. For example, the set of tasks to be followed up can include: customer feedback on the recommended products or services, the follow-up results of the product being "unsuitable", whether the service advisor recommends other more suitable products, customer feedback on the recommended products or services, and whether the service advisor provides further explanation, etc.

[0057] Optionally, the aforementioned executing entity can generate a set of task information to be followed up based on the embedded structured text using the following steps: The first step is to extract keywords from the embedded structured text to generate a key information set of the structured text.

[0058] As an example, the aforementioned execution entity can use natural language processing techniques (e.g., term frequency-inverse document frequency, TF-IDF) to extract keywords from the embedded structured text to generate a set of key information about the structured text.

[0059] The second step is to input the above-mentioned structured text key information set into the semantic annotation model to obtain the annotated structured text key information set.

[0060] Here, the input of the semantic annotation model is a set of key information in structured text, and the output of the semantic annotation model is a set of key information in annotated structured text.

[0061] Here, the aforementioned semantic annotation model can be a model used to semantically annotate the aforementioned set of key information in structured text. This semantic annotation model can be used to represent the correspondence between the set of key information in structured text and the annotated set of key information in structured text. The semantic annotation model can sequentially compare the set of key information in structured text with multiple pre-set sets of key information in structured text in a pre-defined semantic annotation relationship table. This pre-defined semantic annotation relationship table can be created based on the analysis of a large number of pre-set sets of key information in structured text. Each set of pre-set sets of key information in structured text corresponds to a pre-defined set of key information in structured text after annotation. This pre-defined set of key information in structured text after annotation can be a pre-defined set of key information in structured text after annotation. For example, the aforementioned semantic annotation model can refer to a Bidirectional Encoder Representations from Transformers (BERT) model.

[0062] As an example, the aforementioned executing entity can input each piece of structured text key information from the aforementioned structured text key information set into the semantic annotation model to generate annotated structured text key information, thus obtaining an annotated structured text key information set.

[0063] The third step is to perform syntactic analysis on the above-mentioned annotated structured text key information set to generate an analyzed key information set, which will serve as the information set for the tasks to be followed up.

[0064] As an example, the aforementioned execution entity can utilize the Natural Language Toolkit (NLTK) in Python to perform syntactic analysis on the key information set of the annotated structured text to generate an analyzed key information set as the information set for follow-up tasks. For instance, the key information set of the annotated structured text could be {["text": "Customer complained about vehicle brake failure and requested immediate handling.", "label": "serious problem"], ["text": "Customer reported a good vehicle maintenance service experience but mentioned the next maintenance time.", "label": "routine feedback"]}. The syntactic analysis could refer to part-of-speech tagging (POS tagging). The analyzed key information set could be {["customer", "nsubj", "complaint"], ["brake failure", "dobj", "vehicle"], ["request", "nsubj", "immediate handling"]}. nsubj represents the subject. dobj represents the object.

[0065] Step 204: Perform task weight identification on the above-mentioned task information set to be followed up, so as to generate a task weight information set to be followed up.

[0066] In some embodiments, the execution entity may identify task weights in the set of tasks to be followed up in order to generate a set of task weight information to be followed up.

[0067] Here, the task weight information in the aforementioned task weight information set can refer to the weight information of the tasks that service advisors follow up with customers.

[0068] In the process of adopting technical solutions to address the problems mentioned in the background section, the following issues often arise: Traditional manual operations struggle to quickly and accurately extract crucial information from large volumes of dialogues and fail to rapidly identify high-priority tasks, leading to longer execution cycles for handling high-priority user requests. Furthermore, the lack of consideration for historical user behavior data can result in inefficient resource allocation during execution, leading to lower user satisfaction.

[0069] Faced with the above-mentioned technical problems, the inventors decided to adopt the following solution: Optionally, the aforementioned executing entity may perform task weight identification on the aforementioned task-to-follow-up information set through the following steps to generate a task-to-follow-up weight information set: The first step is to perform task dimension detection on the above set of tasks to be followed up in order to generate a set of tasks to be followed up information.

[0070] Here, the aforementioned task dimensions may include, but are not limited to, at least one of the following: problem severity, urgency, and demand level. Problem severity can refer to the severity of a quality issue; for example, a tire blowout is more serious than a paint scratch. Problem severity, urgency, and demand level are each divided into three levels: low, medium, and high. For example, a tire blowout is a high level, while a paint scratch is a low level.

[0071] The second step is to identify the priority of the above-mentioned set of tasks to be followed up in order to generate a priority set of tasks to be followed up.

[0072] As an example, the aforementioned executing entity can determine the level of each task information group in the set of task information groups to be followed up, and then calculate the average of the determined levels to generate a task priority set. For instance, the levels of each task information group in the set of task information groups to be followed up are medium, medium, low, high, and medium. Here, low level represents 1, medium level represents 2, and high level represents 3. Then the average of the determined levels is (2+2+1+3+2) / 5=2.

[0073] The third step is to acquire the user's historical behavior data corresponding to each task in the above task information set to generate a user historical behavior data set.

[0074] As an example, the aforementioned executing entity can obtain the user's historical behavior data corresponding to each task in the task information set to be followed up from the Customer Relationship Management (CRM) system to generate a user historical behavior data set.

[0075] The fourth step is to dynamically weight the above-mentioned user historical behavior data set to generate a weight set of tasks to be followed up for the corresponding user historical behavior data set.

[0076] As an example, the aforementioned execution entity can perform a first weighted processing on the user historical behavior data set that meets the condition of an average order value exceeding a preset value, to generate a first task weight to be followed up. The preset value can be 5000. The first weighted processing can be weight × 1.5. Then, a second weighted processing is performed on the user historical behavior data set that meets the condition of a repurchase frequency exceeding a preset number, to generate a second task weight to be followed up. The preset number of repurchases can be 8 times. The first weighted processing can be weight × 1.2. Next, a third weighted processing is performed on the user historical behavior data set that meets the condition of a refund frequency less than a preset refund frequency, to generate a third task weight to be followed up. The preset number of refunds can be 2 times. The first weighted processing can be weight × 0.8. Finally, the first, second, and third task weights to be followed up are determined as the task weight set to be followed up.

[0077] The fifth step is to perform a preset keyword set detection on the above set of tasks to be followed up, and in response to the detection of preset keywords in the preset keyword set, to increase the priority of at least one task to be followed up corresponding to the preset keywords, so as to generate a priority set of tasks to be followed up.

[0078] Here, the aforementioned preset keyword set can refer to a pre-defined set of keywords. For example, the preset keyword set may include, but is not limited to, at least one of the following: expedited, complaint, after-sales, refund. The aforementioned priority increase can refer to a weight increment of 1. For example, if an "expedited" task is detected from the aforementioned task information set, the weight of the "expedited" task is increased by 1. The detection method is keyword matching. For example, keyword matching is used to detect whether the task information set contains the preset keyword set. If a preset keyword from the preset keyword set is detected, the corresponding task information is given an increased priority.

[0079] The sixth step is to merge the above priority set of tasks to be followed up, the above weight set of tasks to be followed up, and the above priority set of tasks to be followed up after the addition of tasks to obtain the merged comprehensive score set of tasks to be followed up.

[0080] As an example, the aforementioned executing entity can add the priority of each task after the priority set of the tasks to be followed up is increased, the weight of the task corresponding to the priority of the task after ...

[0081] The seventh step is to classify the scoring information of the above-mentioned integrated score set of tasks to be followed up according to the preset scoring threshold conditions, so as to generate a score information category set of tasks to be followed up.

[0082] Here, the aforementioned preset scoring threshold condition can refer to the pre-set "a score value less than 60 is a normal work order, and a score value greater than or equal to 60 is an emergency work order".

[0083] As an example, the aforementioned implementing entity can classify the integrated score set of tasks to be followed up according to the score values ​​to generate a set of score information categories for tasks to be followed up. This set of score information categories includes ordinary work order categories and urgent work order categories.

[0084] Step 8: Based on the above-mentioned set of task rating information categories, the above-mentioned integrated set of task ratings after fusion is labeled with categories to generate a labeled set of task rating information, which serves as the task rating information set.

[0085] As an example, the aforementioned executing entity can label each of the following task scoring information categories in the aforementioned following task scoring information category set to the fused following task comprehensive score set corresponding to the aforementioned following task scoring information category, so as to generate labeled following task scoring information and obtain a labeled following task scoring information set, which serves as the following task scoring information set.

[0086] The content in steps one through eight above constitutes an inventive point of this disclosure, solving the following technical problem: "The inability to quickly identify high-priority tasks leads to longer execution cycles for handling high-priority user requests, resulting in unreasonable resource allocation and poor user satisfaction." Factors leading to longer execution cycles and unreasonable resource allocation for handling high-priority user requests often include: traditional manual operations struggle to quickly and accurately extract important information from large amounts of dialogue, and the inability to quickly identify high-priority tasks, thus extending the execution cycle. Furthermore, the lack of consideration for historical user behavior data can lead to unreasonable resource allocation during execution, resulting in poor user satisfaction. Solving these factors can shorten the execution cycle for handling high-priority user requests. To achieve this, firstly, task dimension detection is performed on the aforementioned set of tasks to be followed up to generate a set of task information groups. This allows for multi-dimensional detection of different tasks, facilitating subsequent resource allocation. Secondly, priority identification is performed on the aforementioned set of task information groups to be followed up to generate a priority set of tasks to be followed up. This provides convenience for subsequent processing. The user's historical behavior data corresponding to each task in the aforementioned task-to-follow-up information set is acquired to generate a user historical behavior data set. This allows for the identification of a user's historical behavior, facilitating resource allocation and improving user satisfaction. The user historical behavior data set is then dynamically weighted to generate a corresponding task weight set. A preset keyword set is detected in the task-to-follow-up information set. In response to the detection of a preset keyword in the preset keyword set, at least one task corresponding to that preset keyword is prioritized, generating a priority set for the tasks after priority increase. This shortens the execution cycle for handling high-priority user requests. The priority set, weight set, and priority set for the tasks after priority increase are then fused to obtain a fused comprehensive score set for the tasks. Based on a preset scoring threshold, the fused comprehensive score set for the tasks is categorized to generate a task scoring information category set. Based on the aforementioned category set of task follow-up scoring information, the fused comprehensive score set of tasks follow-up is categorized to generate a labeled score information set for follow-up tasks. Therefore, through automated task dimension detection, priority identification, dynamic weighting, and keyword detection, high-priority tasks can be quickly identified and responded to promptly. Simultaneously, considering historical user behavior data, important tasks are ensured to be prioritized. This not only improves task processing efficiency but also optimizes resource allocation and enhances user satisfaction.

[0087] Step 205: Generate a follow-up task scoring report based on the above-mentioned set of follow-up task information and the above-mentioned set of follow-up task weight information.

[0088] In some embodiments, the aforementioned executing entity may generate a follow-up task scoring report based on the aforementioned set of follow-up task information and the aforementioned set of follow-up task weight information.

[0089] Optionally, the aforementioned executing entity may generate a follow-up task scoring report based on the aforementioned set of follow-up task information and the aforementioned set of follow-up task weight information through the following steps: The first step is to merge the above-mentioned set of tasks to be followed up and the above-mentioned set of weights of tasks to be followed up to obtain a summary set of tasks.

[0090] Here, the above-mentioned task summary information set is an information set that summarizes the above-mentioned task follow-up information set and the above-mentioned task follow-up weight information set.

[0091] As an example, the aforementioned executing entity can bind each task weight information in the aforementioned task weight information set to the corresponding task information in the same set, thereby generating bound task information and obtaining a bound task information set, which serves as the task summary information set. Each task information has a unique task weight information. For example, if the task information is for a service consultant to follow up with a customer, then the task weight information is 0.4.

[0092] The second step is to populate the report template with the above task summary information set to generate a populated task report.

[0093] Here, the aforementioned report template can refer to a template that includes a fixed title, tables, and populated content areas. The fixed title could be "Follow-up Project Scoring Report." The tables could be tables with rows representing the sets of tasks to be followed up and columns representing the sets of weights for those tasks. The populated content areas could be areas for filling in summary content or task analysis conclusions.

[0094] As an example, the aforementioned executing entity can use the Python language to populate the report template of the above task summary information set to generate a populated task report.

[0095] The third step is to verify the completed task report to generate verification results.

[0096] Here, the above test results can characterize whether the test passes or fails.

[0097] As an example, the aforementioned executing entity can use automated scripts to verify the populated task report and generate verification results.

[0098] The fourth step is to determine that the above test results indicate that the test has passed, and then to designate the corresponding filled task report as the follow-up task scoring report.

[0099] Step 206: Identify task requirements from the above-mentioned follow-up task scoring report to generate follow-up task requirement information, and control the execution device to execute the tasks based on the above-mentioned follow-up task requirement information.

[0100] In some embodiments, the execution entity may identify task requirements from the follow-up task scoring report to generate follow-up task requirement information, and control the execution device to perform tasks based on the follow-up task requirement information.

[0101] In the process of adopting technical solutions to address the problems mentioned in the background section, the following issues often arise: When manually assessing task priorities, it is difficult to comprehensively consider multiple factors, leading to inaccurate priority assessments and unreasonable allocation of resources required to execute tasks. Furthermore, because it is difficult to comprehensively consider the urgency and importance of tasks, response times for high-priority user requests are often longer.

[0102] Faced with the above-mentioned technical problems, the inventors decided to adopt the following solution: Optionally, the aforementioned executing entity may identify task requirements in the aforementioned follow-up task scoring report through the following steps to generate follow-up task requirement information, and control the executing device to execute the task based on the aforementioned follow-up task requirement information: The first step is to extract preset semantic words from the above-mentioned follow-up task scoring report to generate a semantic word set for the follow-up task scoring report.

[0103] Here, the aforementioned preset semantic words can refer to pre-defined semantic words. For example, the aforementioned preset semantic words could refer to "immediately, urgently, complaint".

[0104] As an example, the aforementioned executing entity can use natural language processing technology (e.g., term frequency-inverse document frequency, TF-IDF) to extract preset semantic words from the aforementioned follow-up task scoring report in order to generate a semantic word set for the follow-up task scoring report.

[0105] The second step is to identify the business requirements of the follow-up tasks corresponding to the semantic words of the follow-up task scoring reports that represent problems to be processed, in response to the determination that there are semantic words of follow-up task scoring reports that represent problems to be processed in the above-mentioned semantic word set, so as to generate the first requirement information set.

[0106] Here, the semantic terms in the aforementioned follow-up task scoring report representing pending issues can refer to "immediately" or "complaint." The aforementioned business requirements can refer to the need to follow up on user tasks. For example, these business requirements could refer to tire repair needs and car cover maintenance needs.

[0107] As an example, the aforementioned implementing entity can match the follow-up tasks corresponding to the semantic words in the follow-up task scoring report representing the problem to be processed with a preset set of business requirement keywords, and determine the matched business requirement keywords as the first set of requirement information. The preset set of business requirement keywords may include, but is not limited to, at least one of the following: "immediately," "complaint," "urgent," and "insufficient timeliness." The first set of requirement information may refer to "complaining about poor service attitude" or "arranging repairs immediately."

[0108] The third step is to identify the business requirements of the follow-up tasks corresponding to the semantic words in the follow-up task scoring report that indicate insufficient timeliness in the aforementioned semantic word set, in order to generate a second set of requirements information.

[0109] Here, the semantic term in the above-mentioned follow-up task scoring report that indicates insufficient timeliness can refer to "insufficient timeliness".

[0110] As an example, the aforementioned implementing entity can match the follow-up tasks corresponding to the semantic words in the follow-up task scoring report indicating insufficient timeliness with a preset set of business requirement keywords, and determine the matched business requirement keywords as a second set of requirement information. The second set of requirement information could refer to "Insufficient timeliness, please resolve quickly" or "Urgent task, please resolve quickly".

[0111] The fourth step is to identify the business requirements of the follow-up tasks corresponding to the follow-up task scoring report semantic words that represent high priority in the set of semantic words of the follow-up task scoring report, in order to generate a third requirement information set.

[0112] As an example, the aforementioned implementing entity can match the follow-up tasks corresponding to the semantic words in the follow-up task scoring report representing high priority with a preset set of business requirement keywords, and determine the matched business requirement keywords as a third set of requirement information. The third set of requirement information could refer to "handle immediately" or "urgent tasks, please resolve quickly".

[0113] The fifth step is to perform correlation analysis on the first set of demand information, the second set of demand information, and the third set of demand information to obtain the correlated demand information set.

[0114] As an example, the aforementioned executing entity can combine the first, second, and third sets of demand information to obtain a combined demand information set. Then, it can merge the demand information of similar demands within the combined demand information set to obtain a merged demand information set, which serves as the associated demand information set. For example, the aforementioned similar demands could refer to all being urgent tasks requiring immediate resolution.

[0115] The sixth step is to create a graph data structure for the above-mentioned associated demand information set to generate the associated demand graph information.

[0116] Here, the graph data structure described above can represent the relationships between nodes.

[0117] As an example, the aforementioned executing entity can use the associated demand information in the aforementioned associated demand information set as nodes and the relationships between demand information as edges to create a graph data structure, thereby generating associated demand graph information. For example... Figure 5 As shown, Figure 5 The associated demand graph information includes: demand information 1, demand information 2, demand information 3, demand information 4, demand information 5, demand information 6, and demand information 7. The associated demand information in the associated demand information set can refer to either demand information 1 or demand information 3 in the graph. Demand information 1, demand information 2, demand information 3, demand information 4, demand information 5, demand information 6, and demand information 7 are nodes. The edges connecting the nodes represent the relationships between them: demand information 1 has the same priority as demand information 5 and demand information 4; demand information 1 and demand information 2 have the same problem; demand information 2 and demand information 3 have the same source; demand information 3 and demand information 6 have the same priority; demand information 3 and demand information 4 have the same source; demand information 5 and demand information 7 have the same priority; demand information 4 and demand information 7 have the same source; and demand information 5 and demand information 7 have the same problem.

[0118] The seventh step is to identify key nodes in the above-mentioned associated requirement graph information to generate associated key requirement information, which will serve as the follow-up task requirement information.

[0119] As an example, the aforementioned executing entity can use centrality analysis to identify key nodes in the associated demand graph information to generate associated key demand information, which can then be used as follow-up task demand information. These key nodes can refer to nodes with a degree greater than 1.

[0120] Step 8: Based on the above follow-up task scoring report, update the weights of the above follow-up task requirement information to generate a weighted set of related requirement information.

[0121] Here, we assume that the base weight of the aforementioned follow-up task requirement information in the aforementioned follow-up task scoring report is 0.5. Then, the weighted set of the associated requirement information is the sum of the weight of the follow-up task corresponding to the first requirement information set × 1.5, the weight of the follow-up task corresponding to the second requirement information set × 1.2, and the weight of the follow-up task corresponding to the third requirement information set × 0.8. For example, the weight of the follow-up task corresponding to the first requirement information set is {0.2, 0.3, 0.5}. The weight of the follow-up task corresponding to the second requirement information set is {0.1, 0.2, 0.7}. The weight of the follow-up task corresponding to the third requirement information set is {0.6, 0.2, 0.2}. The weight set of the associated demand information is then {0.2×1.5+0.1×1.2+0.6×0.8=0.9, 0.3×1.5+0.2×1.2+0.2×0.8=0.85, 0.5×1.5+0.7×1.2+0.2×0.8=1.75}, which is {0.9, 0.85, 1.75}.

[0122] The ninth step is to sort the aforementioned set of associated demand information weights to generate a sequence of associated demand information weights.

[0123] As an example, the aforementioned executing entity can sort the set of associated demand information weights from largest to smallest to generate a sequence of associated demand information weights. For example, the sequence of associated demand information weights could be (1.75, 0.9, 0.85).

[0124] Step 10: Control the execution device to execute the follow-up task requirement information corresponding to the weights of the associated requirement information that meet the preset conditions in the above-mentioned associated requirement information weight sequence.

[0125] Here, the aforementioned preset condition may refer to a pre-set weight greater than 1. The aforementioned execution device may refer to maintenance equipment or diagnostic equipment.

[0126] As an example, the aforementioned execution entity can control repair equipment via an API interface to perform repairs on the follow-up task requirement information corresponding to the weighted related post-associated requirement information that meets preset conditions in the aforementioned post-associated requirement information weight sequence. It can also control diagnostic equipment via an API interface to diagnose the follow-up task requirement information corresponding to the weighted related post-associated requirement information that meets preset conditions in the aforementioned post-associated requirement information weight sequence. The aforementioned repair equipment can refer to a car lift. A car lift is used to lift a vehicle for repair personnel to perform chassis repairs. The aforementioned diagnostic equipment can refer to an OBD-II diagnostic tool (On-Board Diagnostics II). The OBD-II diagnostic tool reads the vehicle's fault codes and real-time data by connecting to the vehicle's OBD-II interface. For example, starting the car lift via an API call. Reading data from the OBD-II diagnostic tool via an API call.

[0127] The content in steps one through ten above constitutes an inventive point of this disclosure, solving the technical problem of "long response cycles for high-priority user requests." Factors contributing to long response cycles for high-priority user requests often include: difficulty in comprehensively considering multiple factors during manual task priority assessment, leading to inaccurate priority evaluation and unreasonable resource allocation for task execution; and difficulty in comprehensively considering the urgency and importance of tasks, resulting in long response cycles for high-priority user requests. Solving these factors can shorten the response cycle for high-priority user requests. To achieve this, firstly, semantic words representing unresolved issues, time constraints, and high priority are accurately identified from the dialogue text using preset semantic word extraction and natural language processing technologies, generating a semantic word set. Then, through automated business requirement identification and correlation analysis, a first, second, and third requirement information set are quickly generated, and a graph data structure is created and key nodes are identified to ensure that high-priority tasks can be identified and processed in a timely manner. Finally, through weight updates and sorting, combined with dynamic weighted processing of user historical behavior data, a weighted set and weight sequence of correlated requirement information are generated to ensure accurate task priority assessment and reasonable resource allocation. Then, by controlling the execution equipment, high-priority tasks that meet preset conditions are automatically executed, improving task processing efficiency and accuracy while reducing human intervention. Next, through automated task identification and execution, the response cycle for user requests is shortened, improving user satisfaction and loyalty. Furthermore, through dynamic weighted processing and weight ranking, combined with historical user behavior data, resources are rationally allocated to ensure that high-priority tasks are processed first. Therefore, the response cycle for high-priority user requests is shortened.

[0128] Further reference Figure 3As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a demand information recognition device based on voice information and a large model. These device embodiments are similar to... Figure 2 Corresponding to the method embodiments shown, this demand information recognition device based on voice information and large models can be specifically applied to various electronic devices.

[0129] like Figure 3 As shown, some embodiments of the demand information recognition device 300 based on voice information and large model include: acquisition unit 301, input unit 302, first generation unit 303, scoring unit 304, second generation unit 305 and recognition unit 306. The system includes: an acquisition unit 301 configured to acquire dialogue voice-text information for a consultation scenario from a recording storage database, wherein the dialogue voice-text information is text information converted from the voice-text of conversations between different users; an input unit 302 configured to input the dialogue voice-text information and a preset prompt information set into a prompt word embedding model to obtain embedded structured text; a first generation unit 303 configured to generate a set of tasks to be followed up based on the embedded structured text; a scoring unit 304 configured to perform task weight recognition on the set of tasks to be followed up to generate a set of task weight information; a second generation unit 305 configured to generate a follow-up task scoring report based on the set of tasks to be followed up and the set of task weight information; and a recognition unit 306 configured to perform task requirement recognition on the follow-up task scoring report to generate follow-up task requirement information and control the execution device to execute the tasks based on the follow-up task requirement information.

[0130] It is understandable that the units and references described in the demand information recognition device 300 based on voice information and a large model are... Figure 2 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method are also applicable to the demand information recognition device 300 based on speech information and a large model, and the units contained therein, and will not be repeated here. The following is for reference. Figure 4 It shows a schematic diagram of the structure of an electronic device 400 (e.g., a computing device) suitable for implementing some embodiments of the present disclosure. Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0131] like Figure 4As shown, the electronic device 400 may include a processing unit 401 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a storage device 408 into a random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the electronic device 400. The processing unit 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0132] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic device 400 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 An electronic device 400 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 4 Each box shown can represent a device or multiple devices as needed.

[0133] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 409, or installed from storage device 408, or installed from ROM 402. When the computer program is executed by processing device 401, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0134] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0135] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0136] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently without being assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: retrieve dialogue voice-text information for a consultation scenario from a recording storage database, wherein the dialogue voice-text information is text information converted from dialogue voice between different users; input the dialogue voice-text information and a preset prompt information set into a prompt word embedding model to obtain embedded structured text; generate a set of tasks to be followed up based on the embedded structured text; identify task weights in the set of tasks to be followed up to generate a set of task weight information; generate a follow-up task scoring report based on the set of tasks to be followed up and the set of task weight information; identify task requirements in the follow-up task scoring report to generate follow-up task requirement information; and control an execution device to execute tasks based on the follow-up task requirement information.

[0137] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0138] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0139] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0140] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for identifying demand information based on voice information and a large model, characterized in that, include: Retrieve the dialogue voice text information for the consultation scenario from the audio recording storage database, wherein the dialogue voice text information is the text information after the dialogue voice between different users is converted into text; The dialogue voice text information and the preset prompt information set are input into the prompt word embedding model to obtain embedded structured text; Based on the embedded structured text, a set of task information to be followed up is generated; Task weights are identified in the set of tasks to be followed up in order to generate a set of task weight information. Based on the set of tasks to be followed up information and the set of tasks to be followed up weight information, a follow-up task scoring report is generated; The task requirement is identified in the follow-up task scoring report to generate follow-up task requirement information, and the execution device is controlled to execute the task based on the follow-up task requirement information.

2. The method according to claim 1, characterized in that, The step of retrieving dialogue voice and text information for the consultation scenario from the audio recording storage database includes: Collect voice dialogue information between different users in the consultation scenario; The dialogue voice information is input into the voiceprint recognition model to obtain user voice information, wherein the user voice information is voice information containing voice user tags; The character's voice information is transcribed into text to generate dialogue voice text information.

3. The method according to claim 1, characterized in that, The step of generating a set of task information to be followed up based on the embedded structured text includes: Keyword extraction is performed on the embedded structured text to generate a key information set of the structured text; The structured text key information set is input into the semantic annotation model to obtain the annotated structured text key information set; Syntactic analysis is performed on the labeled structured text key information set to generate an analyzed key information set, which serves as the information set for the task to be followed up.

4. The method according to claim 1, characterized in that, The step of generating a follow-up task scoring report based on the set of follow-up task information and the set of follow-up task weight information includes: The task information set to be followed up and the task weight information set to be followed up are merged to obtain a task summary information set; The task summary information set is populated with a report template to generate a populated task report; The populated task report is then validated to generate validation results. In response to the determination that the test result indicates that the test has passed, the corresponding filled task report is determined as the follow-up task scoring report.

5. The method according to claim 1, characterized in that, The step of inputting the dialogue voice text information and a preset prompt information set into the prompt word embedding model to obtain embedded structured text includes: The dialogue voice text information is segmented into words to generate a segmented voice text sequence; Stop words are removed from the segmented speech-text sequence to generate a speech-text sequence with stop words removed. The removed speech-text sequence is converted into a vector to obtain a speech-text vector sequence; The speech-text vector sequence is normalized to generate a processed speech-text vector sequence. The processed speech-text vector sequence is encoded to generate encoded text semantic information; The preset prompt information set is converted into a vector to obtain a prompt information vector set; Generate a sequence of prompt information vectors based on the set of prompt information vectors; The prompt information vector sequence is fused with the speech text vector sequence to obtain fused text semantic information; The fused text semantic information and the encoded text semantic information are transformed into structured information to obtain the transformed structured text, which is used as the embedded structured text.

6. A demand information recognition device based on voice information and a large model, characterized in that, include: The acquisition unit is configured to acquire dialogue voice-text information for a consultation scenario from a recording storage database, wherein the dialogue voice-text information is text information after the dialogue voice between different users is converted into text. The input unit is configured to input the dialogue voice text information and a preset prompt information set into the prompt word embedding model to obtain embedded structured text; The first generation unit is configured to generate a set of task information to be followed up based on the embedded structured text; The scoring unit is configured to identify task weights in the set of task information to be followed up, so as to generate a set of task weight information to be followed up. The second generation unit is configured to generate a follow-up task scoring report based on the follow-up task information set and the follow-up task weight information set. The identification unit is configured to identify task requirements from the follow-up task scoring report to generate follow-up task requirement information, and to control the execution device to execute the follow-up task requirement information.

7. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 5.

8. A computer-readable medium, characterized in that, It stores a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1 to 5.

Images