Text processing method and computing device

Abstract

The text processing method comprises the following steps: receiving a text for describing a flow of a task; a first clause group is obtained from the text used for describing the process of the task, the first clause group comprises a plurality of clauses arranged according to the occurrence sequence of the process, and each clause in the plurality of clauses comprises a workflow component; determining a workflow label corresponding to each clause in the first clause group; generating the workflow of the task according to the workflow labels corresponding to the multiple clauses in the first clause group; a visual representation of the generated workflow is generated for display by the workflow generation application.

Description

Cross-reference to related applications

[0001] This application claims the benefit of U.S. Patent Application No. 18 / 480,342, filed October 3, 2023, the disclosure of which is incorporated herein by reference in its entirety. Technical Field

[0002] This invention relates to the field of natural language processing (NLP) technology, and more particularly to a text processing method and a computing device. Background Technology

[0003] With the development of artificial intelligence (AI) technology, using AI models for natural language processing has become a major focus in the industry.

[0004] Currently, the Transformer model is the most widely used AI model in the field of natural language processing. In the Transformer model, which includes an encoder and a decoder, the encoder processes the input text, converting it into numerical vectors (called hidden state vectors) that are computer-understandable but not human-understandable. These hidden state vectors contain contextual information related to the input text. This process is called natural language understanding (NLU). Subsequently, the encoder sends these hidden state vectors to the decoder, which uses these vectors to generate responses similar to human language—in other words, human-like responses. This process is called natural language generation (NLG).

[0005] Natural language processing is increasingly focusing on enhancing the comprehensibility of its output for humans. Finding methods to generate human-understandable descriptions remains a field worthy of further research. Summary of the Invention

[0006] In a first aspect, a text processing method is provided. When applied to a computing device, the method includes: receiving text describing a process for which a workflow is to be generated, as input to a workflow generation application executed on the computing device; obtaining a first clause group from the text describing the process of the task, the first clause group including a plurality of clauses arranged in the order in which the process occurs, each of the plurality of clauses in the first clause group including a workflow component, the workflow component indicating information of a corresponding functional unit among a plurality of functional units for performing the task; determining a workflow tag corresponding to each clause in the first clause group, the workflow tag corresponding to the plurality of clauses in the first clause group including a first workflow tag indicating a workflow component or a second workflow tag indicating a workflow mode; generating the workflow of the task based on the workflow tags corresponding to the plurality of clauses in the first clause group; and generating a visual representation of the generated workflow for display by the workflow generation application.

[0007] In one possible implementation, obtaining the first clause group from the text describing the process of the task includes: decomposing the text describing the process of the task to obtain a third clause group, each of the plurality of clauses included in the third clause group including at most one workflow component; filtering the plurality of clauses in the third clause group to obtain a fourth clause group, each of the plurality of clauses included in the fourth clause group including one workflow component; and reordering the plurality of clauses in the fourth clause group to obtain the first clause group.

[0008] In one possible implementation, obtaining the first clause group from the text describing the process of the task includes: decomposing the text describing the process of the task to obtain a third clause group, each of the plurality of clauses included in the third clause group including at most one workflow component; filtering the plurality of clauses in the third clause group to obtain a fourth clause group, each of the plurality of clauses included in the fourth clause group including a workflow component; reordering the plurality of clauses in the fourth clause group to obtain a second clause group; and inserting at least one identifier of at least one workflow pattern into the second clause group to obtain the first clause group, the at least one workflow pattern being used to indicate the structure of the process of the task and the relationship between the plurality of functional units.

[0009] In one possible implementation, decomposing the text used to describe the process of the task to obtain the third clause group includes: performing sentence boundary detection on the text to obtain multiple sentences of the text; and splitting each of the multiple sentences of the text into one or more clauses using a first encoder-based model to obtain the third clause group.

[0010] In one possible implementation, filtering the plurality of clauses in the third clause group to obtain the fourth clause group includes: determining, by a second encoder-based model, whether each clause in the plurality of clauses in the third clause group includes a workflow component; if the clause in the third clause group includes a zero workflow component, then deleting the clause in the third clause group to obtain the fourth clause group.

[0011] In one possible implementation, reordering the plurality of clauses in the fourth clause group includes: reordering the plurality of clauses in the fourth clause group by means of a third encoder-based model, wherein the input of the third encoder-based model is two clauses in the fourth clause group, and the output of the third encoder-based model is information indicating whether the order of the two clauses is correct.

[0012] In one possible implementation, the identifier of each workflow mode in the at least one workflow mode includes a workflow mode boundary indication sentence and / or a workflow mode indication; inserting the at least one identifier of the at least one workflow mode into the second clause group to obtain the first clause group includes: determining the at least one workflow mode of the second clause group; for each workflow mode in the at least one workflow mode, obtaining the first clause group by performing at least one of the following on the second clause group: if there is no clause in the second clause group for indicating the boundary of the workflow mode, then inserting the workflow mode boundary indication sentence of the workflow mode into the second clause group; or if there is no clause in the second clause group that serves as the workflow mode indication, then inserting the workflow mode indication of the workflow mode into the second clause group.

[0013] In one possible implementation, determining the at least one workflow pattern of the second clause group includes: performing pattern keyword detection on a plurality of clauses in the second clause group; and determining the at least one workflow pattern based on the detected at least one pattern keyword, wherein each of the at least one pattern keyword is used to indicate a workflow pattern in the at least one workflow pattern.

[0014] In one possible implementation, determining the workflow tag corresponding to each clause in the first clause group includes: matching each clause in the first clause group with a plurality of workflow tags included in a tag dictionary to determine the workflow tag corresponding to each clause in the first clause group from the tag dictionary. Workflow tags are simplified representations of workflow components and workflow patterns used to generate workflows.

[0015] In one possible implementation, the tag dictionary further includes predefined values ​​corresponding to the plurality of workflow tags; matching each clause in the first clause group with the plurality of workflow tags included in the tag dictionary to determine the workflow tag corresponding to each clause in the first clause group from the tag dictionary includes: performing semantic similarity processing to match each clause in the first clause group with the predefined values ​​in the tag dictionary; and determining the workflow tag corresponding to the predefined value with the highest similarity score for the clause among the predefined values ​​as the workflow tag corresponding to the clause.

[0016] In a second aspect, a computing device is provided. The computing device includes a memory and at least one processor connected to the memory. The memory stores computer instructions that, when executed by the at least one processor, cause the computing device to: receive text describing a flow for which a workflow is to be generated, as input to a workflow generation application executed on the computing device; obtain a first set of clauses from the text describing the flow of the task, the first set of clauses including a plurality of clauses arranged in the order in which the flow occurs, each of the plurality of clauses in the first set of clauses including a workflow component, the workflow component indicating information of a corresponding functional unit among a plurality of functional units for performing the task; determine a workflow label corresponding to each clause in the first set of clauses, the workflow label corresponding to the plurality of clauses in the first set of clauses including a first workflow label indicating a workflow component or a second workflow label indicating a workflow mode; generate a workflow for the task based on the workflow label corresponding to the plurality of clauses in the first set of clauses; and generate a visual representation of the generated workflow for display by the workflow generation application.

[0017] In one possible implementation, the computer instructions, when executed by the at least one processor, cause the computing device to: decompose the text describing the flow of the task into a third clause group, each of the plurality of clauses included in the third clause group comprising at most one workflow component; filter the plurality of clauses in the third clause group to obtain a fourth clause group, each of the plurality of clauses included in the fourth clause group comprising a workflow component; reorder the plurality of clauses in the fourth clause group to obtain a first clause group, or reorder the plurality of clauses in the fourth clause group to obtain a second clause group, and insert at least one identifier of at least one workflow pattern into the second clause group to obtain the first clause group, the at least one workflow pattern being used to indicate the structure of the flow of the task and the relationship between the plurality of functional units.

[0018] In one possible implementation, the computer instructions, when executed by the at least one processor, cause the computing device to: perform sentence boundary detection on the text to obtain multiple sentences of the text; and split each of the multiple sentences of the text into one or more clauses using a first encoder-based model to obtain the third clause group.

[0019] In one possible implementation, the computer instructions, when executed by the at least one processor, cause the computing device to: determine, by a second encoder-based model, whether each of the plurality of clauses in the third clause group includes a workflow component; if the clause in the third clause group includes a zero workflow component, then delete the clause in the third clause group to obtain the fourth clause group.

[0020] In one possible implementation, the computer instructions, when executed by the at least one processor, cause the computing device to: reorder the plurality of clauses in the fourth clause group by means of a third encoder-based model, wherein the input of the third encoder-based model is two clauses in the fourth clause group, and the output of the third encoder-based model is information indicating whether the order of the two clauses is correct.

[0021] In one possible implementation, the identifier of each workflow mode in the at least one workflow mode includes a workflow mode boundary indication sentence and / or a workflow mode indication. The computer instructions, when executed by the at least one processor, cause the computing device to: determine the at least one workflow mode of the second clause group; for each workflow mode in the at least one workflow mode, obtain the first clause group by performing at least one of the following on the second clause group: if there is no clause in the second clause group indicating the boundary of the workflow mode, insert the workflow mode boundary indication sentence of the workflow mode into the second clause group; or if there is no clause in the second clause group serving as the workflow mode indication, insert the workflow mode indication of the workflow mode into the second clause group.

[0022] In one possible implementation, the computer instructions, when executed by the at least one processor, cause the computing device to: perform pattern keyword detection on a plurality of clauses in the second clause group; and determine the at least one workflow pattern based on the detected at least one pattern keyword, each of the at least one pattern keyword indicating a workflow pattern in the at least one workflow pattern.

[0023] In one possible implementation, the computer instructions, when executed by the at least one processor, cause the computing device to: match each clause in the first clause group with a plurality of workflow tags included in a tag dictionary to determine from the tag dictionary the workflow tag corresponding to each clause in the first clause group.

[0024] In one possible implementation, the tag dictionary further includes predefined values ​​corresponding to the plurality of workflow tags. The computer instructions, when executed by the at least one processor, cause the computing device to: perform semantic similarity processing to match each clause in the first clause group with the predefined values ​​in the tag dictionary; and determine the workflow tag corresponding to the predefined value among the predefined values ​​that has the highest similarity score for the clause as the workflow tag corresponding to the clause.

[0025] Thirdly, a non-transitory computer-readable storage medium is provided. The non-transitory computer-readable storage medium stores computer instructions. When executed by a computer, the computer causes the computer to: receive text describing a process for which a workflow is to be generated, as input to a workflow generation application executed on the computing device; obtain a first set of clauses from the text describing the process of the task, the first set of clauses comprising a plurality of clauses arranged in the order in which the process occurs, each of the plurality of clauses in the first set of clauses comprising a workflow component, the workflow component indicating information of a corresponding functional unit among a plurality of functional units for performing the task; determine a workflow label corresponding to each clause in the first set of clauses, the workflow label corresponding to the plurality of clauses in the first set of clauses including a first workflow label indicating a workflow component or a second workflow label indicating a workflow mode; generate a workflow for the task based on the workflow label corresponding to the plurality of clauses in the first set of clauses; and generate a visual representation of the generated workflow for display by the workflow generation application.

[0026] In one possible implementation, the computer instructions, when executed by a computer, cause the computer to: decompose the text describing the flow of the task into a third clause group, each of the plurality of clauses included in the third clause group comprising at most one workflow component; filter the plurality of clauses in the third clause group to obtain a fourth clause group, each of the plurality of clauses included in the fourth clause group comprising a workflow component; reorder the plurality of clauses in the fourth clause group to obtain a first clause group, or reorder the plurality of clauses in the fourth clause group to obtain a second clause group, and insert at least one identifier of at least one workflow pattern into the second clause group to obtain the first clause group, the at least one workflow pattern being used to indicate the structure of the flow of the task and the relationship between the plurality of functional units. Attached Figure Description

[0027] Figure 1 This is an architecture diagram of an intelligent integrated system provided in some embodiments of the present invention; Figure 2 This is a flowchart of a text processing method provided in some embodiments of the present invention; Figure 3 This is a schematic diagram of the hardware structure of a mobile phone provided in some embodiments of the present invention; Figure 4 This is a schematic diagram of a text processing method provided in some embodiments of the present invention; Figure 5AThis is a schematic diagram of another text processing method provided in some embodiments of the present invention; Figure 5B This is a schematic diagram of yet another text processing method provided in some embodiments of the present invention; Figure 6 These are structural diagrams of encoder-based models provided in some embodiments of the present invention; Figure 7 This is a structural diagram of another encoder-based model provided in some embodiments of the present invention; Figure 8 This is a structural diagram of another encoder-based model provided in some embodiments of the present invention; Figure 9 This is a schematic diagram of yet another text processing method provided in some embodiments of the present invention; Figure 10 This is a schematic diagram of a visual representation of a workflow provided in some embodiments of the present invention; Figure 11 These are structural diagrams of computing devices provided in some embodiments of the present invention. Detailed Implementation

[0028] The technical solutions of some embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. However, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art according to the embodiments of the present invention should be included within the protection scope of the present invention.

[0029] Unless the context otherwise requires, throughout the specification and claims, the term "comprise" and its other forms, such as the third-person singular "comprises" and the present participle "comprising," are interpreted as open-ended and inclusive, meaning "including but not limited to." In the specification, terms such as "one embodiment," "some embodiments," "exemplary embodiment," "example," "specific example," or "some examples" are intended to indicate that a particular feature, structure, material, or characteristic associated with one or more embodiments or examples is included in at least one embodiment or example of the invention. The illustrative representations of the above terms do not necessarily refer to the same one or more embodiments or examples. Furthermore, a particular feature, structure, material, or characteristic may be included in any suitable manner in any one or more embodiments or examples.

[0030] In the description of some embodiments, the term "connection" and its derivatives may be used. For example, in describing some embodiments, the term "connection" may be used to indicate that two or more components are in direct physical or electrical contact with each other. However, the term "connection" or "communication connection" may also refer to two or more components that are not in direct contact with each other but still cooperate or interact with each other. The embodiments disclosed herein are not necessarily limited to those described herein.

[0031] The term "and / or" describes the association of related objects only, including three cases. For example, "A and / or B" refers to three cases: A alone, A and B alone, and B alone.

[0032] In the following text, the terms “first” and “second” are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, features defined with “first” and “second” may explicitly or implicitly include one or more features, and the terms “first” and “second” are not used to describe a specific order of objects. For example, “first clause group” and “second clause group” are used to distinguish different clause groups, not to describe a specific order of clause groups.

[0033] In the description of embodiments of the present invention, unless otherwise stated, the terms "multiple", "aplurality of" or "the plurality of" refer to two or more, and "multiple", "a plurality of" or "the plurality of" can also be described as "at least two".

[0034] As used herein, depending on the context, the term "if" may optionally be interpreted as "when," "in the case of," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determined" or "if [the condition or event] is detected" may optionally be interpreted as "when determined," "in response to determination," "when [the condition or event] is detected," or "in response to detection."

[0035] The use of the phrases “applies to” or “is used for” in this article is an open and inclusive language that does not exclude the applicability to or use of devices for performing additional tasks or steps.

[0036] Furthermore, the phrase “according to” as used in this article has an open and inclusive meaning, because a process, step, calculation or other action “according to” one or more of the stated conditions or values ​​may in practice be based on additional conditions or values ​​beyond the stated conditions or values.

[0037] Some embodiments of this invention provide a text processing method. This text processing method is a natural language processing (NLP) method, whose main purpose is to convert text describing a task's process into a concise workflow that is easily understood by humans, based on artificial intelligence (AI) technology. This method represents a new breakthrough in natural language generation (NLG).

[0038] The following section will first explain some of the technical terms involved in the text processing method and device provided in the embodiments of the present invention.

[0039] 1. Workflow A workflow is an abstract and generalized description of the various operational steps of a task according to certain rules (such as the order and logical relationships of the steps). In computers, workflows are represented by workflow models, such as workflow diagrams.

[0040] By using workflows, tasks can be broken down into different subtasks (each subtask includes one or more operation steps), and tasks can be executed and monitored according to certain rules and processes, making it easier and more efficient to control and manage the task process.

[0041] 2. Workflow Components Workflow components are defined as information related to functional units used to perform tasks. This information includes workflow entities and workflow attributes. A workflow entity is a functional unit necessary for the normal operation of a workflow to perform a specific task. Workflow attributes are the parameters / configurations associated with a workflow entity.

[0042] Understandably, the types of workflow entities can include processors, triggers (e.g., schedulers), connectors (e.g., databases), etc.

[0043] For example, sentence 1, "Connect to MySQL database to fetch employee IDs and phone numbers," includes three workflow components: "MySQL," "employee IDs," and "employee phone numbers." Here, "MySQL" is the workflow entity, its type is "database," and "employee IDs" and "employee phone numbers" are workflow attributes.

[0044] For example, sentence 2, "Add filter and splitter processors to process the data and send the output to the Oracle database," includes three workflow components: "filter," "splitter," and "Oracle." Here, "filter" and "splitter" are both workflow entities, and their type is "processor." "Oracle" is also a workflow entity, but its type is "database."

[0045] 3. Workflow Mode Workflow patterns are used to indicate the structure of a task's flow and the relationships between multiple functional units of the task (i.e., workflow entities). Optionally, workflow patterns include, but are not limited to, serial, conditional, parallel, and cyclic patterns.

[0046] Serial mode refers to multiple functional units of a task executing corresponding operations in a serially connected manner. Conditional mode includes one or more conditional branches. Parallel mode includes one or more parallel processor branches. Loop mode includes one or more iterators.

[0047] Understandably, each workflow pattern can be distinguished by a pattern keyword, meaning that pattern keywords are used to differentiate workflow patterns. For example, the pattern keyword for the conditional pattern mentioned above could be "if / else", the pattern keyword for the parallel pattern could be "parallel task", and the pattern keyword for the loop pattern could be "while / this many times".

[0048] 4. Workflow Tags Workflow labels are used to indicate workflow components or workflow patterns. In other words, workflow labels can include workflow labels for indicating workflow components and workflow labels for indicating workflow patterns.

[0049] For example, the workflow label "MySQL" is a label that indicates a workflow component, specifically "MySQL" (a workflow entity of type "database"); the workflow label "conditional branch" is a label that indicates a workflow mode, specifically a conditional mode.

[0050] In the field of natural language processing, a widely used AI model is the Transformer model based on deep learning. There are three types of Transformer models: encoder, decoder, and encoder + decoder. The Transformer model incorporates a self-attention mechanism, which allows it to focus on key information, thereby saving resources and quickly acquiring the most effective information.

[0051] The encoder takes the input text, encodes the words in the input text, and converts the input text into a numerical vector (i.e., a hidden state vector) that a computer can understand (i.e., can be processed programmatically) and that contains the text's context. This process is called natural language understanding (NLU). It's important to note that natural language understanding means that the encoder's output can be understood by a computer, but it is not something humans can understand. Furthermore, the encoder passes the output hidden state vector to the decoder, which converts the hidden state vector into a human-like, comprehensible result. This process is called natural language generation (NLG).

[0052] Currently, natural language processing focuses more on enhancing the comprehensibility of its output for humans. Therefore, the ability of AI models to understand human descriptions has become a major focus in the industry. How to generate descriptions that humans can understand remains a field worthy of further research.

[0053] Furthermore, in related technologies, the decoder in the Transformer model is extremely large in order to generate descriptions that humans can understand, requiring the training of billions of parameters. This results in a significant time commitment for model training, consuming substantial computational resources and leading to a high system overhead in natural language processing.

[0054] Therefore, embodiments of the present invention provide a text processing method that can convert text describing a task's process into a workflow. Compared with related technologies that use encoders and decoders to obtain human-understandable results similar to human-like language descriptions, the present invention can obtain a workflow that is also human-understandable, and this workflow is a more concise description compared to human-like language descriptions.

[0055] The text processing method provided in this invention converts text describing a task's process into a workflow. This method enables users with only basic task-level skills to build workflows using text descriptions, thereby eliminating the need for advanced technical proficiency. Therefore, it lowers the skill level required for users to create workflows. Furthermore, this method effectively solves the problem of synchronization between technical documentation and actual production workflows. Essentially, the text processing method provided in this invention ensures synchronization between technical documentation and workflows.

[0056] For example, the text processing method provided in the embodiments of the present invention can be applied to Figure 1 The intelligent integrated system shown. (Reference) Figure 1 The intelligent integration system 100 provided in this embodiment of the invention includes an intelligent integration engine 101, a cloud server 102, and a storage system 103. A communication connection is established between any two of the intelligent integration engine 101, the cloud server 102, and the storage system 103.

[0057] The Intelligent Integration Engine 101 controls the training process of the AI ​​model (for text processing) and executes the text processing process (i.e., the model inference process) based on the AI ​​model. For example... Figure 1 As shown, the intelligent integration engine 101 includes a model fine-tuning training module, a model inference module, and a data transmission module. The model fine-tuning training module sends a fine-tuning request to the cloud server 102, instructing the cloud server 102 to begin training the AI ​​model. The model inference module processes new text (referred to as test text) based on the trained AI model to obtain processing results. In this embodiment, the processing results may be a stream generation instruction / inference ID used to generate a workflow. The data transmission module sends the test text from the model inference process and the test results obtained from the test text as new samples to a sample library and stores them. Subsequently, the new training samples can be used to update the AI ​​model, for example, to fine-tune the AI ​​model.

[0058] For example, the intelligent integration engine 101 can be a computing device such as a server or desktop computer, or it can be a module or component with processing capabilities within that computing device. For instance, the intelligent integration engine 101 is a processor (such as a central processing unit (CPU)). In this embodiment of the invention, the intelligent integration engine 101 is primarily responsible for the logical calculation and logical control functions of the workload. For complex, large-scale computing tasks, the intelligent integration engine 101 will send the computing tasks to the cloud server 102, which will then execute the computing tasks.

[0059] The cloud server 102 is used to receive computing tasks sent by the intelligent integration engine 101 and execute those tasks. For example, the cloud server 102 receives a request from the intelligent integration engine 101 to fine-tune the training AI model and executes the training process of the AI ​​model. For example, the AI ​​accelerator in the cloud server 102 executes the training process of the AI ​​model.

[0060] AI accelerators, also known as AI chips, AI wafers, computing cards, heterogeneous acceleration units, etc., are modules specifically designed for handling massive computational tasks in the field of artificial intelligence. An AI accelerator is also a type of processor.

[0061] For example, AI accelerators may include graphics processing units (GPUs), field programmable gate arrays (FPGAs), and application-specific integrated circuits (ASICs).

[0062] A GPU, also known as a visual processor or display chip, is a microprocessor specifically designed for performing image and graphics-related computations on personal computers, workstations, game consoles, and some mobile devices such as tablets and smartphones. A GPU contains thousands of cores and can process thousands of threads simultaneously, allowing for massive amounts of computation to run in parallel.

[0063] An FPGA is a programmable logic processor that can be reprogrammed by the user to meet their needs. FPGAs excel at handling short, intensive data accesses. FPGAs allow users to program circuits using tiny blocks of logic, enabling them to handle any type of digital function.

[0064] An ASIC is a special-specification integrated circuit designed for a specific purpose. It is designed and manufactured according to specific user requirements and the needs of a particular electronic system. For example, tensor processing units (TPUs) and neural network processing units (NPUs) are ASICs. TPUs and NPUs are chips specifically designed for machine learning, and with their powerful computing capabilities, they are well-suited for processing massive amounts of data.

[0065] Storage system 103 is used to store data and includes a sample library and a model library. The sample library stores training datasets for training AI models. The training datasets include a large number of training samples, each containing raw data (e.g., sentences and other information such as timestamps, inference IDs, etc.) and descriptive information. The model library stores trained AI models. The model inference module can retrieve AI models from the model library to perform text processing. Cloud server 102 can also retrieve and update AI models from the model library.

[0066] refer to Figure 1 The text processing process based on AI models can include three stages: model training, model inference, and model update.

[0067] During the model training phase, the intelligent integration engine 101 sends a fine-tuning request to the cloud server 102 to trigger the AI ​​accelerator in the cloud server 102 to start AI model training. For example, the fine-tuning request may include a preset AI model (which may be called a base AI model or an initial AI model). The AI ​​accelerator obtains a training sample set from the sample library of the storage system 103 and trains the preset AI model based on the training sample set, that is, updates the parameters of the preset AI model.

[0068] During the model inference phase, when there is a text processing requirement, the console of the text processing platform sends a text processing request to the intelligent integration engine 101 through the manager. The text processing request carries the text to be processed (i.e., the test text). Then, the model inference module of the intelligent integration engine 101 retrieves the trained AI model from the model library, processes the test text according to the AI ​​model, obtains the prediction result of the test text, and then generates a workflow based on the prediction result.

[0069] Optionally, the test text and the prediction results of the test text obtained through model inference can be stored as new training samples in the sample library of the storage system 103.

[0070] During the model update phase, which is the process of continuing to train the pre-trained AI model based on new samples, the AI ​​accelerator of cloud server 102 reads newly added training samples from the sample library and continues to train the AI ​​model based on the new training samples. This is the process of fine-tuning the AI ​​model.

[0071] It is important to note that Figure 1The illustrated intelligent integration system 100 is an example of a system architecture for which the text processing method provided in this embodiment of the invention can be applied. In this system 100, general processing functions, computing functions, and storage functions are implemented by different devices; that is, the general processing functions, computing functions, and storage functions are configured on different devices. For example, the general processing functions are implemented by the intelligent integration engine 101, the computing functions by the AI ​​accelerator 102, and the storage functions by the storage system 103. Of course, any two or three of the general processing functions, computing functions, and storage functions can also be configured on the same device; this embodiment of the invention does not limit this. It is understood that since training AI models typically requires a long time, in some embodiments, the training process of the AI ​​model can be an offline process, while the inference process of the AI ​​model can be an online process.

[0072] In some examples, the text processing method can also be applied to other systems, including general-purpose processors, AI accelerators, storage devices, etc., and the embodiments of the present invention do not specifically limit this.

[0073] In some embodiments of the present invention, the AI ​​model for text processing includes at least a clause splitting model, a clause filtering model, and a clause reordering model. The clause splitting model is used to split multiple sentences of the input text into multiple clauses. The clause filtering model is used to filter the multiple clauses of the input text, deleting clauses without workflow components. The clause reordering model is used to reorder the multiple clauses of the input text according to the execution order.

[0074] refer to Figure 1 The clause splitting model, clause filtering model, and clause reordering model can be trained offline according to the model training process described in the above embodiments. After the models are trained, text can be processed based on these models to generate workflows.

[0075] refer to Figure 2 As shown in the flowchart, in some embodiments of the present invention, after a text processing device for performing a text processing method acquires text describing a task's flow, it first decomposes the text into multiple sentences; then, according to a clause splitting model, each sentence in the multiple sentences is split into clauses, resulting in multiple clauses; the multiple clauses are then reordered and filtered according to a clause reordering model and a clause filtering model to obtain a first clause group; furthermore, a workflow mode identifier is inserted into the first clause group to obtain a second clause group; a workflow label corresponding to each clause in the second clause group is determined; then, the task's workflow is generated according to the workflow labels corresponding to the multiple clauses in the second clause group, and a visual representation of the generated workflow is generated to display the workflow.

[0076] In some embodiments of the present invention, the clause splitting model, the clause filtering model, and the clause reordering model are all encoder-based models. For ease of description, in the following embodiments, the clause splitting model is referred to as the first encoder-based model, the clause filtering model as the second encoder-based model, and the clause reordering model as the third encoder-based model.

[0077] In some embodiments of the present invention, it is used to perform Figure 2 The text processing device described in the text processing method can be a workflow generation application in a computing device. For example, the computing device can be a server, desktop computer, mobile phone, tablet computer, or personal computer (ultra-mobile personal computer (UMPC)). Alternatively, the computing device can also be other desktop devices, laptop devices, handheld devices, wearable devices, smart home devices, in-vehicle devices, and other electronic devices, such as netbooks, smartwatches, smart cameras, personal digital assistants (PDAs), portable multimedia players (PMPs), dedicated media players, or augmented reality (AR) / virtual reality (VR) devices. The embodiments of this invention do not limit the specific type and structure of the computing device.

[0078] Taking a mobile phone as an example, the workflow generation application is an application within the mobile phone. The following section, using a mobile phone as an example, describes the hardware structure of a computing device that can be used to execute the text processing provided in the embodiments of this invention.

[0079] Figure 3 This is a schematic diagram of the hardware structure of a mobile phone 300 provided in some embodiments of the present invention. For example... Figure 3 As shown, the mobile phone 300 includes a processor 310, an external memory interface 320, an internal memory 321, a universal serial bus (USB) interface 330, a charging management module 340, a power management module 341, a battery 342, an antenna 1, an antenna 2, a mobile communication module 350, a wireless communication module 360, an audio module 370, a speaker 370A, a receiver 370B, a microphone 370C, a headphone jack 370D, a sensor module 380, a keypad 390, a motor 391, an indicator device 392, one or more cameras 393, one or more displays 394, and one or more subscriber identification module (SIM) card interfaces 395.

[0080] It is understood that the structure shown in the embodiments of the present invention does not constitute a specific limitation on the mobile phone 300. In other embodiments of the present invention, the mobile phone 300 may include more or fewer components than those shown; or, some components may be combined, some components may be separated, or different components may be provided. The components shown may be implemented in hardware, software, or a combination of software and hardware.

[0081] Processor 310 may include one or more processing units, such as a central processing unit (CPU), application processor (AP), modem processor, image signal processor (ISP), controller, video codec, digital signal processor (DSP), baseband processor, graphics processing unit (GPU), neural network processing unit (NPU), or tensor processing unit (TPU). Different processors may be independent devices or integrated into one or more devices.

[0082] The processor 310 may include a controller, which serves as the nerve center and command center of the mobile phone 300. The controller can generate operation control signals based on instruction opcodes and timing signals, thereby completing the control of instruction extraction and execution.

[0083] The processor 310 may also include a memory for storing instructions and data. In some embodiments, the memory in the processor 310 is a cache memory. This memory can store instructions or data that the processor 310 has just used or retrieved. If the processor 310 needs to use these instructions or data again, it can retrieve them directly from this memory. This avoids repeated access, reduces the waiting time of the processor 310, and improves the efficiency of the system.

[0084] The external storage interface 320 can be used to connect an external storage card, such as a Micro SD card, to expand the storage capacity of the mobile phone 300. The external storage card communicates with the processor 310 through the external storage interface 320 to perform data storage functions, such as saving music and video files on the external storage card.

[0085] Internal memory 321 is used to store computer-executable program code, including instructions. Processor 310 executes the instructions stored in internal memory 321 to perform various functional applications and data processing of mobile phone 300. Internal memory 321 may include a program storage partition and a data storage partition. The program storage partition may store the operating system, applications required for at least one function (such as sound playback function, image playback function), etc. The data storage partition may store data created during the use of mobile phone 300 (such as audio data, phonebook, etc.). Furthermore, internal memory 321 may include high-speed random access memory, or non-volatile memory, such as at least one disk storage device, flash memory device, or general-purpose flash memory.

[0086] The charging management module 340 receives charging input from the charger. The charger can be connected to the mobile phone 300 via USB interface 330. While the charging management module 340 is charging the battery 342, it can also simultaneously supply power to the electronic device via the power management module 341.

[0087] The power management module 341 is used to connect the battery 342, the charging management module 340, and the processor 310. The power management module 341 receives input from the battery 342 and / or the charging management module 340 and supplies power to the processor 310, internal memory 321, external memory, one or more displays 394, one or more cameras 393, wireless communication module 360, etc.

[0088] The wireless communication function of mobile phone 300 can be implemented through antenna 1, antenna 2, mobile communication module 350, wireless communication module 360, modem processor, and baseband processor. Antenna 1 and antenna 2 are used to transmit and receive electromagnetic wave signals.

[0089] The mobile communication module 350 can provide wireless communication solutions, including 2G / 3G / 4G / 5G, for use on the mobile phone 300.

[0090] The wireless communication module 360 ​​can provide wireless communication solutions for mobile phone 300, including wireless local area network (WLAN), such as Wi-Fi, Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC) and infrared (IR).

[0091] In some embodiments, antenna 1 of mobile phone 300 is coupled to mobile communication module 350, and antenna 2 is coupled to wireless communication module 360, enabling mobile phone 300 to communicate with networks and other devices through wireless communication technology.

[0092] The mobile phone 300 implements display functions through a GPU, one or more displays 394, and an application processor. The GPU is an image processing microprocessor connected to one or more displays 394 and the application processor. The GPU is used to perform mathematical and geometric calculations for graphics rendering, model training, etc. The processor 310 may include one or more GPUs that execute program instructions to generate or modify display information.

[0093] The display screen 394 is used to display images, videos, etc. The display device includes a display panel. In some embodiments, the mobile phone 300 may include one or N displays 394, where N is a positive integer greater than 1.

[0094] The mobile phone 300 can perform shooting functions through an ISP, one or more cameras 393, a video codec, a GPU, one or more displays 394, and an application processor. The mobile phone 300 can perform audio functions such as music playback and recording through an audio module 370, a speaker 370A, a receiver 370B, a microphone 370C, a headphone jack 370D, and an application processor.

[0095] The audio module 370 is used to convert digital audio information into analog audio signal output, and also to convert analog audio input into digital audio signal. The audio module 370 can also be used to encode and decode audio signals. In some embodiments, the audio module 370 may be located in the processor 310, or some functional modules of the audio module 370 may be located in the processor 310.

[0096] It is understood that in the embodiments of the present invention, the mobile phone 300 may execute some or all of the steps of the embodiments of the present invention. These steps or operations are merely examples, and the mobile phone 300 may also execute other operations or variations thereof. Furthermore, the execution order of the various steps may differ from the order presented in the embodiments of the present invention, and it is possible that not all operations of the embodiments of the present invention are executed. The embodiments of the present invention may be implemented independently or in any combination, and the embodiments of the present invention do not limit this.

[0097] The text processing method provided in this embodiment of the invention can be applied to text processing applications with... Figure 3 The hardware structure shown is applicable to computing devices or computing devices with similar structures. Alternatively, it can also be applied to computing devices with other structures, and the embodiments of the present invention are not limited thereto.

[0098] Based on the above description, the text processing method provided in this embodiment of the invention can be executed by a computing device. For example, it can be executed by a workflow generation application running on a computing device. Figure 4 As shown, the text processing method provided in this embodiment of the invention includes steps S401-S404.

[0099] In S401, text describing the process of the task is received.

[0100] In this embodiment of the invention, after a workflow generation application in a computing device receives text describing the flow of a task, the text is used as input to the workflow generation application, and then the text is processed in the workflow generation application. For ease of description, in the following embodiments, the text describing the flow of a task is uniformly referred to as "input text".

[0101] In some examples, the text used to describe the process of a task may be text stored locally on the computing device or text received by the computing device from other devices; this embodiment of the invention does not limit this.

[0102] In S402, the first clause group is obtained from the input text. The first clause group includes multiple clauses arranged in the order in which the process occurs, and each clause in the first clause group includes a workflow component.

[0103] Workflow components are used to indicate information about a corresponding functional unit among multiple functional units used to perform a task. Workflow components may include workflow entities or workflow attributes. A more detailed description of workflow components can be found in the description of the definition of workflow components in the above embodiments.

[0104] In some embodiments of the present invention, the first clause group is a group containing multiple clauses obtained after processing the input text, and each clause in the first clause group includes a workflow component. This simplifies the structure of each clause, facilitating a one-to-one mapping between clauses and workflow labels in the next stage based on the workflow components. Furthermore, the multiple clauses in the first clause group are arranged in the order in which the process occurs, which helps in generating the correct workflow subsequently.

[0105] For example, the following text (Text 1) used to describe the process of a task is used as input text.

[0106] Text 1: "After the scheduler is triggered, connect to the MySQL database to query data. Add a condition processor; if condition A is met, add a parallel processor. In parallel task 1, add a filter processor and a splitter processor to process the data and send the output to the Oracle database. In parallel task 2, when sorting is complete, send the data to DB2. Then merge the parallel processing branches. If condition B is met, convert the data to the desired format and send the output to PostgreSQL, then merge the results of the router branch. Finally, send a notification via the SMN service." The clause group obtained after processing the above input text in step S402 (which can be used as the second clause group mentioned below) is as follows: "After triggering the scheduler, connect to the MySQL database. Add a condition processor. If condition A is met, add a parallel processor. In parallel task 1, add a splitter processor. Add a filter processor. Send the output to the Oracle database. In parallel task 2, when sorting is complete, send the data to DB2. Then merge the parallel processing branches. If condition B is met, convert the data to the desired format. Send the output to PostgreSQL. Merge the results of the router branches. Finally, send a notification via the SMN service." In the example above, each clause in the clause group contains only one workflow component, and the clauses are arranged in the order in which the process occurs.

[0107] Table 1 below shows the workflow components in each clause of the above clause group and the type of the workflow components.

[0108] Table 1

[0109] In some embodiments, reference Figure 5A The above step S402 includes steps S4021 to S4023.

[0110] In S4021, the input text is decomposed to obtain the third clause group.

[0111] Each clause in the third clause group contains at most one workflow component.

[0112] In some embodiments, continue to refer to Figure 5A and Figure 5B S4021 can be implemented through S4021a and S4021b.

[0113] In S4021a, sentence boundary detection is performed on the input text to obtain multiple sentences of the input text.

[0114] In some embodiments of the present invention, sentence boundary detection (SBD) technology can be used. Optionally, rule-based, unsupervised learning, or supervised learning methods can be used to segment the input text into multiple sentences.

[0115] In S4021b, a third clause group is obtained by splitting each sentence in the text into one or more clauses through a first encoder-based model.

[0116] It is understandable that the first encoder-based model (i.e., the clause splitting model) is a Transformer model, and the first encoder-based model is a lexical classification model. The input of the first encoder-based model is a sentence. Each word or sub-word in the sentence can be considered a lexical. The output of the first encoder-based model is the start / inside / outside / end (BIOE) label for each lexical (which can be a word or a sub-word) in the input sentence. It can be seen that the first encoder-based model is a multi-classification model. In this embodiment of the invention, a sentence can be split into one or more clauses based on the output BIOE labels.

[0117] refer to Figure 6 The first encoder-based model comprises multiple encoders connected sequentially and multiple heads. Each encoder has multiple inputs and multiple outputs, with each output corresponding to a head. The head includes a feed-forward neural network (FFNN), a softmax function, and an argmax function. Each input to the encoder is a word from the sentence to be split, and each output outputs a hidden state vector corresponding to that word. The hidden state vectors are input to the FFNN and normalized using the softmax function to obtain the probability values ​​of the word's BIOE labels. Specifically, the probability values ​​of words belonging to the four labels B, I, O, and E are obtained. Then, after processing by the argmax function, the label corresponding to the highest probability value is determined as the BIOE label for that word. Next, based on the BIOE label of each word in the input sentence, the beginning of a clause is marked with a B label, and the end of a clause is marked with an E label, thus splitting the input sentence into one or more clauses.

[0118] For example, the encoder in the first encoder-based model can be any one of BERT, ALBERT, DistilBERT, ELECTAR, RoBERTa, etc. This embodiment of the invention does not limit the first encoder-based model.

[0119] For example, when the input sentence is "The tickets were sold out so we watched a movie on TV," each word in the input sentence is used as a word unit. Figure 6 The BIOE labels corresponding to each word in the input sentence after processing by the first encoder-based model are shown in Table 2 below.

[0120] Table 2

[0121] Combining the output results of the first encoder-based model shown in Table 2, the output results are aggregated to obtain two clauses: "The tickets were sold out" and "we watched a movie on TV".

[0122] In some examples, the first encoder-based model is trained on an annotated training dataset that includes multiple original sentences and the correct annotated BIOE label for each word in the original sentences.

[0123] In S4022, multiple clauses in the third clause group are filtered to obtain the fourth clause group. Each clause in the fourth clause group includes a workflow component.

[0124] In some embodiments of the present invention, clauses that do not include workflow components are invalid information during workflow generation. Therefore, clauses that do not include workflow components (i.e., clauses that include zero workflow components) are deleted from the third clause group. Thus, each of the remaining clauses includes a workflow component, which helps to generate a more concise and accurate workflow.

[0125] In some embodiments, continue to refer to Figure 5A and Figure 5B The above S4022 can be implemented through S4022a and S4022b.

[0126] In S4022a, a second encoder-based model is used to determine whether each clause in the third clause group includes a workflow component.

[0127] In S4022b, if it is determined that the clauses in the third clause group include zero workflow components, then the clauses in the third clause group that include zero workflow components are deleted, and the clauses in the third clause group that include workflow components are retained, thus obtaining the fourth clause group.

[0128] It is understandable that the second encoder-based model (i.e., the clause filtering model) is a Transformer model, and thus a binary classification model. The input to the second encoder-based model is a clause. Each word or sub-word in the clause can be considered a lexical unit. The output of the second encoder-based model is indication information indicating whether the input clause contains workflow components. Therefore, it can be seen that the second encoder-based model is a binary classification model.

[0129] refer to Figure 7 The second encoder-based model comprises multiple encoders connected sequentially and a head. All encoders except the last one include multiple inputs and multiple outputs. The last encoder has one output connected to the head. The head includes a feedforward neural network (FFNN), a softmax function, and an argmax function. For each input of the first encoder, each word in the clause is input, and each output of the encoder outputs a hidden state vector corresponding to that word. The output of the last encoder is the hidden state vector corresponding to the input clause. The hidden state vector is then input into the FFNN and normalized by the softmax function to obtain a probability value (Y or N) indicating whether the input clause includes a workflow component. The indication Y indicates that the clause includes a workflow component, and N indicates that the clause includes zero workflow components. Next, after processing by the argmax function, the indication corresponding to the maximum probability value is determined as the indication corresponding to the clause. Then, based on the indication, it is determined whether the input clause includes a workflow component or includes zero workflow components.

[0130] For example, when the input clause is "connect to MySQL database", an indicator "CLS" to indicate the start of the sentence and an indicator "Sep" to separate the sentences are added at the beginning and end of the clause, respectively. Each word in the input sentence is input as a lexical unit into a second encoder-based model. If the second encoder-based model outputs an indicator "Y", it can be determined that the clause includes workflow components, and the clause is retained.

[0131] Optionally, the second encoder-based model is trained on an annotated training dataset that includes multiple sentences and annotated labels indicating whether a clause includes a workflow component.

[0132] In S4023, the clauses in the fourth clause group are rearranged to obtain the first clause group.

[0133] In some embodiments of the present invention, by reordering multiple clauses in the fourth clause group, the multiple clauses in the fourth clause group can be arranged in the order in which the task occurs, making it possible to generate an accurate workflow based on the reordered clauses in subsequent steps.

[0134] In some embodiments, continue to refer to Figure 5A S4023 can be implemented through S4023a.

[0135] In S4023a, the first clause group is obtained by reordering multiple clauses in the fourth clause group through the third encoder-based model.

[0136] The input to the third encoder-based model is two clauses from the fourth clause group, and the output of the third encoder-based model is information indicating whether the order of these two clauses is correct.

[0137] It's understandable that the third encoder-based model (i.e., the clause reordering model) is a Transformer model, and therefore a binary classification model. The input to the third encoder-based model is two clauses. Each word or sub-word in a clause can be considered a lexical unit. The output of the third encoder-based model is an indication of whether the two input clauses are in the correct order. It can be seen that the third encoder-based model is also a binary classification model.

[0138] refer to Figure 8 The third encoder-based model comprises multiple encoders connected sequentially and a head. All encoders except the last one include multiple inputs and multiple outputs. The last encoder has one output connected to the head. The head includes a feedforward neural network (FFNN), a softmax function, and an argmax function. Each word in the two clauses is input to each input of the first encoder, and each output of the encoder outputs a hidden state vector corresponding to that word. The output of the last encoder is a hidden state vector representing the category corresponding to the two input clauses. The hidden state vectors are then input into the FFNN and normalized using the softmax function to obtain probability values ​​(Y or N) indicating whether the order of the two clauses is correct. The indication for correct clause order is Y, and the indication for incorrect clause order is N. Finally, after processing by the argmax function, the indication corresponding to the highest probability value is determined as the indication for the clause. Next, based on the instruction information, determine whether the order of the two input clauses is correct, that is, whether the two clauses need to be rearranged.

[0139] For example, if the two input clauses are "connect to the MySQL database" and "after triggering the scheduler", the original order of these two clauses is "connect to the MySQL database" before "after triggering the scheduler". Adding "CLS" and "Sep" after these clauses, and then feeding each word in the clauses as a token into a third encoder-based model, the output of the third encoder-based model is "N". It can be determined that the order of these two clauses is incorrect and needs to be reversed, placing "after triggering the scheduler" before "connect to the MySQL database".

[0140] It is important to note that the fourth clause group comprises multiple clauses. When the fourth clause group contains at least three clauses, these clauses can be rearranged using bubble sorting based on the third encoder-based model.

[0141] Optionally, the third encoder-based model is trained on an annotated training dataset that includes multiple pairs of clauses and annotated labels indicating whether the order of each pair of clauses is correct.

[0142] Based on the descriptions in S4021 and S4023 above, the first encoder-based model, the second encoder-based model, and the third encoder-based model can all be pre-trained models. Most parameters in the pre-trained models have been well optimized. In this embodiment of the invention, the first encoder-based model, the second encoder-based model, and the third encoder-based model can be fine-tuned according to small to medium-sized datasets, so that each model adapts to the current dataset.

[0143] In some embodiments of the present invention, the models for clause splitting, clause filtering, and clause reordering are all Transformer models that include only an encoder and no decoder. Compared to methods that use models including both encoders and decoders to convert input text into human-understandable descriptions, the models used in the methods provided in these embodiments are much simpler. Therefore, computational, time, and resource costs can be reduced during model training and application.

[0144] It should be noted that the execution order of S4022 and S4023 is not limited in the embodiments of the present invention. For example, S4022 can be executed before S4023, or S4022 can be executed after S4023.

[0145] See in some examples Figure 5B The above S402 includes S4021 and S4022, and also includes S4024 and S403.

[0146] In S4024, the clauses in the fourth clause group are rearranged to obtain the second clause group.

[0147] The description of S4024 can be found in the description of S4023 above, and will not be repeated here. For example, the third encoder-based model reorders multiple clauses in the fourth clause group to obtain the second clause group.

[0148] In S403, at least one identifier of at least one workflow mode is inserted into the second clause group to obtain the first clause group.

[0149] In this embodiment of the invention, the input text is text used to describe the flow of a task. Typically, a flow has a certain structure, and there are certain relationships between the multiple functional units (i.e., workflow entities) that execute the task. In this embodiment, the information indicating the structure of the task flow and the relationships between the multiple functional units (i.e., workflow entities) of the task is defined as a workflow pattern. Workflow patterns include, but are not limited to, serial patterns, conditional patterns, parallel patterns, and cyclic patterns. A more detailed description of workflow patterns can be found in the above embodiments' definition of workflow patterns, and will not be repeated here.

[0150] Understandably, for at least one workflow pattern in the second clause group, each workflow pattern corresponds to an identifier that includes a workflow pattern boundary indication sentence and / or a workflow pattern indication. A workflow pattern boundary indication sentence is a sentence (or clause) associated with the start or end of a workflow pattern. That is, a workflow pattern boundary indication sentence can be a sentence (or clause) that can indicate the start or end of a workflow pattern. A workflow pattern indication is used to indicate the type of workflow pattern and the boundaries of the workflow pattern.

[0151] For example, for a conditional pattern, the workflow pattern boundary indicator sentence indicating the start of the conditional pattern could be "Add a conditional processor," and the workflow pattern boundary indicator sentence indicating the end of the conditional pattern could be "Merge the results of the router branches." The workflow pattern indicator for a conditional pattern could be "Create a conditional branch."

[0152] In one possible implementation, the above S403 is implemented by the following S1 and S2.

[0153] In S1, at least one workflow mode of the second clause group is determined.

[0154] For the input text to include a description of at least one workflow pattern, it can also be understood that the process described by multiple clauses in the second clause group includes at least one workflow pattern. This at least one workflow pattern can be referred to as at least one workflow pattern of the input text or at least one workflow pattern of the second clause group. For example, the workflow patterns of the text 1 above include a conditional pattern and a parallel pattern.

[0155] In embodiments of the invention, at least one workflow pattern of the input text can be determined based on multiple clauses in the second clause group. In some embodiments, at least one workflow pattern can be determined by pattern keyword extraction, i.e., extracting at least one pattern keyword from multiple clauses. A pattern keyword is used to indicate a workflow pattern. For example, the pattern keyword for a conditional pattern can be "if", the pattern keyword for a parallel pattern can be "parallel tasks", and the pattern keyword for a loop pattern can be "when".

[0156] Optionally, machine learning methods can be used to determine at least one workflow pattern of the input text. For example, an AI model can be trained to determine workflow patterns. Clauses in the second clause group can be used as input to the AI ​​model. The AI ​​model can analyze the input clauses, extract one or more pattern keywords, and output at least one workflow pattern of the second clause group (i.e., at least one workflow pattern of the input text).

[0157] For example, for the second clause group in the example above, the extracted pattern keywords include "if" and "parallel". For example, "if" is extracted from the clause "If condition A is true, then add a parallel processor", and "parallel task" is extracted from the clauses "In parallel task 1, add a splitter processor" and "In parallel task 2, when reordering is complete". The workflow pattern corresponding to the pattern keyword "if" is the conditional pattern, and the workflow pattern corresponding to the pattern keyword "parallel task" is the parallel pattern; therefore, at least one workflow pattern for determining the input text includes the conditional pattern and the parallel pattern.

[0158] In S2, for each workflow pattern in at least one workflow pattern, at least one of the following processing methods 1 and 2 is performed on the second clause group to obtain the first clause group.

[0159] In processing method 1, if there is no clause in the second clause group that indicates the boundary of the workflow pattern, then the workflow pattern boundary indication clause is inserted into the second clause group.

[0160] It is understood that workflow pattern boundary indication sentences can be clauses indicating the start or end of a workflow pattern. In embodiments of the present invention, pattern keywords are determined based on one or more clauses. For example, the conditional pattern in the above example is determined based on the clause "If condition A is true, then add a parallel processor". For ease of description, in the following embodiments, the clauses used to determine pattern keywords are referred to as pattern clauses.

[0161] Optionally, after determining at least one workflow pattern as described above, a search can be performed based on pattern keywords to determine whether there are clauses in the second clause group that indicate the boundaries of each workflow pattern. For example, one or more clauses preceding the pattern clause can be searched to determine whether there are clauses related to the start of the workflow pattern, and one or more clauses following the pattern clause can be searched to determine whether there are clauses related to the end of the workflow pattern.

[0162] For example, for the second clause group in the above example, the workflow modes determined in the above embodiment include conditional mode and parallel mode. For the conditional mode, its pattern keyword is "if," and the sentences containing the pattern keyword (i.e., pattern clauses) are "If condition A, then add a parallel processor" and "If condition B, then convert the data to the desired format." Before these two clauses, a clause related to the start of the conditional mode, "add a condition processor," is found through a search; after these two clauses, a clause related to the end of the conditional mode, "merge the results of the router branches," is found through a search. That is, the second clause group includes workflow mode boundary indication sentences for the conditional mode. For the parallel mode, the pattern keyword is "parallel task," and the sentences containing the pattern keyword are "In parallel task 1, add a splitter processor" and "In parallel task 2, when sorting is complete." Before these two clauses, a clause related to the start of the parallel mode, "add a parallel processor," is found through a search; after these two clauses, a clause related to the end of the parallel mode, "then merge the parallel processing branches," is found through a search. In other words, the second clause group includes workflow mode boundary indicator sentences for parallel mode.

[0163] For any workflow pattern in at least one workflow pattern, if there is a clause in the second clause group that indicates the boundary of the workflow pattern, then the second clause group is not processed; if there is no clause in the second clause group that indicates the boundary of the workflow pattern, then the clause that indicates the boundary of the workflow pattern, i.e., the workflow pattern boundary indication sentence, is inserted into the second clause group.

[0164] During the insertion of workflow pattern boundary indicator sentences into the second clause group, if the second clause group does not contain a clause indicating the start of the workflow pattern, a workflow pattern boundary indicator sentence indicating the start of the workflow pattern is inserted somewhere before the pattern clause; conversely, if the second clause group does not contain a clause indicating the end of the workflow pattern, a workflow pattern boundary indicator sentence indicating the end of the workflow pattern is inserted somewhere after the pattern clause. The insertion position of the workflow pattern boundary indicator sentence is related to the context of the second clause group.

[0165] For input text 1 in the example above, processing input text 1 yields the second clause group. When processing using method 1, since the second clause group contains clauses indicating the boundaries of the workflow pattern, there is no need to insert a workflow pattern boundary indicator sentence. Therefore, the result of processing the second clause group according to method 1 is still the second clause group, which is used as the first clause group.

[0166] In processing method 2, if there is no clause in the second clause group that serves as a workflow mode indicator, then the workflow mode indicator of the workflow mode is inserted into the second clause group.

[0167] In some embodiments of the present invention, a workflow mode indicator is a clause used to indicate the type and boundaries of a workflow mode. For example, for a conditional mode, the workflow mode indicator may be "create conditional branch"; for a parallel mode, the workflow mode indicator may be "create parallel task branch".

[0168] Based on the above description of the pattern clause in processing method 1, in this embodiment of the invention, the method for determining whether there is a clause in the second clause group that can be used as a workflow pattern indicator may be to determine whether the pattern clause itself is a workflow pattern indicator. If the pattern clause is a workflow pattern indicator, then there is no need to process the second clause group. If the pattern clause is not a workflow indicator, then a corresponding workflow pattern indicator needs to be inserted before the pattern clause.

[0169] Taking the second clause group of the above example as an example, as can be seen from the description of the above embodiment, the pattern clauses in the second clause group are as follows: "If it is condition A, then add a parallel processor."

[0170] If condition B is met, the data will be converted to the desired format.

[0171] In parallel task 1, add a splitter processor.

[0172] In parallel task 2, when the reordering is complete. None of the four pattern clauses mentioned above are workflow pattern indicators. Therefore, the corresponding workflow pattern indicator needs to be inserted before these pattern clauses.

[0173] For example, after inserting the workflow pattern instructions ("Create Conditional Branch" and "Create Parallel Task Branch") into the second clause group, the resulting first clause group is: "After triggering the scheduler, connect to the MySQL database. Add a condition processor. Create a condition branch. If condition A is met, add a parallel processor. Create a parallel task branch. In parallel task 1, add a splitter processor. Add a filter processor. Send the output to the Oracle database. Create a parallel task branch. In parallel task 2, when sorting is complete, send the data to DB2. Then merge the parallel processing branches. Create a condition branch. If condition B is met, convert the data to the desired format. Send the output to PostgreSQL. Merge the results of the router branch. Finally, send a notification via the SMN service." In some embodiments of the present invention, the first clause group obtained by S403 includes clauses arranged in the order in which the task flow occurs, workflow mode boundary indication sentences, and workflow mode indications. The first clause group contains elements for generating a workflow.

[0174] In S404, determine the workflow label corresponding to each clause in the first clause group.

[0175] Workflow labels corresponding to the multiple clauses in the first clause group include a first workflow label indicating a workflow component or a second workflow label indicating a workflow pattern. It is understood that workflow labels are simplified representations of workflow components and workflow patterns, and are used to generate workflows.

[0176] In some embodiments of the present invention, the workflow tag matching method can be used to match a corresponding workflow tag for each clause in the first clause group. For example, refer to Figure 4 and Figure 9 The above S404 can be implemented through S4041.

[0177] In S4041, each clause in the first clause group is matched with multiple workflow tags included in the tag dictionary to determine the workflow tag corresponding to each clause in the first clause group from the tag dictionary.

[0178] In some embodiments, a tag dictionary for workflow tag matching can be pre-created. The tag dictionary includes multiple workflow tags and predefined values ​​corresponding to these workflow tags. The predefined values ​​are based on workflow components or workflow patterns. That is, in the tag dictionary, workflow tags are "keys," and predefined values ​​are "values." The tag dictionary is associated with the input text to be processed in these embodiments of the invention, and includes all workflow components and workflow patterns of the input text.

[0179] In some examples, the above S4041 includes S4041a and S4041b.

[0180] In S4041a, semantic similarity processing is performed to match each clause in the first clause group with a predefined value in the label dictionary.

[0181] In some embodiments of the present invention, semantic similarity detection technology is used to perform similarity matching between clauses and predefined values ​​in the tag dictionary, resulting in multiple similarity scores. Optionally, the semantic similarity detection method can be a similarity calculation method based on cosine similarity or Euclidean distance, etc., and the choice can be made according to actual needs. The embodiments of the present invention do not limit this.

[0182] In S4041b, the workflow label corresponding to the predefined value with the highest similarity score for the clause is determined as the workflow label corresponding to the clause.

[0183] In other words, the workflow label corresponding to the predefined value with the highest similarity to the clause is the workflow label that matches the clause.

[0184] Summarizing S4041a and S4041b, it can be seen that this embodiment of the invention uses an indirect semantic similarity detection method to annotate sentences. That is, instead of directly matching clauses with workflow tags (which is originally a multi-level classification task), the clauses are first matched with predefined values, and then the workflow tag corresponding to the predefined value that matches the clause is used as the workflow tag corresponding to the clause. Since there are many types and a large number of workflow tags, directly matching clauses with workflow tags is computationally intensive, slow, and has low accuracy. Therefore, this embodiment of the invention uses a faster and more accurate indirect method to match clauses with workflow tags.

[0185] For example, for the first clause group in the above example, the workflow labels matched by each clause in the first clause group are shown in Table 3 below.

[0186] Table 3

[0187] In S405, a workflow for the task is generated based on the workflow labels corresponding to the multiple clauses in the first clause group.

[0188] In some embodiments of the present invention, the generated workflow includes workflow labels corresponding to multiple clauses in the first clause group and tag numbers corresponding to the workflow labels, which are used to indicate the logical relationship between the workflow labels.

[0189] For example, based on the workflow tags corresponding to multiple clauses in the first clause group in the example shown in Table 3 above, the content of the generated workflow is shown in Table 4 below.

[0190] Table 4

[0191] Referring to Table 4, in the workflow label markers, numbers indicate the execution order, the symbol "." indicates a branch, and the symbol "-" indicates a serial connection.

[0192] In S406, a visual representation of the generated workflow is generated for display by the workflow generation application.

[0193] In some embodiments of the present invention, visualization tools can be used to visually represent the workflow obtained in S405 above, enabling the workflow to be displayed and allowing users to intuitively perceive and understand it. Optionally, the visual representation of the workflow can take the form of a workflow flowchart or other forms, and the embodiments of the present invention do not limit this. Figure 10 This is an example of a visual representation of the workflow shown in Table 4 above.

[0194] Based on the above description, in the text processing method provided in this embodiment of the invention, a workflow generation application in a computing device receives text describing the flow of a task and uses this text as input to the workflow generation application. The text is then processed to obtain a first clause group. The first clause group includes multiple clauses arranged in the order in which the flow described in the text occurs, and each clause in the first clause group includes a workflow component. Furthermore, a workflow label corresponding to each clause in the first clause group, which can indicate a workflow component or workflow pattern, is determined. Then, based on the workflow labels corresponding to the multiple clauses in the first clause group, a workflow for the task is generated; a visual representation of the generated workflow is generated, enabling the workflow generation application on the computing device to display the workflow. The text processing method provided in this embodiment of the invention can convert text describing the flow of a task into a workflow. Compared with technical solutions that use models including encoders and decoders to obtain human-understandable results similar to human-like language descriptions, the technical solution of this invention can obtain a workflow that is also human-understandable, and compared to human-like language descriptions, this workflow is a more concise description. In other words, the text processing method provided in this embodiment of the invention can convert text describing the process of a task into a concise workflow that humans can understand.

[0195] Some embodiments of the present invention also provide a computer device for implementing the text processing method described in the above embodiments. Figure 11 These are structural diagrams of computing devices provided in some embodiments of the present invention. In some examples, Figure 11 The computing device shown is Figure 1 The intelligent integration engine 101 in the system architecture shown is also used to implement... Figure 4 , Figure 5A , Figure 5B and Figure 9 The computing device described in the method embodiment is shown in the figure. The computing device includes at least one processor 1101 and a memory 1102.

[0196] The processor 1101 includes one or more central processing units (CPUs). The CPU is a single-core CPU (single CPU) or a multi-core CPU (multi-CPU).

[0197] The memory 1102 includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, or optical memory. The operating system code is stored in the memory 1102.

[0198] In some examples, processor 1101 implements the methods in the above embodiments by reading instructions stored in memory 1102; or, processor 1101 implements the methods in the above embodiments by internally stored instructions. When processor 1101 implements the methods in the above embodiments by reading instructions stored in memory 1102, memory 1102 stores instructions for implementing the text processing method provided in the embodiments of the present invention.

[0199] After the program code stored in memory 1102 is read by at least one processor 1101, the computing device performs the following operations: receiving text describing the flow of a task for which a workflow is to be generated, as input to a workflow generation application executed on the computing device; obtaining a first clause group from the text describing the flow of the task, the first clause group comprising a plurality of clauses arranged in the order in which the flow occurs, each of the plurality of clauses in the first clause group comprising a workflow component, the workflow component being used to indicate information of a corresponding functional unit among a plurality of functional units for performing the task; determining a workflow label corresponding to each clause in the first clause group, the workflow label corresponding to the plurality of clauses in the first clause group including a first workflow label for indicating a workflow component or a second workflow label for indicating a workflow mode; generating a workflow for the task based on the workflow labels corresponding to the plurality of clauses in the first clause group; and generating a visual representation of the generated workflow for display by the workflow generation application.

[0200] In some embodiments, Figure 11 The computing device shown also includes a communication interface 1103. The communication interface 1103 is a wired interface, such as a fiber distributed data interface (FDDI) or a gigabit Ethernet (GE) interface. Alternatively, the communication interface 1103 is a wireless interface. The communication interface 1103 is used to receive text describing the flow of a task. Alternatively, the communication interface 1103 is used to send the generated workflow to a display device or other device.

[0201] The memory 1102 is used to store text received by the communication interface 1103. Optionally, the memory 1102 is also used for training datasets and related AI models. At least one processor 1101 also executes the methods described in the foregoing method embodiments based on the contents stored in the memory 1102. How the processor 1101 implements the functions described above can be referred to the description in the foregoing method embodiments, and will not be repeated here.

[0202] In some examples, the computing device also includes a bus 1104. The processor 1101, memory 1102, and communication interface 1103 are typically interconnected via the bus 1104, or otherwise.

[0203] In some examples, the computing device also includes an input / output interface 1105 for connecting to an input device to receive text input by a user. Input devices include, but are not limited to, a keyboard, a touchscreen, and a microphone. The input / output interface 1105 is also used to connect to an output device to output and display the workflow generated by the processor 1101. Output devices include, but are not limited to, a monitor and a printer.

[0204] The computing device provided in this embodiment of the invention is used to execute one or more steps of the text processing method provided in any of the above embodiments, thereby converting text into a concise workflow that humans can understand. It should be noted that the beneficial effects of this computing device are the same as those of the text processing methods described in some of the above embodiments, and will not be repeated here.

[0205] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any combination thereof. When the above embodiments are implemented using software programs, the software programs can be implemented, in whole or in part, in the form of a computer program product. A computer program product includes one or more computer instructions. When loaded and executed on a computer, the computer instructions generate some of the processes or functions provided in the embodiments of the present invention. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or any other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave) methods. A computer-readable storage medium can be any available medium accessible to a computer, or a server, data center, or any other data storage device that includes one or more available media. Available media can be magnetic media (e.g., floppy disks, magnetic disks, or magnetic tapes), optical media (e.g., digital versatile disks (DVDs)), semiconductor media (e.g., solid-state drives (SSDs)), etc.

[0206] Through the description of the above embodiments, those skilled in the art will clearly recognize that, for the sake of convenience and brevity, the illustrative examples are based solely on the functional module division described above. In practical applications, the above functions can be assigned to different functional modules as needed. That is, the internal structure of the device can be divided into different functional modules to perform all or part of the functions described above. The specific working processes of the above systems, devices, and units can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0207] In the several embodiments provided by this invention, it is understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the above device embodiments are merely exemplary. For example, the division of functional modules or units is only a logical functional division. In actual implementation, there may be other division methods. For example, in some embodiments, multiple devices or components may be combined or integrated into another system, or some features may be ignored or not performed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be an electrical connection, a mechanical connection, or other forms of connection.

[0208] Units described as individual components may or may not be physically separate, and components shown as units may or may not be physical units. That is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiment scheme according to actual needs.

[0209] The functional units in the embodiments of the present invention can be integrated into a single processing unit; or these units can be separate physical units; or two or more units can be integrated into a single unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0210] 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 readable storage medium. Based on this understanding, the technical solution of the present invention can essentially be all or part of a technical solution, and can be embodied in the form of a software product. The 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.) or processor to execute all or part of the steps of the methods described in various embodiments of the present invention. The storage medium includes various types of media capable of storing program code, such as flash memory (USB flash drive), portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, etc.

[0211] Some embodiments of the present invention provide a computer-readable storage medium (e.g., a non-transitory computer-readable storage medium). The computer-readable storage medium stores computer program instructions that, when executed on a computer, cause the computer to perform one or more steps of the text processing method as described in any of the above embodiments.

[0212] For example, computer-readable storage media include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, or magnetic tapes), optical disks (e.g., compact disks, CDs, or DVDs), smart cards, and flash memory devices (e.g., erasable programmable read-only memory (EPROMs), cards, memory sticks, or key drives). The various computer-readable storage media described in embodiments of the present invention can refer to one or more devices and / or other machine-readable storage media for storing information. The term "computer-readable storage medium" can include, but is not limited to, wireless channels and various other media capable of storing, including, and / or carrying instructions and / or data.

[0213] Some embodiments of the present invention also provide a computer program product. The computer program product includes program instructions carried on a non-transitory computer-readable storage medium. When executed on a computer, the computer program instructions cause the computer to perform one or more steps of the text processing method described in the above embodiments.

[0214] The beneficial effects of computer-readable storage media and computer program products are the same as those of the text processing methods described in the above embodiments, and will not be repeated here.

[0215] The foregoing description is merely a specific implementation of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions falling within the technical scope of the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A text processing method executed by a computing device, characterized in that, include: Receive text describing the flow of a task for which a workflow is to be generated, as input to a workflow generation application executed on the computing device; Obtain a first clause group from the text describing the process of the task, wherein the first clause group includes a plurality of clauses arranged in the order in which the process occurs, and each of the plurality of clauses in the first clause group includes a workflow component for indicating information of a corresponding functional unit among a plurality of functional units for performing the task; Determine a workflow label corresponding to each clause in the first clause group, wherein the workflow label corresponding to the plurality of clauses in the first clause group includes a first workflow label for indicating a workflow component or a second workflow label for indicating a workflow mode; The workflow for the task is generated based on the workflow tags corresponding to the plurality of clauses in the first clause group; Generate a visual representation of the generated workflow for display by the workflow generation application.

2. The text processing method according to claim 1, characterized in that, Obtaining the first clause group from the text used to describe the process of the task includes: The text describing the process of the task is decomposed to obtain a third clause group, wherein each of the multiple clauses included in the third clause group includes at most one workflow component. The clauses in the third clause group are filtered to obtain a fourth clause group, wherein each clause in the fourth clause group includes a workflow component; The clauses in the fourth clause group are rearranged to obtain the first clause group; or Obtaining the first clause group from the text used to describe the process of the task includes: The text describing the process of the task is decomposed to obtain the third clause group, wherein each of the plurality of clauses included in the third clause group includes at most one workflow component; The clauses in the third clause group are filtered to obtain the fourth clause group, wherein each clause in the fourth clause group includes a workflow component; The clauses in the fourth clause group are rearranged to obtain the second clause group; Insert at least one identifier of at least one workflow pattern into the second clause group to obtain the first clause group, wherein the at least one workflow pattern is used to indicate the structure of the process of the task and the relationship between the plurality of functional units.

3. The text processing method according to claim 2, characterized in that, Decomposing the text describing the process of the task to obtain the third clause group includes: Sentence boundary detection is performed on the text to obtain multiple sentences of the text; The third clause group is obtained by splitting each sentence in the multiple sentences of the text into one or more clauses using a first encoder-based model.

4. The text processing method according to claim 2, characterized in that, The fourth clause group is obtained by filtering the multiple clauses in the third clause group, including: The second encoder-based model is used to determine whether each clause in the plurality of clauses in the third clause group includes a workflow component; If the clause in the third clause group contains zero workflow components, then the clause in the third clause group is deleted to obtain the fourth clause group.

5. The text processing method according to claim 2, characterized in that, Reordering the clauses in the fourth clause group includes: The plurality of clauses in the fourth clause group are reordered by a third encoder-based model, wherein the input of the third encoder-based model is two clauses in the fourth clause group, and the output of the third encoder-based model is information indicating whether the order of the two clauses is correct.

6. The text processing method according to claim 2, characterized in that, The identifier of each workflow mode in the at least one workflow mode includes a workflow mode boundary indication sentence and / or a workflow mode indication; inserting the at least one identifier of the at least one workflow mode into the second clause group yields the first clause group, which includes: Determine at least one workflow pattern of the second clause group; For each of the at least one workflow pattern, the first clause group is obtained by performing at least one of the following on the second clause group: If there is no clause in the second clause group indicating the boundary of the workflow pattern, then the workflow pattern boundary indication clause is inserted into the second clause group; or If there is no clause in the second clause group that serves as an indication of the workflow mode, then the workflow mode indication of the workflow mode is inserted into the second clause group.

7. The text processing method according to claim 6, characterized in that, Determining the at least one workflow pattern of the second clause group includes: Perform pattern keyword detection on multiple clauses in the second clause group; The at least one workflow pattern is determined based on at least one detected pattern keyword, each of the at least one pattern keyword being used to indicate a workflow pattern in the at least one workflow pattern.

8. The text processing method according to any one of claims 1 to 7, characterized in that, Determining the workflow label corresponding to each clause in the first clause group includes: Each clause in the first clause group is matched with a plurality of workflow tags included in the tag dictionary to determine the workflow tag corresponding to each clause in the first clause group from the tag dictionary.

9. The text processing method according to claim 8, characterized in that, The tag dictionary also includes predefined values ​​corresponding to the plurality of workflow tags; matching each clause in the first clause group with the plurality of workflow tags included in the tag dictionary to determine the workflow tag corresponding to each clause in the first clause group from the tag dictionary includes: Perform semantic similarity processing to match each clause in the first clause group with the predefined value in the tag dictionary; The workflow label corresponding to the predefined value that has the highest similarity score with respect to the clause is determined as the workflow label corresponding to the clause.

10. A computing device, characterized in that, include: Memory; At least one processor, said at least one processor being connected to the memory; The memory is used to store computer instructions, which, when executed by the at least one processor, cause the computing device to perform the following: Receive text describing the flow of a task for which a workflow is to be generated, as input to a workflow generation application executed on the computing device; Obtain a first clause group from the text describing the process of the task, wherein the first clause group includes a plurality of clauses arranged in the order in which the process occurs, and each of the plurality of clauses in the first clause group includes a workflow component for indicating information of a corresponding functional unit among a plurality of functional units for performing the task; Determine a workflow label corresponding to each clause in the first clause group, wherein the workflow label corresponding to the plurality of clauses in the first clause group includes a first workflow label for indicating a workflow component or a second workflow label for indicating a workflow mode; The workflow for the task is generated based on the workflow tags corresponding to the plurality of clauses in the first clause group; Generate a visual representation of the generated workflow for display by the workflow generation application.

11. The computing device according to claim 10, characterized in that, The computer instructions, when executed by the at least one processor, cause the computing device to perform: The text describing the process of the task is decomposed to obtain a third clause group, wherein each of the multiple clauses included in the third clause group includes at most one workflow component. The clauses in the third clause group are filtered to obtain a fourth clause group, wherein each clause in the fourth clause group includes a workflow component; The clauses in the fourth clause group are rearranged to obtain the first clause group, or the clauses in the fourth clause group are rearranged to obtain the second clause group, and at least one identifier of at least one workflow mode is inserted into the second clause group to obtain the first clause group, wherein the at least one workflow mode is used to indicate the structure of the process of the task and the relationship between the multiple functional units.

12. The computing device according to claim 11, characterized in that, The computer instructions, when executed by the at least one processor, cause the computing device to perform: Sentence boundary detection is performed on the text to obtain multiple sentences of the text; The third clause group is obtained by splitting each sentence in the multiple sentences of the text into one or more clauses using a first encoder-based model.

13. The computing device according to claim 11, characterized in that, The computer instructions, when executed by the at least one processor, cause the computing device to perform: The second encoder-based model is used to determine whether each clause in the plurality of clauses in the third clause group includes a workflow component; If the clause in the third clause group contains zero workflow components, then the clause in the third clause group is deleted to obtain the fourth clause group.

14. The computing device according to claim 11, characterized in that, The computer instructions, when executed by the at least one processor, cause the computing device to perform: The plurality of clauses in the fourth clause group are reordered by a third encoder-based model, wherein the input of the third encoder-based model is two clauses in the fourth clause group, and the output of the third encoder-based model is information indicating whether the order of the two clauses is correct.

15. The computing device according to claim 11, characterized in that, The identifier of each workflow mode in the at least one workflow mode includes a workflow mode boundary indication sentence and / or a workflow mode indication; the computer instructions, when executed by the at least one processor, cause the computing device to implement: Determine at least one workflow pattern of the second clause group; For each of the at least one workflow pattern, the first clause group is obtained by performing at least one of the following on the second clause group: If there is no clause in the second clause group that indicates the boundary of the workflow pattern, then the workflow pattern boundary indication clause is inserted into the second clause group. or If there is no clause in the second clause group that serves as an indication of the workflow mode, then the workflow mode indication of the workflow mode is inserted into the second clause group.

16. The computing device according to claim 15, characterized in that, The computer instructions, when executed by the at least one processor, cause the computing device to perform: Perform pattern keyword detection on multiple clauses in the second clause group; The at least one workflow pattern is determined based on at least one detected pattern keyword, wherein each of the at least one pattern keyword is used to indicate a workflow pattern in the at least one workflow pattern.

17. The computing device according to any one of claims 10 to 16, characterized in that, The computer instructions, when executed by the at least one processor, cause the computing device to perform: Each clause in the first clause group is matched with a plurality of workflow tags included in the tag dictionary to determine the workflow tag corresponding to each clause in the first clause group from the tag dictionary.

18. The computing device according to claim 10, characterized in that, The tag dictionary also includes predefined values ​​corresponding to the plurality of workflow tags; the computer instructions, when executed by the at least one processor, cause the computing device to implement: Perform semantic similarity processing to match each clause in the first clause group with the predefined value in the tag dictionary; The workflow label corresponding to the predefined value that has the highest similarity score with respect to the clause is determined as the workflow label corresponding to the clause.

19. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions, when executed by the computer, cause the computer to perform the following: Receive text describing the flow of a task for which a workflow is to be generated, as input to a workflow generation application executed on the computing device; Obtain a first clause group from the text describing the process of the task, wherein the first clause group includes a plurality of clauses arranged in the order in which the process occurs, and each of the plurality of clauses in the first clause group includes a workflow component for indicating information of a corresponding functional unit among a plurality of functional units for performing the task; Determine a workflow label corresponding to each clause in the first clause group, wherein the workflow label corresponding to the plurality of clauses in the first clause group includes a first workflow label for indicating a workflow component or a second workflow label for indicating a workflow mode; The workflow for the task is generated based on the workflow tags corresponding to the plurality of clauses in the first clause group; Generate a visual representation of the generated workflow for display by the workflow generation application.

20. The non-transitory computer-readable storage medium according to claim 19, characterized in that, The computer instructions, when executed by the computer, cause the computer to perform the following: The text describing the process of the task is decomposed to obtain a third clause group, wherein each of the multiple clauses included in the third clause group includes at most one workflow component. The clauses in the third clause group are filtered to obtain a fourth clause group, wherein each clause in the fourth clause group includes a workflow component; The clauses in the fourth clause group are rearranged to obtain the first clause group, or the clauses in the fourth clause group are rearranged to obtain the second clause group, and at least one identifier of at least one workflow mode is inserted into the second clause group to obtain the first clause group, wherein the at least one workflow mode is used to indicate the structure of the process of the task and the relationship between the multiple functional units.

Images