Work order allocation methods and devices, electronic equipment and storage media
By integrating the text and audio features of work orders, business and sentiment tags are generated. The model is then trained using historical data, which solves the problem of low accuracy in work order allocation and achieves precise allocation and efficient processing of work orders.
Patent Information
- Application Number
- CN202610845147.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-11
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, work order allocation relies on manual or static rules, resulting in low allocation accuracy and difficulty in accurately assigning work orders to the corresponding business personnel terminals.
By receiving and parsing work order text and audio, a feature fusion model is used to combine text and audio features to generate business tags and sentiment tags. Based on these tags, a work order allocation strategy is generated, including identifying business type and sentiment intensity. Historical work order data is used to train the model to improve allocation accuracy.
It achieves dual-dimensional intelligent recognition of work orders, improves the accuracy and efficiency of work order allocation, breaks through the traditional work order system's reliance on text information, and avoids priority misjudgment caused by the lack of emotional perception.
Smart Images

Figure CN123117572A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology or other related technical fields. Specifically, it relates to a work order allocation method and apparatus, electronic device and storage medium. Background Technology
[0002] A work order is a standardized electronic document used to record, track, and manage service requests or business tasks. It is typically created by the system or manually after a customer raises a question, requests a service, or reports an anomaly. It clarifies the content of the matter, attribution of responsibility, processing flow, and completion deadline. Essentially, a work order is a process-driven work mechanism that ensures every customer request or internal task is fully recorded, flows smoothly, and is closed within a specified timeframe. It is widely used in service industries such as finance and telecommunications.
[0003] In related technologies, work order allocation relies on manual assignment or automated engines based on static rules. For example, by extracting keywords from the work order, the work order is allocated according to pre-set rules. However, this method can only simply identify the content of the work order and allocate it accordingly. The accuracy is low, and it is difficult to accurately allocate the work order to the corresponding business personnel's terminal.
[0004] There is currently no effective solution to the above problems. Summary of the Invention
[0005] This invention provides a work order allocation method, apparatus, electronic device, and storage medium to at least solve the technical problem of low accuracy in work order allocation based on manual or static rules in related technologies.
[0006] According to one aspect of the present invention, a work order allocation method is provided, comprising: receiving a work order and parsing the work order to obtain work order text and work order audio; preprocessing the work order text and the work order audio to obtain text features and audio features; inputting the text features and the audio features into a feature fusion model, and performing feature fusion on the text features and the audio features through the feature fusion model to obtain fused features, wherein the feature fusion model is a pre-constructed neural network model for fusing multimodal features; inputting the fused features into a work order allocation model, configuring business tags and sentiment tags for the work order through the work order allocation model, and generating a work order allocation strategy based on the business tags and the sentiment tags, wherein the work order allocation model is a pre-constructed neural network model for allocating the work order.
[0007] Furthermore, the preprocessing steps for the work order text include: de-identifying the work order text; correcting the de-identified work order text by calling business terms from the business database; performing semantic verification on the corrected work order text using a large language model; and extracting features from the work order text that passes the verification.
[0008] Further, the preprocessing steps for the work order audio include: performing activity detection on the work order audio using an audio activity detection algorithm, and extracting valid audio based on the detection results; extracting acoustic features from the valid audio; inputting the valid audio into a speech recognition model, converting the audio content into a text sequence through the speech recognition model, and extracting features from the text sequence to obtain text sequence features, wherein the acoustic features and the text sequence features constitute audio features.
[0009] Furthermore, the step of fusing the text features and the audio features through the feature fusion model to obtain the fused features includes: using the text features as a query vector and the audio features as key-value pairs to construct cross-modal attention weights; using the feature fusion model to perform weighted aggregation of the audio features based on the cross-modal attention weights to generate an enhanced audio representation that is semantically aligned with the work order text; and using the feature fusion model to concatenate the enhanced audio representation with the text features to form the fused features.
[0010] Further, the step of configuring business tags and emotion tags for the business work order through the work order allocation model includes: identifying the first business type to which the business work order belongs through the work order allocation model, wherein the first business type includes at least one of the following: account management, loan service, service feedback; based on the first business type, identifying the second business type to which the business work order belongs through the work order allocation model, wherein the second business type is a subtype of the first business type; configuring the business tag for the business work order based on the first business type and the second business type; dividing the emotion intensity in the work order audio through the work order allocation model to obtain an emotion intensity level, and configuring the emotion tag for the business work order based on the emotion intensity level.
[0011] Furthermore, the step of generating a work order allocation strategy based on the business tag and the emotion tag includes: matching a preset list of processing terminals according to the business tag to determine the processing terminal with business permissions; determining the processing time limit according to the emotion tag; retrieving the processing strategy from the knowledge base based on the business tag and the emotion tag; and generating the work order allocation strategy based on the processing terminal, the processing time limit, and the processing strategy.
[0012] Furthermore, the work order allocation model is pre-built. The steps of building the work order allocation model include: obtaining historical business work orders and corresponding historical work order allocation strategies; performing feature processing on the historical business work orders to obtain fused features, and configuring labels on the historical business work orders based on the historical work order allocation strategies, wherein the feature processing includes at least feature extraction and feature alignment, and the label configuration includes at least business label configuration, sentiment label configuration, and allocation strategy configuration; constructing instruction response pairs based on the fused features and configured labels, wherein the instruction response pairs use the fused features as input instructions and the business labels, sentiment labels, and allocation strategies as output responses; inputting the instruction response pairs into the initial work order allocation model, and using a batch training mechanism to iteratively train the initial work order allocation model until the training batch is completed, thereby obtaining the trained work order allocation model, wherein during each batch of iterative training, the likelihood loss between the predicted output of the work order allocation model and the output response is calculated through backpropagation, and the parameters of the work order allocation model are adjusted based on the likelihood loss.
[0013] According to another aspect of the present invention, a work order allocation device is also provided, comprising: a receiving unit, configured to receive a work order and parse the work order to obtain work order text and work order audio; a processing unit, configured to preprocess the work order text and the work order audio respectively to obtain text features and audio features; a fusion unit, configured to input the text features and the audio features into a feature fusion model, and perform feature fusion on the text features and the audio features through the feature fusion model to obtain fused features, wherein the feature fusion model is a pre-built neural network model for fusing multimodal features; and an allocation unit, configured to input the fused features into a work order allocation model, configure business tags and sentiment tags for the work order through the work order allocation model, and generate a work order allocation strategy based on the business tags and the sentiment tags, wherein the work order allocation model is a pre-built neural network model for allocating the work order.
[0014] Furthermore, the processing unit includes: a first desensitization module for desensitizing the work order text; a first correction module for correcting the desensitized work order text by calling business terms from the business database; a first verification module for semantically verifying the corrected work order text using a large language model; and a first extraction module for extracting features from the verified work order text.
[0015] Furthermore, the processing unit further includes: a first detection module, used to perform activity detection on the work order audio using an audio activity detection algorithm, and extract valid audio based on the detection result; a second extraction module, used to extract acoustic features from the valid audio; and a third extraction module, used to input the valid audio into a speech recognition model, convert the audio content into a text sequence through the speech recognition model, and extract features from the text sequence to obtain text sequence features, wherein the acoustic features and the text sequence features constitute audio features.
[0016] Furthermore, the fusion unit includes: a first construction module, used to construct cross-modal attention weights by using the text features as query vectors and the audio features as key-value pairs through the feature fusion model; a first aggregation module, used to perform weighted aggregation of the audio features by the feature fusion model based on the cross-modal attention weights to generate an enhanced audio representation that is semantically aligned with the work order text; and a first concatenation module, used to concatenate the enhanced audio representation with the text features by the feature fusion model to form the fused feature.
[0017] Further, the allocation unit includes: a first identification module, configured to identify the first business type to which the business work order belongs through the work order allocation model, wherein the first business type includes at least one of the following: account management, loan service, service feedback; a second identification module, configured to identify the second business type to which the business work order belongs based on the first business type through the work order allocation model, wherein the second business type is a subtype of the first business type; a first configuration module, configured to configure the business tag for the business work order based on the first business type and the second business type; and a second configuration module, configured to divide the emotional intensity in the work order audio through the work order allocation model to obtain an emotional intensity level, and configure the emotional tag for the business work order based on the emotional intensity level.
[0018] Furthermore, the allocation unit further includes: a first determining module, configured to determine a processing terminal with business permissions by matching a preset list of processing terminals according to the business tag; a second determining module, configured to determine a processing time limit based on the emotion tag; a first retrieval module, configured to retrieve a processing strategy from a knowledge base based on the business tag and the emotion tag; and a first generating module, configured to generate the work order allocation strategy based on the processing terminal, the processing time limit, and the processing strategy.
[0019] Furthermore, the work order allocation device further includes: a first acquisition module, used to acquire historical business work orders and corresponding historical work order allocation strategies; a first processing module, used to perform feature processing on the historical business work orders to obtain fused features, and to configure labels on the historical business work orders based on the historical work order allocation strategies, wherein the feature processing includes at least feature extraction and feature alignment, and the label configuration includes at least business label configuration, sentiment label configuration, and allocation strategy configuration; a second construction module, used to construct instruction response pairs based on the fused features and configured labels, wherein the instruction response pairs use the fused features as input instructions and the business labels, sentiment labels, and allocation strategies as output responses; and a first training module, used to input the instruction response pairs into an initial work order allocation model, and to iteratively train the initial work order allocation model using a batch training mechanism until the training batch is completed, thereby obtaining the trained work order allocation model, wherein during each batch of iterative training, the likelihood loss between the predicted output of the work order allocation model and the output response is calculated through backpropagation, and the parameters of the work order allocation model are adjusted based on the likelihood loss.
[0020] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute any of the above-described work order allocation methods.
[0021] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any of the above-described work order allocation methods.
[0022] According to another aspect of the present invention, a computer program product is also provided, the computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements any of the above-described work order allocation methods.
[0023] In this application, the following steps are performed: receiving a work order and parsing it to obtain work order text and audio; preprocessing the work order text and audio to obtain text features and audio features; inputting the text features and audio features into a feature fusion model to fuse the text features and audio features to obtain fused features, wherein the feature fusion model is a pre-built neural network model for fusing multimodal features; inputting the fused features into a work order allocation model to configure business tags and sentiment tags for the work order, and generating a work order allocation strategy based on the business tags and sentiment tags, wherein the work order allocation model is a pre-built neural network model for allocating work orders.
[0024] This application systematically integrates text and audio information to achieve a joint understanding of the semantic intent and emotional state of business work orders, forming text and audio features. A finely tuned network model, based on massive historical work orders and expert-annotated data, learns the intrinsic relationship between business semantics and emotional response, identifies the primary and secondary business categories of work orders, and simultaneously judges the customer's emotional level. This comprehensive approach generates a precise work order allocation strategy, overcoming the traditional work order system's reliance on text information and the priority misjudgment caused by the lack of emotional perception. It achieves dual-dimensional intelligent recognition and automated response to customer requests, improving classification accuracy and thus enhancing work order allocation accuracy and efficiency. This solves the technical problem of low accuracy in work order allocation methods based on manual or static rules in related technologies. Attached Figure Description
[0025] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:
[0026] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a work order allocation method is shown.
[0027] Figure 2 This is a flowchart of an optional work order allocation method according to an embodiment of the present invention;
[0028] Figure 3 This is a schematic diagram of an optional work order allocation process according to an embodiment of the present invention;
[0029] Figure 4 This is a schematic diagram of an optional work order allocation device according to an embodiment of the present invention;
[0030] Figure 5This is a hardware structure block diagram of an electronic device (or mobile device) for an optional work order allocation method according to an embodiment of the present invention. Detailed Implementation
[0031] 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 should fall within the scope of protection of the present invention.
[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0033] It should be noted that the work order allocation method and apparatus in this application can be used in the field of artificial intelligence technology to automatically allocate business work orders based on artificial intelligence, and can also be used in any field other than the field of artificial intelligence technology to automatically allocate business work orders based on artificial intelligence. This application does not limit the application field of the work order allocation method and apparatus.
[0034] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with relevant laws, regulations, and standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding access points are provided for users to choose to authorize or refuse. For example, interfaces are set up between this system and relevant users or organizations, providing users with corresponding access points to choose to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.
[0035] The following embodiments of the present invention can be applied to various work order allocation systems / applications / devices. The present invention relates to a work order allocation mechanism based on multimodal data of audio and text and intelligent classification of large models. By integrating the text description and audio information of work orders, it achieves fine-grained classification of work orders and dynamically optimizes work order allocation strategies and processing time, thereby improving the accuracy of work order allocation and business processing efficiency.
[0036] The present invention will now be described in detail with reference to various embodiments.
[0037] Example 1
[0038] According to an embodiment of the present invention, an embodiment of a work order allocation method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.
[0039] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a work order allocation method is shown. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.
[0040] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).
[0041] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the work order allocation method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned work order allocation method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
[0042] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.
[0043] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).
[0044] Under the aforementioned operating environment, this application provides the following: Figure 2 The work order allocation method shown is implemented by a work order allocation system. The work order allocation method of this application can be applied to the allocation of business work orders in the financial field.
[0045] Figure 2 This is a flowchart of an optional work order allocation method according to an embodiment of the present invention, such as... Figure 2As shown, the method includes the following steps:
[0046] Step S201: Receive the service work order and parse the service work order to obtain the work order text and work order audio.
[0047] In step S201 above, the system receives a work order from the customer service channel. This work order is a multimodal data packet, containing text description information submitted by the customer and the original audio file associated with the work order. The system automatically parses the work order data packet through an internal interface, separating the structured text fields and the unstructured audio stream, and outputting them as independent work order text and work order audio, respectively.
[0048] Unified access to service requests from different sources ensures that subsequent processing modules can simultaneously acquire both text and voice information, avoiding the omission of key semantic clues conveyed by customers in their voice messages due to reliance on a single modality. This lays a data foundation for subsequent multimodal analysis. The work order text refers to the problem content described by the customer in natural language, while the work order audio refers to the raw sound wave signals collected during the conversation between the customer and customer service personnel. Together, they constitute a complete expression of the service intent.
[0049] Step S202: Preprocess the work order text and work order audio respectively to obtain text features and audio features.
[0050] In step S202 above, the work order text and audio output from step S201 are independently preprocessed to extract structured, computable text and audio features. For the work order text, preprocessing includes at least de-identification, technical terminology correction, semantic verification, and feature extraction. After these processes, the work order text is transformed into semantically clear and formatted text features, serving as the foundational input for subsequent semantic understanding. For the work order audio, preprocessing includes at least audio activity detection, streaming feature extraction, and text sequence conversion, ultimately outputting structured audio features whose dimensions include temporal acoustic parameters and semantic transcription information, reflecting the customer's emotional expression and semantic features.
[0051] In one optional embodiment, the preprocessing steps for the work order text include: de-identifying the work order text; correcting the de-identified work order text by calling business terms from the business database; performing semantic verification on the corrected work order text using a large language model; and extracting features from the work order text that passes the verification.
[0052] Furthermore, when preprocessing the work order text, the work order text is first de-identified. Through a predefined set of regular expression patterns, sensitive personal information contained in the text, such as ID card numbers, bank card numbers, and mobile phone numbers, is automatically identified and masked, and replaced with uniform placeholders to prevent customer personal information from being illegally leaked or misused during the system process.
[0053] Secondly, the de-identified work order text is corrected by calling business terms from the business library. That is, based on the pre-built standard terminology dictionary, colloquial, non-standard or vague expressions are mapped to standardized business terms. The terminology normalization is achieved by calculating the minimum edit distance match. The purpose is to eliminate the semantic ambiguity caused by the diversity of expressions and improve the recognizability and consistency of the text in subsequent semantic analysis.
[0054] Subsequently, the corrected work order text is semantically validated using a large language model. A lightweight large language model is used here to judge the semantic integrity and logical rationality of the corrected text, identify and filter invalid content with semantic breaks, disordered word order, and no actual meaning, and retain only statements with clear service requests to ensure that the text in subsequent processing has clear, effective, and understandable business semantics and avoid noise interfering with model inference.
[0055] Finally, feature extraction is performed on the verified work order texts. A pre-trained text embedding model is used to transform the anonymized, corrected and verified text sequence into a high-dimensional semantic vector, forming structured text features. These features encode the core business intent of the work order in the semantic space, such as account freezing and loan approval delay, providing accurate semantic input for the post-feature fusion model.
[0056] Through the above steps, while ensuring data security and compliance, the original, messy, and colloquial customer input is transformed into standardized, semantically clear, and noise-free high-quality text features. This significantly improves the usability and expressive accuracy of text information in multimodal systems, providing a reliable language understanding foundation for accurate work order classification and intelligent dispatch.
[0057] In one optional embodiment, the preprocessing steps for work order audio include: performing activity detection on the work order audio using an audio activity detection algorithm, extracting valid audio based on the detection results; extracting acoustic features from the valid audio; inputting the valid audio into a speech recognition model, converting the audio content into a text sequence using the speech recognition model, and extracting features from the text sequence to obtain text sequence features, wherein the acoustic features and the text sequence features constitute the audio features.
[0058] Furthermore, when preprocessing the work order audio, the audio is first tested for activity using an audio activity detection algorithm. This algorithm uses energy threshold and spectrum change analysis mechanism to identify whether there is effective speech activity in the audio signal, filters out non-dialogue segments such as long silences, background noise, and key sounds, and retains only speech segments with semantic carrying capacity. The output is a number of continuous effective audio segments, avoiding invalid signals from interfering with the accuracy and efficiency of feature extraction.
[0059] Secondly, acoustic features are extracted from the valid audio clips. For each valid audio segment, a series of acoustic parameters related to emotion, speech rate, and tone are calculated, including fundamental frequency, mean and variance of energy, zero crossover rate, Mel-frequency cepstral coefficients and their first-order differences, forming a temporal-dimensional acoustic feature vector. These features reflect the customer's emotional fluctuations, tension, and intensity of expression during the speech process. For example, faster speech rate and higher pitch usually correspond to anxiety or dissatisfaction. The purpose is to quantify the nonverbal emotional cues implied in the speech, serving as a key basis for judging the urgency of the customer's needs.
[0060] Subsequently, the valid audio is input into a speech recognition model. This model employs a convolutional neural network architecture and possesses streaming processing capabilities, enabling it to convert speech signals segment by segment into corresponding text transcription content, generating a text sequence aligned with the audio time. The same embedding model used for work order text is then applied to this text sequence for feature extraction, outputting semantically encoded text sequence features. This process restores the speech content to a computable semantic expression, achieving a unified representation of speech and text information in the semantic space, and providing a semantic bridge for subsequent cross-modal alignment.
[0061] Step S203: Input the text features and audio features into the feature fusion model, and perform feature fusion on the text features and audio features to obtain fused features.
[0062] The aforementioned feature fusion model is a pre-built neural network model used to fuse multimodal features.
[0063] In step S203 above, the text features and audio features output from step S202 are used as input and fed into a pre-constructed feature fusion model. The core mechanism of this feature fusion model is a cross-attention mechanism, used to establish semantic associations between text features and audio features. Subsequently, the feature fusion model concatenates or fuses the aligned text features and enhanced audio features element-wise, and outputs a unified, semantically dense fused feature through layer normalization and linear transformation. This fused feature not only includes the business semantics of the work order but also embeds the emotional intensity of the customer when expressing the request, forming a joint representation that combines semantic and emotional features.
[0064] In one optional embodiment, the steps of fusing text features and audio features through a feature fusion model to obtain fused features include: using text features as query vectors and audio features as key-value pairs to construct cross-modal attention weights; using the feature fusion model to perform weighted aggregation of audio features based on the cross-modal attention weights to generate an enhanced audio representation that is semantically aligned with the work order text; and using the feature fusion model to concatenate the enhanced audio representation with the text features to form fused features.
[0065] Furthermore, during feature fusion, firstly, based on a cross-attention mechanism, text features are used as query vectors to actively guide the model to focus on the acoustic segments in the audio features most relevant to the current semantics; while audio features, as key-value pairs, provide cues of tone, rhythm, and emotion that correspond to the text semantics. By calculating attention weights, the model dynamically weights the audio features, achieving semantic alignment of text-controlled audio, so that the originally independent acoustic fluctuations and text semantics form a complementary expression in a unified vector space.
[0066] Secondly, the feature fusion model weights and sums the audio feature vector sequences according to attention weights to obtain an audio representation that is synchronously enhanced with the text semantics. This representation not only retains the original acoustic features (such as speech rate and pitch fluctuations) but is also given contextual labels consistent with the text semantics. Thus, the originally isolated acoustic features are transformed into emotional semantic embeddings with semantic context, making the audio no longer just a sound signal but an emotional supplement to the text's appeal.
[0067] Finally, the feature fusion model concatenates the enhanced audio representation with the text features to form a fused feature. The concatenation process directly links the encoding results of the two modalities along the vector dimension, ensuring that the final output fused feature simultaneously contains the precise semantic content of the work order and its corresponding emotional intensity. This fused feature, as a unified multimodal representation, preserves the structured semantics of the text while incorporating nonverbal emotional cues from the audio, resulting in a complete model of customer intent.
[0068] Step S204: Input the fused features into the work order allocation model, configure business tags and sentiment tags for business work orders through the work order allocation model, and generate work order allocation strategies based on business tags and sentiment tags.
[0069] In step S204 above, the fused features are input into the work order allocation model. This work order allocation model is a pre-built neural network model based on a large language model architecture, used to allocate business work orders. Its input layer receives high-dimensional fused features, and internally, through multi-layer nonlinear transformations and attention mechanisms, it achieves joint semantic parsing of business intent and emotional state. The model first identifies and outputs business labels based on the semantic content encoded in the fused features. These labels are structured classification results, including primary labels such as credit card business, loan business, customer service and complaints, and secondary labels such as bill dispute handling and repayment plan adjustment. This achieves accurate classification of the specific business types involved in the work order, ensuring that subsequent processing units have professional matching.
[0070] Meanwhile, based on the emotional intensity information embedded in the fusion features, the model outputs an emotion label, which is a five-level discrete score representing different emotional states. Specifically, it can include calm, slight dissatisfaction, moderate dissatisfaction, high anxiety, and extreme anger. The value is obtained by the model through collaborative analysis of acoustic indicators such as fundamental frequency fluctuation, speech rate change, and energy variance in the enhanced audio representation with text semantics. The setting of emotion labels aims to quantify the customer's emotional state and provide an objective basis for prioritizing service response.
[0071] Based on this, the work order allocation model automatically outputs work order allocation strategies through the strategy generation module, based on the joint input of business tags and sentiment tags. The work order allocation strategies specifically include: recommended processing units, estimated processing time, and recommended processing methods.
[0072] In one optional embodiment, the work order allocation model is pre-built. The steps of building the work order allocation model include: obtaining historical business work orders and corresponding historical work order allocation strategies; performing feature processing on the historical business work orders to obtain fused features, and configuring labels for the historical business work orders based on the historical work order allocation strategies, wherein the feature processing includes at least feature extraction and feature alignment, and the label configuration includes at least business label configuration, sentiment label configuration, and allocation strategy configuration; constructing instruction-response pairs based on the fused features and configured labels, wherein the instruction-response pairs use the fused features as input instructions and the business labels, sentiment labels, and allocation strategies as output responses; inputting the instruction-response pairs into the initial work order allocation model, and using a batch training mechanism to iteratively train the initial work order allocation model until the training batch is completed, thereby obtaining a trained work order allocation model, wherein during each batch of iterative training, the likelihood loss between the predicted output and the output response of the work order allocation model is calculated through backpropagation, and the parameters of the work order allocation model are adjusted based on the likelihood loss.
[0073] Furthermore, when constructing the work order allocation model, firstly, historical business work orders and their corresponding historical work order allocation strategies are obtained. The historical business work orders include text descriptions that have been anonymized and original audio records. The historical work order allocation strategies are the actual processing results executed manually or by the system, including the allocation unit, processing time limit, processing personnel, and final disposal method, providing the model with supervision signals under real business scenarios.
[0074] Secondly, historical work orders undergo feature processing to obtain fused features, and labels are configured for these work orders based on historical work order allocation strategies. Feature processing includes feature extraction and feature alignment. Feature extraction generates text features using a lightweight language model, extracts acoustic features using liveness detection technology and acoustic feature extraction tools, and then fuses them into unified fused features via a cross-attention module. Feature alignment maps features from different sources to the same semantic space, ensuring consistent input dimensions. Label configuration corresponds to three types of outputs: business label configuration classifies work orders into predefined primary and secondary business categories (e.g., primary label: loan business; secondary label: repayment plan adjustment); sentiment label configuration labels sentiment levels based on acoustic features in the audio and sentiment words in the text; and allocation strategy configuration structures historical allocation results into triplets of processing unit, response time, and processing method. This transforms the original work orders into structured samples that the model can learn, achieving end-to-end mapping modeling of input and output.
[0075] Subsequently, instruction-response pairs are constructed based on the fusion features and configured labels. These pairs take the fusion features as input instructions and business labels, sentiment labels, and allocation strategies as output responses, forming training samples such as [fusion features] → {Business label: Loan business - Repayment plan adjustment; Sentiment label: 4; Allocation strategy: Customer service group two, within 24 hours, suggest appeasement and submitting an extension application template}. The purpose of constructing instruction-response pairs is to model the work order processing task as an instruction-following language generation problem, enabling the model to understand and generate compliant and accurate processing decisions based on context, thereby enhancing the model's generalization ability and interpretability.
[0076] Finally, the instruction response is input into the initial work order allocation model, and iterative training is performed using a batch training mechanism. During each batch training, the model performs forward prediction on the fused features of the input, outputting the joint probability distribution of the corresponding business label, sentiment label, and allocation strategy. The negative log-likelihood loss between the predicted output and the true label is calculated to measure the deviation between the model's generated result and the true strategy. The model parameters are then updated using a backpropagation algorithm to gradually reduce the prediction error. Training continues until a preset number of rounds are completed, at which point the loss value tends to converge, resulting in the trained work order allocation model.
[0077] Therefore, by driving with real historical data, guiding with structured labels, modeling with instruction response pairs, and supervising training to the end, an intelligent work order allocation model with semantic understanding, emotion perception, and strategy generation capabilities was constructed.
[0078] In one optional embodiment, the steps of configuring business tags and emotion tags for business work orders using a work order allocation model include: identifying a first business type to which the business work order belongs using the work order allocation model, wherein the first business type includes at least one of the following: account management, loan service, service feedback; identifying a second business type to which the business work order belongs using the work order allocation model based on the first business type, wherein the second business type is a subtype of the first business type; configuring business tags for the business work order based on the first business type and the second business type; and classifying the emotion intensity in the work order audio using the work order allocation model to obtain an emotion intensity level, and configuring emotion tags for the business work order based on the emotion intensity level.
[0079] Furthermore, in real-time work order allocation scenarios, firstly, the work order allocation model identifies the primary business type to which the work order belongs. This primary business type is a coarse-grained business classification, including types such as account management, loan services, and service feedback. The model completes the first-level coarse-grained classification based on the collaborative discrimination of semantic keywords and acoustic emotion features in the fused features. This narrows the scope of business and improves classification efficiency and stability.
[0080] Secondly, based on the first business type, the work order allocation model further identifies the second business type to which the work order belongs, dividing the work order into more granular subcategories. For example, when the first business type is account management, the model further determines it as a secondary label such as mobile APP login failure or biometric verification failure based on specific semantic and contextual clues in the fused features; when the first type is loan services, the model combines keywords such as incomplete application materials and disputes over overdue interest with the anxious tone in the customer's voice to pinpoint specific subcategories such as loan application material consultation or disputes over overdue interest. Through this hierarchical classification mechanism, it ensures that business labels achieve accurate matching between macro-classification and micro-level needs, avoiding problems such as label ambiguity or overgeneralization.
[0081] After completing the two-layer identification of business types, the model combines the first business type with the second business type into a structured business label, such as Account Management - Biometric (fingerprint / face) verification failure. This label serves as a professional classification identifier for work orders and is used for subsequent targeted assignment to the corresponding business function group.
[0082] Simultaneously, the emotional intensity in the work order audio is categorized using a work order allocation model to obtain emotional intensity levels. This process is based on the co-weighting of acoustic parameters extracted from fusion features (such as fundamental frequency standard deviation, mean energy, and speech rate change rate) and emotional words in the text, outputting a discrete level from 1 to 5, including calm (no emotion), slight dissatisfaction, somewhat dissatisfied, highly anxious, and extremely angry. This emotional intensity level classification is based on expert rules labeled during the training phase and the distribution statistics learned by the model, aiming to transform subjective emotions into quantifiable and comparable objective indicators. Finally, based on this emotional intensity level, the model assigns emotional tags to business work orders as dynamic weighting factors for service priority, used for decision optimization.
[0083] In one optional embodiment, the step of generating a work order allocation strategy based on business tags and sentiment tags includes: matching a preset list of processing terminals according to the business tags to determine the processing terminals with business permissions; determining the processing time limit based on the sentiment tags; retrieving the processing strategy from the knowledge base based on the business tags and sentiment tags; and generating a work order allocation strategy based on the processing terminal, the processing time limit, and the processing strategy.
[0084] Furthermore, when generating work order allocation strategies, the system first matches a pre-defined list of processing terminals based on business tags to determine the processing terminals with the necessary business permissions. The system has a built-in permission mapping table that pre-configures the binding relationships between various business tags and their corresponding processing units. This ensures that work orders are only pushed to functional departments with the appropriate business qualifications and operating permissions, preventing processing delays caused by misassignment, unauthorized processing, or mismatched professional capabilities.
[0085] Secondly, processing time limits are determined based on emotion tags. Emotion tags are a five-level discrete rating system, with response time thresholds strictly corresponding to each emotion level. For example, an emotion tag of 1 corresponds to a processing time limit of 72 hours; an emotion tag of 3 shortens the time limit to 24 hours; and an emotion tag of 5 is forcibly compressed to within 2 hours. This mechanism dynamically adjusts service priority based on emotion level, ensuring rapid responses to customers with high-emotional states and preventing escalation of risks. This transforms customer subjective emotions into actionable system response constraints, enabling flexible scheduling of service resources.
[0086] Subsequently, processing strategies are retrieved from a knowledge base based on business tags and sentiment tags. This knowledge base is a structured strategy database that stores standardized operating procedures refined from historical processing experience. Each strategy includes recommended scripts, process guidelines, required materials, and system operation paths. This automates the embedding of experiential knowledge into the decision-making process, improving processing standardization and consistency, and reducing human judgment errors.
[0087] Finally, a work order allocation strategy is generated based on the processing terminal, processing time limit, and processing policy. This strategy is a structured output containing three core elements: recommended processing unit, mandatory response time limit, and suggested execution plan. Outputted in a machine-readable format, this strategy is directly pushed to the work order scheduling system to drive subsequent automatic work order dispatch and monitoring alerts. This achieves end-to-end automation of accurate identification, intelligent decision-making, and closed-loop execution.
[0088] Through the above steps, a work order is received and parsed to obtain work order text and audio. The work order text and audio are preprocessed to obtain text features and audio features, respectively. These features are then input into a feature fusion model, which fuses the text and audio features to obtain fused features. The feature fusion model is a pre-built neural network model for fusing multimodal features. Finally, the fused features are input into a work order allocation model, which assigns business tags and sentiment tags to the work order and generates a work order allocation strategy based on these tags. The work order allocation model is also a pre-built neural network model for allocating work orders.
[0089] In this embodiment, by systematically integrating two heterogeneous information sources, text and audio, a joint understanding of the semantic intent and emotional state of business work orders is achieved, forming text features and audio features. A network model finely tuned by instructions is employed to learn the inherent correlation between business semantics and emotional response based on massive historical work order data and expert-annotated data. This identifies the primary and secondary business categories to which the work order belongs and simultaneously determines the customer's emotional level, thereby comprehensively generating a precise work order allocation strategy. This overcomes the traditional work order system's reliance on textual information and the priority misjudgment caused by the lack of emotional perception, achieving dual-dimensional intelligent identification and automated response to customer demands. This improves classification accuracy, thereby enhancing the accuracy and efficiency of work order allocation. Ultimately, this solves the technical problem of low accuracy in work order allocation methods based on manual or static rules in related technologies.
[0090] The following describes in detail another optional implementation method.
[0091] Figure 3 This is a schematic diagram of an optional work order allocation process according to an embodiment of the present invention, such as... Figure 3 As shown, the work order allocation process flows from the data layer to the model layer, and finally to the business processing layer. Specifically, for new work orders, multimodal data is collected to obtain work order text and audio, and the data is preprocessed to transform the text data and audio data into text features and audio features that the model can recognize.
[0092] Subsequently, text features and audio features are used as input to the model layer. The work order allocation model is used to analyze the content and perform business segmentation. First-level business tags and second-level business tags are configured for each work order. Then, sentiment analysis is performed and sentiment tags are configured. At the same time, a work order allocation strategy is generated based on the knowledge base matching processing strategy.
[0093] The work order allocation strategy is transferred to the business processing layer for processing. According to the strategy, the business work orders are dispatched to the corresponding business terminals, and the effectiveness of the processed work orders is evaluated.
[0094] This invention relates to a work order allocation mechanism based on multimodal data of audio and text and intelligent classification of large models. By integrating the text description and audio information of work orders, it achieves fine-grained classification of work orders and dynamically optimizes work order allocation strategies and processing time, thereby improving the accuracy of work order allocation and business processing efficiency.
[0095] The following is a detailed description with reference to another embodiment.
[0096] Example 2
[0097] The work order allocation device provided in this embodiment includes multiple implementation units, each of which corresponds to a specific implementation step in the above embodiment one. The specific implementation method and beneficial effects can be referred to the foregoing method embodiment, and will not be repeated here.
[0098] Figure 4 This is a schematic diagram of an optional work order allocation device according to an embodiment of the present invention, such as... Figure 4 As shown, the work order allocation device may include: a receiving unit 41, a processing unit 42, a fusion unit 43, and an allocation unit 44, wherein,
[0099] The receiving unit 41 is used to receive the service work order and parse the service work order to obtain the work order text and the work order audio.
[0100] Processing unit 42 is used to preprocess the work order text and work order audio respectively to obtain text features and audio features;
[0101] The fusion unit 43 is used to input text features and audio features into the feature fusion model, and to fuse the text features and audio features through the feature fusion model to obtain fused features. The feature fusion model is a pre-built neural network model for fusing multimodal features.
[0102] The allocation unit 44 is used to input the fused features into the work order allocation model, configure business tags and sentiment tags for business work orders through the work order allocation model, and generate work order allocation strategies based on business tags and sentiment tags. The work order allocation model is a pre-built neural network model for allocating business work orders.
[0103] The aforementioned work order allocation device receives work orders through receiving unit 41 and parses them to obtain work order text and audio. Processing unit 42 preprocesses the work order text and audio to obtain text features and audio features. Fusion unit 43 inputs the text and audio features into a feature fusion model, which fuses the text and audio features to obtain fused features. The feature fusion model is a pre-built neural network model for fusing multimodal features. Allocation unit 44 inputs the fused features into a work order allocation model, which configures business tags and sentiment tags for the work orders and generates a work order allocation strategy based on the business tags and sentiment tags. The work order allocation model is a pre-built neural network model for allocating work orders.
[0104] In this embodiment, by systematically integrating two heterogeneous information sources, text and audio, a joint understanding of the semantic intent and emotional state of business work orders is achieved, forming text features and audio features. A network model finely tuned by instructions is employed to learn the inherent correlation between business semantics and emotional response based on massive historical work order data and expert-annotated data. This identifies the primary and secondary business categories to which the work order belongs and simultaneously determines the customer's emotional level, thereby comprehensively generating a precise work order allocation strategy. This overcomes the traditional work order system's reliance on textual information and the priority misjudgment caused by the lack of emotional perception, achieving dual-dimensional intelligent identification and automated response to customer demands. This improves classification accuracy, thereby enhancing the accuracy and efficiency of work order allocation. Ultimately, this solves the technical problem of low accuracy in work order allocation methods based on manual or static rules in related technologies.
[0105] Furthermore, the processing unit includes: a first desensitization module for desensitizing the work order text; a first correction module for correcting the desensitized work order text by calling business terms from the business database; a first verification module for performing semantic verification on the corrected work order text using a large language model; and a first extraction module for extracting features from the verified work order text.
[0106] Furthermore, the processing unit also includes: a first detection module, used to perform activity detection on the work order audio using an audio activity detection algorithm, and extract valid audio based on the detection results; a second extraction module, used to extract acoustic features from the valid audio; and a third extraction module, used to input the valid audio into a speech recognition model, convert the audio content into a text sequence through the speech recognition model, and extract features from the text sequence to obtain text sequence features, wherein the acoustic features and the text sequence features constitute the audio features.
[0107] Furthermore, the fusion unit includes: a first construction module, used to construct cross-modal attention weights by using text features as query vectors and audio features as key-value pairs through a feature fusion model; a first aggregation module, used to perform weighted aggregation of audio features based on cross-modal attention weights by the feature fusion model to generate an enhanced audio representation that is semantically aligned with the work order text; and a first splicing module, used to splice the enhanced audio representation with text features by the feature fusion model to form fused features.
[0108] Further, the allocation unit includes: a first identification module, used to identify the first business type to which the business work order belongs through a work order allocation model, wherein the first business type includes at least one of the following: account management, loan service, service feedback; a second identification module, used to identify the second business type to which the business work order belongs based on the first business type through a work order allocation model, wherein the second business type is a subtype of the first business type; a first configuration module, used to configure business tags for the business work order based on the first business type and the second business type; and a second configuration module, used to classify the emotional intensity in the work order audio through the work order allocation model to obtain an emotional intensity level, and configure an emotional tag for the business work order based on the emotional intensity level.
[0109] Furthermore, the allocation unit also includes: a first determining module, used to determine the processing terminal with business permissions by matching a preset list of processing terminals according to the business tags; a second determining module, used to determine the processing time limit based on the sentiment tags; a first retrieval module, used to retrieve the processing strategy from the knowledge base based on the business tags and sentiment tags; and a first generating module, used to generate a work order allocation strategy based on the processing terminal, processing time limit, and processing strategy.
[0110] Furthermore, the work order allocation device also includes: a first acquisition module, used to acquire historical business work orders and corresponding historical work order allocation strategies; a first processing module, used to perform feature processing on historical business work orders to obtain fused features, and to configure labels for historical business work orders based on historical work order allocation strategies, wherein feature processing includes at least feature extraction and feature alignment, and label configuration includes at least business label configuration, sentiment label configuration, and allocation strategy configuration; a second construction module, used to construct instruction response pairs based on fused features and configured labels, wherein the instruction response pairs use fused features as input instructions and business labels, sentiment labels, and allocation strategies as output responses; and a first training module, used to input instruction response pairs into an initial work order allocation model, and to iteratively train the initial work order allocation model using a batch training mechanism until the training batch is completed, thereby obtaining a trained work order allocation model, wherein during each batch of iterative training, the likelihood loss between the predicted output and the output response of the work order allocation model is calculated through backpropagation, and the parameters of the work order allocation model are adjusted based on the likelihood loss.
[0111] It should be noted that the receiving unit 41, processing unit 42, fusion unit 43, and allocation unit 44 mentioned above correspond to steps S201 to S204 in Embodiment 1. The instances and application scenarios implemented by the above units and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules or units can also be part of a device and run in the computer terminal 10 provided in Embodiment 1.
[0112] The invention will now be described in conjunction with another alternative embodiment.
[0113] Example 3
[0114] The present invention can also provide an electronic device. Figure 5 This is a hardware structure block diagram of an electronic device (or mobile device) for an optional work order allocation method according to an embodiment of the present invention, such as... Figure 5 As shown, the electronic device may include: one or more ( Figure 5 (Only one is shown) processor 502, memory 504, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.
[0115] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
[0116] The processor can access information and applications stored in the memory via a transmission device to perform the following steps: receiving a work order and parsing it to obtain work order text and audio; preprocessing the work order text and audio to obtain text features and audio features; inputting the text features and audio features into a feature fusion model to fuse the text features and audio features, obtaining fused features, wherein the feature fusion model is a pre-built neural network model for fusing multimodal features; inputting the fused features into a work order allocation model to configure business tags and sentiment tags for the work order, and generating a work order allocation strategy based on the business tags and sentiment tags, wherein the work order allocation model is a pre-built neural network model for allocating work orders.
[0117] The processor can access information and applications stored in the memory via the transmission device to perform the following steps: de-identify the work order text; correct the de-identified work order text by calling business terms from the business database; perform semantic verification on the corrected work order text using a large language model; and extract features from the work order text that passes the verification.
[0118] The processor can access the information and application programs stored in the memory via the transmission device to perform the following steps: perform activity detection on the work order audio using an audio activity detection algorithm, and extract valid audio based on the detection results; extract acoustic features from the valid audio; input the valid audio into a speech recognition model, convert the audio content into a text sequence through the speech recognition model, and extract features from the text sequence to obtain text sequence features, wherein the acoustic features and the text sequence features constitute the audio features.
[0119] The processor can access the information and application stored in the memory via the transmission device to perform the following steps: using text features as query vectors and audio features as key-value pairs through a feature fusion model to construct cross-modal attention weights; using the feature fusion model to perform weighted aggregation of audio features based on the cross-modal attention weights to generate an enhanced audio representation that is semantically aligned with the work order text; and using the feature fusion model to concatenate the enhanced audio representation with the text features to form a fused feature.
[0120] The processor can access information and applications stored in the memory via a transmission device to perform the following steps: identifying the first business type to which the business work order belongs through a work order allocation model, wherein the first business type includes at least one of the following: account management, loan service, service feedback; based on the first business type, identifying the second business type to which the business work order belongs through the work order allocation model, wherein the second business type is a subtype of the first business type; configuring business tags for the business work order based on the first business type and the second business type; classifying the emotional intensity in the work order audio through the work order allocation model to obtain an emotional intensity level, and configuring an emotional tag for the business work order based on the emotional intensity level.
[0121] The processor can access information and applications stored in the memory via a transmission device to perform the following steps: matching a pre-defined list of processing terminals based on business tags to determine processing terminals with the necessary business permissions; determining processing time limits based on sentiment tags; retrieving processing strategies from the knowledge base based on business tags and sentiment tags; and generating work order allocation strategies based on processing terminals, processing time limits, and processing strategies.
[0122] The processor can access information and applications stored in memory via a transmission device to execute the following steps: acquiring historical work orders and their corresponding historical work order allocation strategies; performing feature processing on historical work orders to obtain fused features, and configuring labels for historical work orders based on the historical work order allocation strategies, wherein feature processing includes at least feature extraction and feature alignment, and label configuration includes at least business label configuration, sentiment label configuration, and allocation strategy configuration; constructing instruction-response pairs based on the fused features and configured labels, wherein the instruction-response pairs use the fused features as input instructions and the business labels, sentiment labels, and allocation strategies as output responses; inputting the instruction-response pairs into an initial work order allocation model, and using a batch training mechanism to iteratively train the initial work order allocation model until the training batch is completed, resulting in a trained work order allocation model, wherein during each batch of iterative training, the likelihood loss between the predicted output and the output response of the work order allocation model is calculated through backpropagation, and the parameters of the work order allocation model are adjusted based on the likelihood loss.
[0123] This invention provides a work order allocation scheme. By systematically integrating text and audio information, it achieves a joint understanding of the semantic intent and emotional state of business work orders, forming text and audio features. Using a finely tuned network model, based on massive historical work order data and expert-annotated data, it learns the inherent correlation between business semantics and emotional response, identifies the primary and secondary business categories of the work order, and simultaneously judges the customer's emotional level. This comprehensively generates a precise work order allocation strategy, breaking through the traditional work order system's reliance on text information and overcoming priority misjudgments caused by the lack of emotional perception. It achieves dual-dimensional intelligent identification and automated response to customer requests, improving classification accuracy and thus enhancing work order allocation accuracy and efficiency. This solves the technical problem of low accuracy in work order allocation methods based on manual or static rules in related technologies.
[0124] Those skilled in the art will understand that Figure 5 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 5 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 5 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 5 The different configurations shown.
[0125] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.
[0126] The invention will now be described in conjunction with another alternative embodiment.
[0127] Example 4
[0128] This invention also provides a computer-readable storage medium. Optionally, in this invention, the computer-readable storage medium can be used to store the program code executed by the work order allocation method provided in Embodiment 1.
[0129] Optionally, in this embodiment of the invention, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.
[0130] This invention also provides a computer program product, which, when executed on a data processing device, is suitable for performing the steps of a work order allocation method: receiving a work order and parsing it to obtain work order text and audio; preprocessing the work order text and audio to obtain text features and audio features; inputting the text features and audio features into a feature fusion model to fuse the text features and audio features to obtain fused features, wherein the feature fusion model is a pre-built neural network model for fusing multimodal features; inputting the fused features into a work order allocation model to configure business tags and sentiment tags for the work order, and generating a work order allocation strategy based on the business tags and sentiment tags, wherein the work order allocation model is a pre-built neural network model for allocating work orders.
[0131] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0132] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0133] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.
[0134] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0135] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0136] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.
[0137] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method of work order distribution, characterized by, include: Receive a service work order and parse the service work order to obtain the work order text and work order audio; The work order text and the work order audio are preprocessed separately to obtain text features and audio features; The text features and the audio features are input into the feature fusion model, and the text features and the audio features are fused by the feature fusion model to obtain fused features. The feature fusion model is a pre-built neural network model for fusing multimodal features. The fused features are input into the work order allocation model, which configures business tags and sentiment tags for the business work orders. A work order allocation strategy is generated based on the business tags and sentiment tags. The work order allocation model is a pre-built neural network model for allocating the business work orders.
2. The method of claim 1, wherein, The steps for preprocessing the work order text include: The work order text is anonymized; The de-identified work order text is corrected by calling business terms from the business library; The corrected work order text was semantically validated using a large language model. Feature extraction is performed on the verified work order text.
3. The method of claim 1, wherein, The steps for preprocessing the audio of the work order include: The work order audio is subjected to activity detection using an audio activity detection algorithm, and valid audio is extracted based on the detection results; Extract acoustic features from the valid audio; The valid audio is input into a speech recognition model, which converts the audio content into a text sequence. The text sequence is then used to extract features to obtain text sequence features, wherein the acoustic features and the text sequence features constitute audio features.
4. The method of claim 1, wherein, The steps of fusing the text features and the audio features using the feature fusion model to obtain the fused features include: The feature fusion model uses the text features as query vectors and the audio features as key-value pairs to construct cross-modal attention weights. The feature fusion model performs weighted aggregation of the audio features based on the cross-modal attention weights to generate an enhanced audio representation that is semantically aligned with the work order text; The enhanced audio representation is concatenated with the text features by the feature fusion model to form the fused feature.
5. The method of claim 1, wherein, The steps of configuring business tags and sentiment tags for the business work orders using the work order allocation model include: The work order allocation model identifies the first business type to which the business work order belongs, wherein the first business type includes at least one of the following: account management, loan service, and service feedback; Based on the first business type, the second business type to which the business work order belongs is identified through the work order allocation model, wherein the second business type is a subtype of the first business type; Configure the service tag for the service work order based on the first service type and the second service type; The emotional intensity in the work order audio is divided by the work order allocation model to obtain an emotional intensity level, and the emotional tag is configured for the work order based on the emotional intensity level.
6. The method of claim 1, wherein, The steps for generating a work order allocation strategy based on the business tags and the sentiment tags include: Based on the business tag, a preset list of processing terminals is matched to determine the processing terminal with the business permission; The processing time limit is determined based on the aforementioned emotion tags; Based on the business tags and the emotion tags, a knowledge base retrieval and processing strategy is employed. The work order allocation strategy is generated based on the processing terminal, the processing time limit, and the processing strategy.
7. The method of claim 1, wherein, The work order allocation model is pre-built, and the steps for building the work order allocation model include: Obtain historical work orders and the corresponding historical work order allocation strategies; The historical work orders are subjected to feature processing to obtain fused features, and the historical work orders are labeled based on the historical work order allocation strategy. The feature processing includes at least feature extraction and feature alignment, and the label configuration includes at least business label configuration, sentiment label configuration and allocation strategy configuration. Based on the fusion features and the configured tags, an instruction-response pair is constructed, wherein the instruction-response pair takes the fusion features as input instructions and takes business tags, sentiment tags and allocation strategies as output responses; The instruction response pair is input into the initial work order allocation model. A batch training mechanism is used to iteratively train the initial work order allocation model until the training batch is completed, resulting in the trained work order allocation model. During each batch of iterative training, the likelihood loss between the predicted output of the work order allocation model and the output response is calculated through backpropagation, and the parameters of the work order allocation model are adjusted based on the likelihood loss.
8. A work order distribution apparatus characterized by comprising: include: The receiving unit is used to receive the service work order and parse the service work order to obtain the work order text and the work order audio. The processing unit is used to preprocess the work order text and the work order audio respectively to obtain text features and audio features; A fusion unit is used to input the text features and the audio features into a feature fusion model, and to perform feature fusion on the text features and the audio features through the feature fusion model to obtain fused features. The feature fusion model is a pre-built neural network model for fusing multimodal features. The allocation unit is used to input the fused features into the work order allocation model, configure business tags and sentiment tags for the business work orders through the work order allocation model, and generate a work order allocation strategy based on the business tags and sentiment tags. The work order allocation model is a pre-built neural network model for allocating the business work orders.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the work order allocation method according to any one of claims 1 to 7.
10. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the work order allocation method according to any one of claims 1 to 7.
11. A computer program product, characterized in that, The computer program product includes a computer program, wherein the computer program, when executed by a processor, implements the work order allocation method according to any one of claims 1 to 7.
