Work order abstract generation method, apparatus, device, medium, and product

Abstract

This application discloses a method for generating work order summaries, relating to the field of computer technology. The method includes the following steps: obtaining a target work order and a corresponding work order summary template; segmenting the dialogue text into text fragments based on the events described in the dialogue text of the target work order; locating statements corresponding to the business dimensions associated with slots in the work order summary template within the text fragments; generating slot content corresponding to the located statements; and filling the slots with the slot content to generate a work order summary for the target work order.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, device, medium and product for generating work order summaries. Background Technology

[0002] In customer service systems, it's necessary to generate work order summaries that conform to business specifications for customer service work orders such as complaint work orders. One related technology involves inputting a work order summary example and relevant business information into a large language model (MLM) via prompts. The MLM then generates a work order summary conforming to business specifications by referring to the example. However, due to the uncertainty of the MLM output, the generated work order summary is difficult to strictly adhere to the example and often fails to meet business specifications. Furthermore, customer service work orders such as complaint work orders typically include the entire voice interaction text between the user and the agent, which is usually quite long. Processing customer service work orders using the above method may result in inaccurate work order summaries, potentially containing issues such as misinterpretations, factual errors, and missing key information. Summary of the Invention

[0003] The purpose of this application is to provide a work order summary method, apparatus, device, medium, and product to improve the accuracy of work order summary generation and business compliance.

[0004] In a first aspect, embodiments of this application provide a method for generating a work order summary, the method comprising: Obtain the target work order and the corresponding work order summary template; the target work order includes the dialogue text between the user and the agent; Based on the events described in the dialogue text, the dialogue text is segmented into text fragments; Based on the business dimension associated with the slot in the work order summary template, locate the statement corresponding to the business dimension in the text fragment, and generate the slot content corresponding to the slot based on the located statement. The content of the slot is filled into the slot to generate a work order summary for the target work order.

[0005] Secondly, embodiments of this application provide a work order summary generation apparatus, the apparatus comprising: The acquisition module is used to acquire the target work order and the work order summary template corresponding to the target work order; the target work order includes the dialogue text between the user and the agent; The segmentation module is used to segment the dialogue text into text fragments based on the events described in the dialogue text; The generation module is used to locate the statement corresponding to the business dimension in the text fragment based on the business dimension associated with the slot in the work order summary template, and generate the slot content corresponding to the slot based on the located statement. The fill module is used to fill the slot with the content of the slot to generate a work order summary of the target work order.

[0006] Thirdly, embodiments of this application provide a work order summary generation device, the device comprising: A memory on which computer programs are stored; A processor is used to process the computer program to implement the above-described work order summary generation method.

[0007] Fourthly, embodiments of this application provide a readable storage medium having a computer program stored thereon, the computer program being executable by a processor to implement the above-described work order summary generation method.

[0008] Fifthly, embodiments of this application provide a computer program product, including a computer program that can be executed by a processor to implement the above-described work order summary generation method.

[0009] The work order summary generation method, apparatus, device, medium, and product provided in this application embodiment obtain a target work order and a corresponding work order summary template. Based on the events described in the dialogue text of the target work order, the dialogue text is segmented into text fragments. Based on the business dimensions associated with slots in the work order summary template, the statements corresponding to the business dimensions are located in the text fragments. Based on the located statements, the corresponding slot content is generated and the slot content is filled into the slot. This can improve the accuracy of work order summary generation while generating corresponding work order summaries that conform to business specifications according to the work order summary template, thereby improving the business standardization of work order summaries. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating a work order summary generation method according to an embodiment of this application; Figure 2 This is a schematic diagram of a work order summary template according to an embodiment of this application; Figure 3 This is a schematic diagram of a text segmentation process according to an embodiment of this application; Figure 4 This is a schematic diagram of the training process of a semantic relevance scoring model according to an embodiment of this application; Figure 5 This is a schematic diagram illustrating the process of determining the business dimension of slot association according to an embodiment of this application; Figure 6 This is a schematic diagram of a business dimension matching process according to an embodiment of this application; Figure 7 This is a schematic diagram of a text reconstruction process according to an embodiment of this application; Figure 8 This is a schematic diagram of the training process of a clue localization model according to an embodiment of this application; Figure 9 This is a schematic diagram of a slot content generation process according to an embodiment of this application; Figure 10 This is a schematic diagram of the structure of a conditional attention model according to an embodiment of this application; Figure 11 This is a schematic diagram of a work order summary generation device according to an embodiment of this application; Figure 12 This is a schematic diagram of a work order summary generation device according to an embodiment of this application. Detailed Implementation

[0012] This application provides a method, apparatus, device, medium, and product for generating work order summaries.

[0013] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0014] This application provides a method for generating work order summaries, which can be executed by an automated work order summary generation device to improve the accuracy and business compliance of the generated work order summaries.

[0015] Figure 1 This is a flowchart illustrating a work order summary generation method according to an embodiment of this application. The work order summary according to this embodiment can be a summary of the work order, such as a work order summary, which can be executed after the work order processing is completed. Figure 1 The method for generating work order summaries shown is used to generate work order summaries. For example... Figure 1As shown, a work order summary generation method according to an embodiment of this application may include the following steps: S102, obtain the target work order and the corresponding work order summary template. The target work order includes the dialogue text between the user and the agent.

[0016] The target work order is the work order for which a work order summary is to be generated, and it can be a customer service work order. The work order summary template corresponding to the target work order is the summary template applicable to the target work order, which can be determined according to the business type of the target work order and varies depending on the business type of the target work order.

[0017] As an example, the target work order can be a complaint work order, and its business type is complaint type. The work order summary template corresponding to the target work order can be a complaint type work order summary template, that is, a complaint work order summary template.

[0018] A target work order is a structured collection of information used to record, transfer, track, and resolve a specific issue. Its content may include: the order details, historical processing status, and the dialogue text between the user and the agent. The dialogue text can be the complete conversation between the user and the agent, transcribed from their voice calls, and can be very long.

[0019] Using the example above, if the target work order is a complaint work order, the user and the agent may have multiple long voice calls regarding the complaint in the complaint work order. The dialogue text in the complaint work order can be a transcript of all the conversations between the user and the agent regarding the complaint, which is usually a very long text.

[0020] A work order summary template is a structured text framework used to generate work order summaries that conform to business specifications. It can include one or more predefined slots, each of which can be associated with a business dimension. A business dimension is a perspective for observation and classification based on a specific business viewpoint or attribute. The business dimension can vary depending on the type of work order summary template.

[0021] Following the example above, the business dimensions of a complaint ticket summary template can include: business name, complaint issue, investigation status, handling plan, satisfaction level, and resolution status. The complaint ticket summary template can include slots corresponding to these business dimensions, with each slot associated with its corresponding business dimension. An example of a complaint ticket summary template could be as follows: Figure 2 As shown.

[0022] S104, based on the events described in the dialogue text, divide the dialogue text into text fragments.

[0023] In this context, an event can be an identifiable and meaningful unit of communication or activity within a dialogue. In practice, the dialogue text can be segmented into multiple sentences with independent semantics. Adjacent sentences describing the same event can be aggregated to obtain one or more text fragments. Each text fragment can correspond to an event, which is an independent situational unit, typically revolving around a core intent or theme, and possesses relatively independent and complete meaning.

[0024] For example, the dialogue text can be sequentially segmented into statements with independent semantics, including statement 1, statement 2, statement 3, statement 4, and statement 5. The semantic relevance between statement 2 and statement 1, statement 3 and statement 2, statement 4 and statement 3, and statement 5 and statement 4 can be calculated. If statement 2 and statement 1 describe the same event, statement 3 and statement 2 describe the same event, statement 4 and statement 3 do not describe the same event, and statement 5 and statement 4 describe the same event, then statements 1, 2, and 3 can be grouped into one text segment, and statements 4 and 5 can be grouped into one text segment.

[0025] In one implementation, a semantic relevance scoring model can be trained. After segmenting the dialogue text into multiple sentences with independent semantics, the semantic relevance scoring model can be used to score whether adjacent sentences describe the same event. If the score is greater than a scoring threshold, it is determined that the adjacent sentences describe the same event. Adjacent sentences describing the same event are then aggregated to obtain one or more text segments. An example of the segmentation process is as follows: Figure 3 As shown.

[0026] The semantic relevance scoring model can be obtained by fine-tuning a pre-trained language model using a dialogue dataset between users and agents. This dialogue dataset can be a labeled dataset. Each sample in the dataset can include a pair of adjacent statements and annotations indicating whether the adjacent statements describe the same event. This dialogue dataset can be used to perform supervised fine-tuning of the pre-trained language model to obtain the semantic relevance scoring model.

[0027] Annotated data can be constructed automatically based on historical dialogue texts between users and agents. For example, real historical dialogue texts between users and agents can be obtained from customer service system logs. These historical dialogue texts are then vectorized and clustered using clustering algorithms such as k-means, grouping semantically / topically similar dialogue texts into the same cluster. Then, within each cluster, each dialogue text undergoes noise reduction processing based on heuristic rules, such as stop word filtering and short sentence merging. Positive examples belonging to the same event are generated based on combinations of adjacent statements or statement units within the same dialogue text, while negative examples belonging to the same event are generated based on combinations of statements or statement units from different dialogue texts.

[0028] An example of the training process for a semantic relevance scoring model can be as follows: Figure 4 As shown.

[0029] In this embodiment, by segmenting the dialogue text into one or more text fragments based on the events described in the dialogue text, the semantic integrity of the text fragments can be preserved while reducing the length of subsequent text processing. This provides event-level context for subsequent sentence location, improves the accuracy of subsequent location, and facilitates the generation of accurate and business-compliant work order summaries for complaint work orders, such as those with excessively long dialogue texts.

[0030] S106: Based on the business dimension associated with the slot in the work order summary template, locate the statement corresponding to the business dimension in the text fragment, and generate the slot content corresponding to the slot based on the located statement.

[0031] The content of the slot corresponding to the slot can be obtained by summarizing the content of the located statement. It can be a summary of the statements corresponding to the business dimensions associated with the slot in each text fragment.

[0032] In implementation, the work order summary template can include one or more predefined slots, each associated with a business dimension. When the work order summary template has multiple slots, step S106 can be executed for each slot: based on the business dimension associated with that slot, the corresponding statement in each text fragment is located, and based on the located statement, the corresponding slot content is generated. Thus, corresponding slot content is generated for each slot.

[0033] Using the example above, one slot in the complaint ticket summary template can be associated with the business dimension "Business Name". The corresponding statement for "Business Name" can be located in each text fragment, and the content of the located statement can be summarized to generate a summary of the "Business Name". This summary is the content for that slot. Similarly, another slot in the complaint ticket summary template can be associated with the business dimension "Complaint Issue". The corresponding statement for "Complaint Issue" can be located in each text fragment, and the content of the located statement can be summarized to generate a summary of the "Complaint Issue". This summary is the content for that slot.

[0034] In this embodiment, the dialogue text is divided into one or more text segments based on the events described in the dialogue text. Statements related to the business dimension associated with the slot are located in each text segment. This can narrow the positioning range, reduce positioning noise, and improve positioning efficiency and accuracy through event-level contextual information.

[0035] S108, fill the slot with the slot content to generate a work order summary for the target work order.

[0036] In practice, when the work order summary template has multiple slots, the content corresponding to that slot can be filled into each slot. After filling all the slots with their corresponding content, the work order summary can be obtained.

[0037] In this embodiment, by obtaining the target work order and its corresponding work order summary template, and based on the events described in the dialogue text of the target work order, the dialogue text is segmented into text fragments. Based on the business dimensions associated with slots in the work order summary template, the statements corresponding to the business dimensions are located in the text fragments. Based on the located statements, information fragments corresponding to the slots are generated. Based on the information fragments corresponding to the slots, slot content for the corresponding slots is generated. The slot content for the corresponding slots is then filled into the corresponding slots. This method can improve the efficiency and accuracy of work order summary generation while generating work order summaries that conform to business specifications according to the work order summary template, thereby improving the business standardization of work order summaries.

[0038] In the above embodiments, the work order summary template can be a template with pre-associated business dimensions, in which each slot is pre-associated with a corresponding business dimension. In one embodiment, the work order summary template can be a template without business dimension association, where each slot is not associated with a business dimension. Before step S106, the association processing between slots and business dimensions can be performed. For example... Figure 5 As shown, the work order summary generation method may also include the following steps: S502, retrieve the work order summary description corresponding to the work order summary template.

[0039] The work order summary description can serve as a guiding or standard document for work order summaries, ensuring adherence to unified summary specifications and generating summaries with consistent content structure, controllable quality, and effective information.

[0040] In practice, the work order summary description can be determined based on the business type of the work order summary template. For example, if the work order summary template is a complaint work order summary template and its business type is complaint, the corresponding work order summary description can be a complaint work order summary description.

[0041] S504 breaks down the work order summary template into multiple slot context fragments.

[0042] In this context, the slot context fragment is a text fragment related to that slot that provides relevant information for filling it. In implementation, a work order summary template typically has multiple slots. Each slot in the work order summary template can be identified, and based on the identified slots and semantics, the work order summary template is decomposed into multiple slot context fragments. Each slot context fragment corresponds to one slot and is a text fragment related to that slot. This can be achieved using a large language model to identify each slot in the work order summary template and, based on the identified slots and semantics, decompose the work order summary template into multiple slot context fragments.

[0043] S506 breaks down the work order summary description into multiple business dimension segments.

[0044] In this context, "business dimension" refers to key information categories from a business perspective. "Business dimension fragment" refers to explanatory fragments related to that business dimension within the work order summary. In implementation, work order summaries typically contain multiple business dimensions. These business dimensions can be identified, and based on these identifications, the work order summary is broken down into multiple dimension fragments. Each business dimension fragment corresponds to one business dimension and is a text fragment related to that specific business dimension. This can be achieved using a large language model to identify the business dimensions in the work order summary and then break it down into multiple business dimension fragments.

[0045] S508 performs semantic matching between multiple slot context fragments and multiple business dimension fragments, and determines the business dimension associated with each slot based on the matching results.

[0046] Specifically, for each slot context fragment, semantic matching can be performed between the slot context fragment and each business dimension fragment. The business dimension fragment with the highest matching degree is determined as the business dimension fragment matched by the slot context fragment. Based on the business dimension fragment, the business dimension associated with the slot corresponding to the slot context fragment is determined.

[0047] In one embodiment, to achieve more accurate matching between slot context fragments and business dimension fragments and improve matching accuracy, when performing semantic matching between multiple slot context fragments and multiple business dimension fragments, for each slot context fragment, each business dimension fragment can be sequentially taken as the current business dimension fragment. Multiple models can be used to calculate the matching score between the slot context fragment and the current business dimension fragment. The matching scores calculated by multiple models can be weighted and summed to obtain the matching score between the slot context fragment and the current business dimension fragment. Among the calculated matching scores between the slot context fragment and each business dimension fragment, the business dimension fragment with the highest matching score is determined as the business dimension fragment matched by the slot context fragment.

[0048] The multiple models can be of different types, such as multiple models from the following categories: small language model, large language model, and vector model. The small language model can be a lightweight pre-trained language model, the large language model can be a generative large language model, and the vector model can be a text embedding model. An example of a business-dimensional matching process could be as follows: Figure 5 As shown. The summary description is a summary of the work order, and the summary template is a template for the work order summary.

[0049] In this embodiment, by obtaining the work order summary description corresponding to the work order summary template, the work order summary template is decomposed into multiple slot context fragments, and the work order summary description is decomposed into multiple business dimension fragments. Semantic matching is performed on the multiple slot context fragments and multiple business dimension fragments. Based on the matching results, the business dimension associated with each slot is determined. This allows for the dynamic establishment of the association between slots and business dimensions in the work order summary template based on semantic matching. When a new business template appears or an existing template changes, this method can automatically adapt slots and business dimensions dynamically through semantic understanding, quickly responding to diverse business needs without requiring extensive manual reconfiguration.

[0050] In the above embodiments, in step S106, the statement corresponding to the business dimension associated with the slot is located in each text fragment, and the slot content corresponding to the slot is generated based on the located statement. During the dialogue between the user and the agent, interruptions, interjections, and unclear expression logic often occur, leading to semantic ambiguity and loss in the dialogue text. In one embodiment, to improve the quality of the text fragments and facilitate the location of relevant statements, locating the statement corresponding to the business dimension in the text fragment in step S106 may include: performing text reconstruction processing on the text fragment, and locating the statement corresponding to the business dimension in the reconstructed text fragment.

[0051] Text reconstruction processing is used to rewrite text, re-express it, or adjust its structure while preserving its core semantics. In implementation, for each text segment, it can be reconstructed semantically. Then, the reconstructed text segment is validated to check for factual errors, illusions, missing information, errors in merging short sentences, etc. A correctness score is generated based on the validation results. If the correctness score is less than a threshold, the cause of the error is analyzed, and reconstruction is performed again based on the cause. If the correctness score is greater than or equal to the threshold, reconstruction stops. Thus, without changing the semantics, text segments are reconstructed into semantically coherent and logically clear expressions. Text segment reconstruction includes, but is not limited to: merging short sentences in a text segment, deleting short sentences from a text segment, and rewriting the text segment according to the contextual semantics. An example of the text reconstruction process is as follows: Figure 7 As shown.

[0052] In the above embodiment, in step S106, the statement corresponding to the business dimension is located in each text fragment. In practice, the statement corresponding to the business dimension can be located in each text fragment using a clue location model. For each business dimension associated with a slot, text fragment-business dimension pairs can be generated based on each text fragment and the business dimension. Based on the prompt word template and the text fragment-business dimension pairs, prompt words can be generated to prompt the location of the statement corresponding to the business dimension from the corresponding text fragment. Then, the prompt words can be input into the clue location model, and the statement corresponding to the business dimension can be located in the corresponding text fragment using the clue location model.

[0053] The cue localization model can be implemented based on a large language model. In one implementation, the cue localization model can be trained using the following two-stage training scheme.

[0054] Phase 1: Training the Clue Location Scoring Model. Specifically, the strong reasoning capabilities of the first major language model can be utilized to synthesize location scoring training data through multi-model voting. This data is then used to post-train the clue location scoring model, enabling it to distinguish the correctness of location results. The first major language model can be a reasoning model with hundreds of billions of parameters.

[0055] The second stage involves training the clue localization model. Specifically, based on the efficient inference speed of the second-largest language model, candidate training data for localization is generated. This candidate data is then quality-filtered using a pre-trained clue localization scoring model, and the high-quality data is used to reinforce the clue localization model. The second-largest language model can be a reasoning model with billions of parameters.

[0056] The clue location scoring model is used to determine whether the clue location result is correct. To ensure the model can accurately judge the clue location result, its training data can include positive and negative samples. The construction of positive and negative samples can specifically include the following two steps: Step one involves pairing text fragments with business dimensions to form "text fragment-business dimension" pairs. Based on a fixed location prompt word template, these pairs are combined to create prompt words indicating the location of the corresponding statement within the business dimension from the text fragments. These prompt words are then input multiple times into the first large language model to obtain several location results. The first large language model is a trillion-parameter inference model; although its inference speed is relatively slow, it possesses powerful inference capabilities. Using this model to generate location results can improve the confidence level of the results.

[0057] Step two: From the above location results, select the result with the highest frequency that is greater than the frequency threshold as the correct result, and regard the other results as incorrect results to construct positive and negative samples. If the highest frequency result occurs less than the frequency threshold, it can be considered that the confidence level of this result as a correct result is low, and this result is discarded to improve the quality of the sample.

[0058] By taking the above two steps, the sample quality of the clue location scoring model can be improved, thereby improving the judgment accuracy of the clue location scoring model.

[0059] An example of the training process for a clue localization model can be as follows: Figure 8 As shown in the figure. Among them, the work order interaction text is the historical dialogue text, and the summary dimension is the business dimension.

[0060] In the above embodiments, in step S108, the information fragments corresponding to the slots are summarized to generate the slot content. In one embodiment, to more accurately focus on relevant information, improve the accuracy of slot content generation, reduce the amount of computation for redundant information, and improve computation speed, in step S108, the business dimension corresponding to the slot can be used as conditional information. Through a conditional attention model, the located statements are summarized for the conditional information to generate the slot content.

[0061] For example, if a slot is associated with a dimension named "Processing Solution," then "Processing Solution" can be used as conditional information. The "Processing Solution" and the corresponding statement located for that slot can be input into a conditional attention model. The conditional attention mechanism calculates the relevance weights of each statement to "Processing Solution," and based on these relevance weights, the content of each statement is summarized to generate summary content focusing on "Processing Solution." This summary content is the slot content corresponding to that slot. An example of the slot content generation process can be as follows: Figure 9 As shown in the figure. The summary dimension is the business dimension, the clue content is the located statement, the summary result is the generated slot content, the summary template is the work order summary template, and the restoration result is the generated work order summary.

[0062] In implementation, the conditional attention model can be a large language model, and it can adopt methods such as... Figure 10 The model structure is shown below. Figure 10 As shown, this model can employ an improved Transformer Decoder architecture, which splits the Transformer Decoder structure into two parts: Transformer Decoder Light and Transformer Decoder Heavy. The Transformer Decoder Light structure contains fewer model layers and a smaller hidden layer dimension. When the current token undergoes an attention operation in the Transformer Decoder Light, all historical tokens are computed, ensuring the globality of the attention operation. The Transformer Decoder Heavy structure has more model layers and a larger hidden layer dimension. When the current token undergoes an attention operation in the Transformer Decoder Heavy structure, only a selected subset of tokens are computed. The Transformer Decoder Heavy contains a routing structure used to filter historical tokens that require attention computation.

[0063] The Transformer Decoder Light represents the current token by performing attention calculations on all historical decoded tokens. The Transformer Decoder Heavy represents the current token by performing attention calculations on only historical decoded tokens selected by routing.

[0064] The routing operation uses a linear gate to calculate the score of each historical decoded token, and selects the top K historical decoded tokens by inputting them into the Transformer Decoder Heavy module for calculation.

[0065] The Token Representation Aggregation calculates the final representation of the next token by weighted summation of the representations of the current token (encoded by Transformer Decoder Light and Transformer Decoder Heavy) and the representations of the current token (encoded by Transformer Decoder Heavy), which is used for token prediction.

[0066] The above describes a work order summary generation method according to an embodiment of this application. Based on the same idea, an embodiment of this application also provides a work order summary generation apparatus. Figure 11 This is a schematic diagram of a work order summary generation device according to an embodiment of this application. Figure 11 As shown, the work order summary generation device 1100 may include: an acquisition module 1110, a segmentation module 1120, a generation module 1130, and a filling module 1140.

[0067] The acquisition module 1110 is used to acquire the target work order and the work order summary template corresponding to the target work order; the target work order includes the dialogue text between the user and the agent.

[0068] The segmentation module 1120 is used to segment the dialogue text into text fragments based on the events described in the dialogue text.

[0069] The generation module 1130 is used to locate the statement corresponding to the business dimension in the text fragment based on the business dimension associated with the slot in the work order summary template, and generate the slot content corresponding to the slot based on the located statement. The filling module 1140 is used to fill the slot content into the slot to generate a work order summary of the target work order.

[0070] In one embodiment, the work order summary generation device 1100 further includes: an association module, used for: Obtain the work order summary description corresponding to the work order summary template; The work order summary template is decomposed into multiple slot context fragments, and each slot context fragment corresponds to one slot. The work order summary description is broken down into multiple business dimension segments, and each business dimension segment corresponds to one business dimension. Semantic matching is performed on the multiple slot context fragments and the multiple business dimension fragments, and the business dimension associated with each slot is determined based on the matching results.

[0071] In one implementation, the association module is specifically used for: For each slot context fragment, each business dimension fragment is sequentially taken as the current business dimension fragment. Multiple models are used to calculate the matching score between the slot context fragment and the current business dimension fragment. The matching scores calculated by the multiple models are weighted and summed to obtain the matching score between the slot context fragment and the current business dimension fragment. Among the calculated matching scores between the slot context fragment and each business dimension fragment, the business dimension fragment with the highest matching score is determined as the business dimension fragment that the slot context fragment matches. The multiple models include large language models, small language models, and vector models.

[0072] In one embodiment, the segmentation module 1120 is specifically used for: The dialogue text is segmented into multiple statements with independent semantics; A semantic relevance scoring model is used to score whether adjacent statements describe the same event. If the score is greater than the scoring threshold, it is determined that the adjacent statements describe the same event. Aggregate adjacent statements describing the same event to obtain one or more text fragments.

[0073] In one embodiment, the generation module 1130 is specifically used for: Perform text reconstruction processing on the text fragment; locate the statement corresponding to the business dimension in the text fragment after text reconstruction processing.

[0074] In one embodiment, the generation module 1130 is specifically used for: Using the business dimension associated with the slot as conditional information, the locationd statement is summarized based on the conditional information using a conditional attention model to generate the slot content.

[0075] The work order summary generation device of this application embodiment can be used as the execution subject of the work order summary generation method of this application embodiment to realize all the functions of the work order summary method of this application embodiment. Since the principle is the same, it will not be described again here.

[0076] The above describes a work order summary generation apparatus according to an embodiment of this application. Based on the same concept, this application also provides a work order summary generation device, such as... Figure 12 As shown.

[0077] The work order summary generation device can be an electronic device that performs the work order summary generation method according to the embodiments of this application.

[0078] The work order summary generation device can vary considerably due to differences in configuration or performance, and may include one or more memory 1210s and processors 1220s. The memory 1210 stores a computer program, and the processor 1220 processes the computer program to implement the work order summary generation method according to embodiments of this application.

[0079] Furthermore, one or more embodiments of this application also provide a storage medium for storing computer-executable instruction information. In a specific embodiment, the storage medium may be a USB flash drive, optical disc, hard disk, etc. When the computer-executable instruction information stored in the storage medium is executed by a processor, it can realize the work order summary generation method of the embodiments of this application.

[0080] Furthermore, one or more embodiments of this application also provide a computer program product, including a computer program that can be executed by a processor to implement the work order summary generation method of the embodiments of this application.

[0081] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0082] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must also be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0083] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0084] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0085] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, when implementing one or more embodiments of this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0086] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, one or more embodiments of this application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] Embodiments of this application are described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable parallel device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable parallel device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable fraud device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions can also be loaded onto a computer or other programmable device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0091] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0092] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0093] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0094] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, one or more embodiments of this application may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0095] One or more embodiments of this application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. One or more embodiments of this application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can reside in local and remote computer storage media, including storage devices.

[0096] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0097] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0098] It should be understood that the training and prediction processes of the AI ​​models involved in the various embodiments of this application all adhere to the principles of legal data sources, compliant data content, and compliant data governance, and comply with the requirements of Article 5 of the Patent Law. Specifically: Data source legitimacy: All datasets used for AI model training were obtained through legal means, covering three categories: publicly authorized data, data authorized by partners, and self-collected compliant data. Publicly authorized data comes from compliant data sources following open-source licenses such as Apache 2.0, with complete copyright attribution and authorization scope clearly marked, and no unauthorized open-source code or data reuse. Data authorized by partners has been subject to formal data usage agreements, clearly defining the scope, duration, and confidentiality obligations, and possessing a complete authorization chain. For self-collected data involving personal information, strict informed consent procedures have been followed, and anonymization processes (including but not limited to field masking, feature anonymization, and differential privacy technology applications) have been implemented to remove personally identifiable information, fully complying with the requirements of relevant laws and regulations such as the "Interim Measures for the Administration of Generative Artificial Intelligence Services" and the "Personal Information Protection Law."

[0099] Data content compliance: The AI ​​model's dataset undergoes multiple screenings and cleaning processes to remove all content that may violate social morality or harm public interests. It contains no obscene, pornographic, violent, discriminatory, or information that endangers national or public safety, nor does it involve the illegal acquisition or use of genetic resources. For data in sensitive fields (such as healthcare and finance), an additional privacy-preserving computation module (including federated learning and secure multi-party computation technologies) ensures that the data is "usable but not visible," avoiding compliance risks during the original data transmission process and ensuring that the data application scenarios and uses comply with public order and good morals and industry regulatory requirements.

[0100] Data governance norms: A complete data traceability system is established during the AI ​​model training process to automatically record the source, collection time, annotation process, cleaning rules, and permission allocation of training data, generating traceable compliance reports to ensure that the data is verifiable throughout its entire lifecycle. The dataset annotation process for AI models is completed by a professional human R&D team, clearly defining the proportion of human creative contributions and avoiding reliance on AI-generated data that has not undergone substantial human modification, thus meeting the examination requirements for "human main contributions" in AI patent applications.

Claims

1. A method for generating work order summaries, characterized in that, The method includes: Obtain the target work order and the corresponding work order summary template; the target work order includes the dialogue text between the user and the agent; Based on the events described in the dialogue text, the dialogue text is segmented into text fragments; Based on the business dimension associated with the slot in the work order summary template, locate the statement corresponding to the business dimension in the text fragment, and generate the slot content corresponding to the slot based on the located statement. The content of the slot is filled into the slot to generate a work order summary for the target work order.

2. The method according to claim 1, characterized in that, Before locating the statement corresponding to the business dimension in the text fragment based on the business dimension associated with the slot in the work order summary template, the method further includes: Obtain the work order summary description corresponding to the work order summary template; The work order summary template is decomposed into multiple slot context fragments, and each slot context fragment corresponds to one slot. The work order summary description is broken down into multiple business dimension segments, and each business dimension segment corresponds to one business dimension. Semantic matching is performed on the multiple slot context fragments and the multiple business dimension fragments, and the business dimension associated with each slot is determined based on the matching results.

3. The method according to claim 2, characterized in that, The semantic matching of the plurality of slot context fragments and the plurality of business dimension fragments includes: For each slot context fragment, each business dimension fragment is sequentially taken as the current business dimension fragment. Multiple models are used to calculate the matching score between the slot context fragment and the current business dimension fragment. The matching scores calculated by the multiple models are weighted and summed to obtain the matching score between the slot context fragment and the current business dimension fragment. Among the calculated matching scores between the slot context fragment and each business dimension fragment, the business dimension fragment with the highest matching score is determined as the business dimension fragment that the slot context fragment matches. The multiple models include large language models, small language models, and vector models.

4. The method according to claim 1, characterized in that, The step of segmenting the dialogue text into text fragments based on the events described in the dialogue text includes: The dialogue text is segmented into multiple statements with independent semantics; A semantic relevance scoring model is used to score whether adjacent statements describe the same event. If the score is greater than the scoring threshold, it is determined that the adjacent statements describe the same event. Aggregate adjacent statements describing the same event to obtain one or more text fragments.

5. The method according to claim 1, characterized in that, Locating the statement corresponding to the business dimension in the text fragment includes: The text fragment is then reconstructed. Locate the statement corresponding to the business dimension in the text fragment after text reconstruction.

6. The method according to claim 1, characterized in that, The process of generating the slot content corresponding to the slot based on the located statement includes: Using the business dimension associated with the slot as conditional information, the locationd statement is summarized based on the conditional information using a conditional attention model to generate the slot content.

7. A work order summary generation device, characterized in that, The device includes: The acquisition module is used to acquire the target work order and the work order summary template corresponding to the target work order; the target work order includes the dialogue text between the user and the agent; The segmentation module is used to segment the dialogue text into text fragments based on the events described in the dialogue text; The generation module is used to locate the statement corresponding to the business dimension in the text fragment based on the business dimension associated with the slot in the work order summary template, and generate the slot content corresponding to the slot based on the located statement. The fill module is used to fill the slot with the content of the slot to generate a work order summary of the target work order.

8. A work order summary generation device, characterized in that, The device includes: A memory on which computer programs are stored; A processor for processing the computer program to implement the work order summary generation method according to any one of claims 1 to 6.

9. A readable storage medium, characterized in that, It stores a computer program that can be executed by a processor to implement the work order summary generation method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes a computer program that can be executed by a processor to implement the work order summary generation method according to any one of claims 1 to 6.

Images