To-do task generation method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of computers, and provides a to-do task generation method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a plurality of labeling tasks and a description text of each labeling task; based on the description text of each labeling task, performing machine language transcription on each labeling task to obtain a transcription task corresponding to each labeling task; determining a task thinking chain based on the description text of each labeling task; and based on the task thinking chain, integrating each labeling task and each transliteration task to obtain a plurality of to-do tasks. According to the method, expression habits and cognitive patterns of human beings can be better captured in the process of converting the labeling tasks into the machine language data, so that the generated to-do tasks can be consistent with thinking modes of the human beings, and the comprehensiveness, orderliness and intelligibility of the to-do tasks are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, electronic device and storage medium for generating to-do tasks. Background Art

[0002] Efficiently organizing to-do tasks is crucial for improving meeting efficiency and execution. However, the application of large models in meeting scenarios currently faces many challenges, especially when processing meeting transcripts and generating to-do tasks.

[0003] Currently, most to-do task generation solutions rely on large-scale language models to extract to-do tasks from meeting transcripts. However, given the complexity of to-do tasks, large-scale language models are often trained using high-quality, manually annotated data. However, this high-quality, manually annotated data suffers from the diverse styles and varying summarization capabilities of annotators. Furthermore, there are significant differences between human and machine expression habits. Directly training models using annotated data is ineffective, resulting in model-generated to-do tasks that are generally incomplete, fragmented, and generally incomprehensible. Summary of the Invention

[0004] The present invention provides a to-do task generation method, device, electronic device and storage medium, which are used to solve the defects in the prior art.

[0005] The present invention provides a method for generating to-do tasks, comprising the following steps: Get multiple annotation tasks and the description text of each annotation task; Based on the description text of each annotation task, each annotation task is transcribed into machine language to obtain the corresponding transcription task of each annotation task; Determine the task thinking chain based on the description text of each labeled task; Based on the task thinking chain, each labeling task and each transcription task are integrated to obtain a plurality of to-do tasks.

[0006] According to a method for generating to-do tasks provided by the present invention, the method of performing machine language transcription on each annotation task based on the description text of each annotation task to obtain a transcription task corresponding to each annotation task includes: Based on the large language model, the description text of each annotation task is applied to polish each annotation task to obtain the first transcription task corresponding to each annotation task; and / or, Based on a large language model, multiple candidate tasks are extracted from the description text of each annotation task, and information completion and logical integration are performed on each annotation task based on the multiple candidate tasks to obtain a second transcription task corresponding to each annotation task.

[0007] According to a to-do task generation method provided by the present invention, extracting multiple candidate tasks from the description text of each labeled task includes: Determine the task extraction requirements of each description text based on its position in the original text; Based on the task extraction requirements of each description text, multiple candidate tasks are extracted from the description text of each annotation task.

[0008] According to a to-do task generation method provided by the present invention, the task extraction requirements based on each description text are used to extract multiple candidate tasks from the description text of each marked task, including: Based on the task extraction requirements of each description text, construct the prompt text of each description text; Based on the hint text of each description text, multiple candidate tasks are extracted from the description text of each annotation task.

[0009] According to a method for generating to-do tasks provided by the present invention, the description text of each marked task is obtained based on the following steps: Obtain the initial description text corresponding to each annotation task from the original text; When the initial description text of any labeling task contains description sentences of other labeling tasks, obtaining the remaining description sentences of the other labeling tasks from the original text; Adding the remaining description sentences to the initial description text of any one of the annotation tasks to obtain candidate description text; A description text of any one of the labeling tasks is determined from a plurality of target description texts, where the target description text is a candidate description text containing a description sentence of any one of the labeling tasks.

[0010] According to a to-do task generation method provided by the present invention, determining the description text of any one of the labeled tasks from a plurality of target description texts includes: Based on the position of each target description text in the original text, the position of the description sentence describing any labeling task in each target description text in the original text, and the number of labeling tasks corresponding to each target description text, the description text of any labeling task is determined from the multiple target description texts.

[0011] According to a method for generating to-do tasks provided by the present invention, the step of obtaining the initial description text corresponding to each annotated task from the original text includes: Obtaining the annotated sentences in the original text for each annotation task; Determining context sentences for each annotation task from the original text based on the position of each annotation sentence in the original text; The context sentence of each annotation task and the annotation sentence of each annotation task are used as the initial description text of each annotation task.

[0012] The present invention also provides a to-do task generation device, comprising the following modules: An acquisition unit, used to acquire multiple labeling tasks and description text of each labeling task; A transcription unit is used to perform machine language transcription on each annotation task based on the description text of each annotation task to obtain a transcription task corresponding to each annotation task; The determination unit is used to determine the task thinking chain based on the description text of each annotation task; The generating unit is used to integrate the labeling tasks and the transcription tasks based on the task thinking chain to obtain a plurality of to-do tasks.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-described methods for generating to-do tasks is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for generating to-do tasks.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for generating to-do tasks.

[0016] The to-do task generation method, device, electronic device, and storage medium provided by the present invention convert manually annotated tasks into machine language data that aligns with human expectations, generating transcription tasks. These tasks then combine the annotation and transcription tasks to generate to-do tasks that align with human cognition. Because the process of converting the annotation tasks into machine language data better captures human expression habits and cognitive patterns, the generated to-do tasks are consistent with human thinking, thereby improving the comprehensiveness, organization, and intelligibility of the to-do tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is one of the flow charts of the to-do task generation method provided by the present invention.

[0019] Figure 2This is the second flow chart of the method for generating to-do tasks provided by the present invention.

[0020] Figure 3 This is the third flow chart of the to-do task generation method provided by the present invention.

[0021] Figure 4 This is the fourth flow chart of the to-do task generation method provided by the present invention.

[0022] Figure 5 This is the fifth flow chart of the to-do task generation method provided by the present invention.

[0023] Figure 6 This is the sixth flow chart of the to-do task generation method provided by the present invention.

[0024] Figure 7 This is the seventh flow chart of the to-do task generation method provided by the present invention.

[0025] Figure 8 It is a flowchart of the method for obtaining description texts for each annotation task provided by the present invention.

[0026] Figure 9 It is a flowchart of the second transcription task generation method provided by the present invention.

[0027] Figure 10 It is a structural diagram of the to-do task generating device provided by the present invention.

[0028] Figure 11 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0029] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0030] Currently, most to-do task generation solutions rely on large-scale language models to extract to-do tasks from meeting transcripts. Since meeting transcripts are typically 20,000 to 30,000 words long, this far exceeds the input length limit of large-scale models, resulting in a decrease in model processing capabilities.

[0031] In addition, considering the complexity of the tasks to be done, large-scale language models are often trained using high-quality data that has been manually annotated. However, high-quality data that has been manually annotated has problems such as diverse annotation styles and uneven inductive abilities of the annotators, and there are significant differences between human expression habits and machine expression habits. Directly using the annotated data to train the model is not very effective, resulting in the to-do tasks generated by the model being generally incomplete, trivial, and having low overall intelligibility.

[0032] To this end, the present invention provides a method for generating to-do tasks, which converts manually annotated annotation tasks into machine language data that meets human expectations to obtain transcription tasks, and combines the annotation tasks and transcription tasks to generate to-do tasks that align with human cognition. Since the process of converting annotation tasks into machine language data can better capture human expression habits and cognitive patterns, the generated to-do tasks can be consistent with human thinking patterns, thereby improving the comprehensiveness, organization and intelligibility of the to-do tasks. Among them, this method can be applied to the generation of to-do tasks in conference scenarios, and can also be applied to the generation of to-do tasks in scenarios such as education and training, schedule management, etc. In order to facilitate the explanation of the technical solution of the present invention, the following embodiments are all described using the generation of to-do tasks in conference scenarios as an example.

[0033] Figure 1 This is one of the flow charts of the method for generating to-do tasks provided by the present invention, such as Figure 1 As shown, the method includes step 110 , step 120 , step 130 and step 140 .

[0034] Step 110: Acquire multiple labeling tasks and description text of each labeling task.

[0035] Here, the annotation task refers to a specific to-do task that is manually annotated. The annotation task can be manually extracted from the original text (the original text here can be a meeting transcription text). For example, if the meeting transcription text records "complete the product design document before next week", then "complete the product design document" can be manually extracted as the annotation task.

[0036] The description text of the annotation task is a text that explains the annotation task in detail. This text may include information such as the task content, background, objectives, requirements, etc. The description text is usually text related to the annotation task obtained from the meeting transcript. It includes the text used to describe the annotation task and the context text of the text. For example, if the annotation task is "Complete the product design document", the description text of the annotation task may be "Based on customer feedback, modify and improve the product design document to ensure that the document is ready before the review meeting next week."

[0037] Since labeling tasks are manually refined and expressed in human natural language, the model faces difficulties in understanding and converting tasks based on labeling tasks due to the significant differences between human natural language and machine language. For example, if the labeling task is "collect customer feedback as soon as possible," the "as soon as possible" is clearly a natural language expression, which the model cannot accurately understand. Consequently, when extracting tasks based on the labeling task, it may mistakenly interpret "as soon as possible" as a specific time period or missing instructions, resulting in incomplete and trivial extracted tasks.

[0038] As an optional embodiment, when manually extracting annotation tasks from conference transcripts, relevant sentences for the annotation task can be identified simultaneously, and context sentences can be extracted from the conference transcripts based on the relevant sentences, with the relevant sentences and context sentences used as the descriptive text for the corresponding annotation task. The context sentences can be determined based on preset rules, such as the first n sentences and the last n sentences of the relevant sentence can be used as context sentences. The context sentences can also be sentences obtained through semantic similarity analysis, which have a strong correlation with the relevant sentence in terms of topic or content, to help more fully understand the task background and details.

[0039] Step 120 : Based on the description text of each annotation task, perform machine language transcription on each annotation task to obtain a transcription task corresponding to each annotation task.

[0040] Specifically, machine language transcription refers to converting annotation tasks expressed in human natural language into a machine language format that computers can understand and process, enabling machines to more accurately interpret task information. A transcription task is a task representation formed after the machine language transcription process. This representation is typically in a standardized, structured information format, making it more suitable for further processing and management by computer programs. Transcription tasks typically include the necessary details and instructions to ensure that the machine understands the specific operations to be performed.

[0041] For example, the labeling task is "Complete product design documents", and the corresponding description text is "Based on customer feedback, modify and improve the product design documents to ensure that the documents are ready before the review meeting next week." Obviously, the labeling task lacks a deadline, priority, and specific execution instructions, which makes it impossible for the computer to understand the urgency of the task and the specific requirements for completion, and thus cannot generate accurate execution instructions. In this regard, an embodiment of the present invention transcribes the labeling task into machine language based on the description text, and the transcribed text obtained is "Generate a task object (task object), the content is "Modify product design documents", the priority is "high", and the deadline is "before the review meeting next week"." The transcribed text provides detailed information about the task, enabling the computer to effectively parse and process the task to ensure timely completion.

[0042] It's important to note that the description text is extracted directly from the original text of the meeting transcript, containing specific details relevant to the annotation task. This detailed information is usually expressed in natural language and directly reflects the content discussed by the participants in the meeting. Therefore, machine language transcription based on this information with clear context and intent can effectively preserve the original meaning of the task.

[0043] As an optional embodiment, the descriptive text of each annotation task and each annotation task can be input into a large language model. The large language model then performs machine language transcription on each annotation task based on the descriptive text, obtaining the corresponding transcription task. Large language models refer to natural language processing models trained using deep learning techniques. These models have powerful text understanding and generation capabilities and can include models such as BERT (Bidirectional Encoder Representation Transformer) and T5 (Text-to-Text Transformer).

[0044] Step 130: Determine the task thinking chain based on the description text of each annotation task.

[0045] Specifically, the task thinking chain refers to a structure that coherently represents the logical relationship and execution order between multiple tasks. This structure helps to clarify the dependencies and execution processes between tasks. It is used to characterize how tasks are connected, influence and depend on each other, ensuring the comprehensiveness and rationality of task execution.

[0046] Considering the logical relationship and execution order between tasks, if the task thinking chain is not determined and the to-do tasks are determined only based on the marking tasks and / or transcription tasks, there may be potential risks such as unsmooth task execution, omission of key steps, or duplication of work.

[0047] For example, the description of annotation task 1 is "Modify and improve the product design document based on customer feedback," while the description of annotation task 2 is "Add new functional specifications to the revised design document." Analysis reveals a causal relationship between annotation tasks 1 and 2, with the former being a prerequisite for the latter. The corresponding task thought chain could be "First, modify the product design document, then add new functional specifications based on this." This suggests that if the task thought chain is unclear, there may be a lack of clear understanding of the dependencies between tasks, making it impossible to effectively execute tasks in the correct order, leading to confusion in the execution of annotation tasks 1 and 2.

[0048] As an optional implementation, each labeling task can be converted into a node based on the descriptive text, and the relationships between nodes (such as causal relationships, parallel relationships, etc.) can be represented by edges, forming a semantic graph. For example, Labeling Task 1 and Labeling Task 2 can be two nodes in the graph. If Labeling Task 1 is a predecessor of Labeling Task 2, the two nodes can be connected by a directed edge. Next, a graph algorithm (such as depth-first search (DFS) or breadth-first search (BFS)) is applied, starting from the starting node. By analyzing the connections between task nodes, the order and dependencies of each task can be determined, and the corresponding task thought chain can be output.

[0049] Step 140: Based on the task thinking chain, integrate the labeling tasks and the transcription tasks to obtain multiple to-do tasks.

[0050] Specifically, labeling tasks are obtained through manual annotation and are used to represent human needs, intentions, and actions, but they may contain information omissions and logical inconsistencies. Transcription tasks are obtained by transcribing labeling tasks into machine language. They can fill in missing information and logical inconsistencies in labeling tasks, but they only reflect the machine's understanding and processing methods, which may not fully conform to human expression habits and, therefore, fail to align with human cognition.

[0051] Based on this, considering that the task thinking chain can provide logical connections and sequential relationships between tasks, when integrating the various labeling tasks and transcription tasks based on the task thinking chain, the transcription tasks can be used to supplement the information omissions and logical inconsistencies in the labeling tasks, and the information can be optimized and adjusted in combination with the expression of human natural language, so that the multiple to-do tasks obtained by integration can also be aligned with human cognition, thereby ensuring the comprehensiveness, orderliness and comprehensibility of the extracted to-do tasks.

[0052] The to-do task generation method provided by an embodiment of the present invention converts manually annotated tasks into machine language data that aligns with human expectations, generating transcription tasks. The transcription tasks are then combined to generate to-do tasks that align with human cognition. Because the process of converting the annotation tasks into machine language data better captures human expression habits and cognitive patterns, the generated to-do tasks are consistent with human thinking, thereby improving the comprehensiveness, organization, and understandability of the to-do tasks.

[0053] Based on the above embodiments, Figure 2 This is the second flow chart of the method for generating to-do tasks provided by the present invention. Figure 2 As shown, the method includes the following steps: Step 210: Acquire multiple labeling tasks and description text of each labeling task.

[0054] Specifically, a labeling task refers to a specific to-do task that has been manually annotated. This labeling task can be manually extracted from the original text (the original text in this case can be a conference transcript). The labeling task description is a text that provides a detailed description of the labeling task. This text may include information such as the task content, background, objectives, and requirements.

[0055] Step 220: Based on the large language model, apply the description text of each annotation task to polish each annotation task to obtain a first transcription task corresponding to each annotation task; and / or, Based on a large language model, multiple candidate tasks are extracted from the description text of each annotation task. Based on the multiple candidate tasks, information is supplemented and logically integrated for each annotation task to obtain the second transcription task corresponding to each annotation task. The above transcription tasks include the first transcription task and / or the second transcription task.

[0056] Specifically, considering that the expression of each annotation task may not be clear or lack necessary contextual information, it is necessary to apply the description text of each annotation task to polish each annotation task to obtain the first transcription task. Polishing refers to the linguistic adjustment and optimization of the annotation task to make it more concise and clear, and conform to standard expression conventions.

[0057] For example, the task labeled "Communicate with customers promptly to understand their needs" is not specific enough and the timeline is unclear, which may lead the executor to misunderstand the urgency of communication. After polishing the task based on the descriptive text, it can be adjusted to "Conduct a demand communication with customers within this week to ensure understanding of their specific needs." This avoids communication delays caused by unclear timelines and ensures more efficient task execution, resulting in the first transcription task: "Conduct a demand communication with customers within this week to ensure understanding of their specific needs."

[0058] Considering that large-scale language models have powerful language understanding and generation capabilities, they can accurately understand the context and semantics of the text, thereby performing more accurate polishing. To this end, the embodiment of the present invention is based on a large-scale language model, applies the description text of each annotation task, polishes each annotation task, and obtains the first transcription task corresponding to each annotation task. Among them, the first transcription task can be understood as the polished annotation task, and the first transcription task is more operational and clear in execution. The large-scale language model can be a BERT model or other models, and the embodiment of the present invention does not specifically limit this.

[0059] As an optional embodiment, a polishing prompt text can be constructed to guide the large language model to generate a complete statement of the annotation task. The polishing prompt text is input into the large language model, and the large language model polishes each annotation task to obtain the first transcription task.

[0060] Among them, the large language model can be trained based on the following steps: First, a large amount of training data needs to be collected and prepared. This training data typically comes from various text sources such as books, news, and social media. This training data is then preprocessed, including steps such as word segmentation and noise removal. Next, the preprocessed training data is fed into the model. Backpropagation and optimization algorithms (such as stochastic gradient descent (SGD) and the Adam optimization algorithm) are used to adjust the model's parameters to minimize the error between the model's predictions and the true labels. Furthermore, during the training process, model selection and evaluation are required to ensure the model's generalization and performance. The true labels here can be task descriptions that have been refined and optimized for expression.

[0061] The above training data can be constructed based on the following steps: Obtain a large amount of text data from different sources. This data can include books, papers, news articles, social media posts, and Wikipedia. This text should cover a variety of topics and styles to increase the diversity and coverage of the model.

[0062] For each text source, multiple tasks are manually labeled to form multiple sample data, which can be regarded as the basic input for model learning.

[0063] Clean and preprocess the sample data, including removing extra whitespace, HTML tags, and non-text symbols, and performing tokenization, stemming, and lemmatization. Optionally, record key metrics for each sample data, such as text length, keyword distribution, and topic categories, to facilitate subsequent analysis and model evaluation.

[0064] Considering that the style and expression of texts may vary, in order to enhance the generalization ability of the model, it is necessary to ensure that the training data contains samples of multiple language styles, cultural backgrounds, and writing habits.

[0065] The processed sample data is divided into training data, validation data, and test data to ensure the effectiveness of model training and the objectivity of evaluation.

[0066] Furthermore, considering that labeling tasks may present issues such as missing information and incoherent logic, the present invention first extracts multiple candidate tasks from the descriptive text of each labeling task based on a large language model. While the candidate tasks initially extracted by the large language model often exhibit high misjudgment rates and fragmented and incomplete content, these candidate tasks are summarized from the descriptive text. This means that the candidate tasks are traceable in the original text and represent condensed and refined information learned by the machine. While these candidate tasks cannot be directly used as to-do tasks, they can be used to effectively integrate the labeling tasks with incomplete information and logic. The resulting second transcription task not only addresses the issues of low machine extraction accuracy, fragmented content that cannot be merged, and incomplete to-do lists, but also highly preserves human cognitive ability to extract meeting to-do items.

[0067] The training process for the large-scale language model corresponding to the second transcription task can be referenced to the training process for the large-scale language model corresponding to the first transcription task, and the corresponding training data construction method can also be referenced to the training data construction method described above. The true labels of the large-scale language model corresponding to the second transcription task are different from the true labels of the first transcription task. The true labels for the second transcription task are the complete task description after information completion and logical integration, while the true labels for the first transcription task are the task description after language polishing and expression optimization.

[0068] It's important to note that the first transcription task focuses on polishing—improving the language and clarifying the existing annotation task to enhance readability and accuracy. The second transcription task focuses on information extraction and completion, mining potential information from the descriptive text and enhancing the completeness and logic of the information by integrating multiple candidate tasks. In other words, the first transcription task aims to make the annotation task more readable and clear, while the second transcription task ensures the completeness and logic of the task information.

[0069] Step 230: Determine the task thinking chain based on the description text of each annotation task.

[0070] Specifically, the task thinking chain refers to a structure that coherently represents the logical relationship and execution order between multiple tasks. It is used to characterize how tasks are connected, influence and depend on each other, ensuring the comprehensiveness and rationality of task execution.

[0071] Step 240: Based on the task thinking chain, integrate the labeling tasks and the transcription tasks to obtain multiple to-do tasks.

[0072] Specifically, considering that the task thinking chain can provide logical connections and sequential relationships between tasks, when integrating various labeling tasks and transcription tasks based on the task thinking chain, the transcription tasks can be used to supplement the information omissions and logical inconsistencies in the labeling tasks, and the information can be optimized and adjusted in combination with the expression of human natural language, so that the multiple to-do tasks obtained by integration can also be aligned with human cognition, thereby ensuring the comprehensiveness, orderliness and comprehensibility of the extracted to-do tasks.

[0073] Based on any of the above embodiments, Figure 3 This is the third flow chart of the method for generating to-do tasks provided by the present invention, such as Figure 3 As shown, the method includes the following steps: Step 310: Acquire multiple labeling tasks and description text of each labeling task.

[0074] Specifically, a labeling task refers to a specific to-do task that has been manually annotated. This labeling task can be manually extracted from the original text (the original text in this case can be a conference transcript). The labeling task description is a text that provides a detailed description of the labeling task. This text may include information such as the task content, background, objectives, and requirements.

[0075] Step 320: Based on the large language model, apply the description text of each annotation task to polish each annotation task to obtain a first transcription task corresponding to each annotation task; and / or, Based on a large language model and the position of each description text in the original text, the task extraction requirements of each description text are determined. Based on the task extraction requirements of each description text, multiple candidate tasks are extracted from the description text of each annotation task. Based on the multiple candidate tasks, information is supplemented and logically integrated for each annotation task to obtain the second transcription task corresponding to each annotation task. The above transcription tasks include the first transcription task and / or the second transcription task.

[0076] The original text here can be understood as the original corpus corresponding to each description text. For example, in the case of a meeting scenario, the original text here can be the transcript of the meeting. Typically, description text related to to-do tasks is concentrated in a certain area of ​​the original text. In other words, description text related to to-do tasks is more likely to appear within this area, and less likely to appear outside this area.

[0077] For example, for meeting transcription texts, statistics show that the first half of the meeting focuses on event discussion, while the second half will strengthen the effective arrangement of meeting agenda. In other words, the content of the second half of the meeting is more likely to involve to-do tasks.

[0078] Based on this, the embodiment of the present invention determines the task extraction requirements of each description text based on the position of each description text in the original text. For example, for positions with a high probability of involving to-do tasks (such as the second half of the conference transcription text), the task extraction requirements are relatively loose, so that the to-do tasks can be fully extracted; for positions with a low probability of involving to-do tasks (such as the first half of the conference transcription text), the task extraction requirements are relatively strict, so that irrelevant content can be avoided from being extracted. Among them, the task extraction requirements can be understood as specific standards and conditions for extracting tasks. For example, a relatively strict task extraction requirement can be "only extracting clearly indicated to-do tasks", and a relatively loose task extraction requirement can be "trying to extract all possible to-do task related information."

[0079] As an optional embodiment, the task extraction requirements corresponding to different positions in the original text can be determined based on preset rules, such as the task extraction requirements corresponding to the first half of the original text are relatively loose, and the task extraction requirements corresponding to the second half of the original text are relatively strict; the task extraction requirements corresponding to different positions in the original text can also be determined through a machine learning model, such as automatically learning the correlation between different positions and tasks based on training data.

[0080] Step 330: Determine the task thinking chain based on the description text of each annotation task.

[0081] Specifically, the task thinking chain refers to a structure that coherently represents the logical relationship and execution order between multiple tasks. It is used to characterize how tasks are connected, influence and depend on each other, ensuring the comprehensiveness and rationality of task execution.

[0082] Step 340: Based on the task thinking chain, integrate the labeling tasks and the transcription tasks to obtain multiple to-do tasks.

[0083] Specifically, considering that the task thinking chain can provide logical connections and sequential relationships between tasks, when integrating various labeling tasks and transcription tasks based on the task thinking chain, the transcription tasks can be used to supplement the information omissions and logical inconsistencies in the labeling tasks, and the information can be optimized and adjusted in combination with the expression of human natural language, so that the multiple to-do tasks obtained by integration can also be aligned with human cognition, thereby ensuring the comprehensiveness, orderliness and comprehensibility of the extracted to-do tasks.

[0084] Based on any of the above embodiments, Figure 4 This is the fourth flow chart of the method for generating to-do tasks provided by the present invention, such as Figure 4 As shown, the method includes the following steps: Step 410: Acquire multiple labeling tasks and description text of each labeling task.

[0085] Specifically, a labeling task refers to a specific to-do task that has been manually annotated. This labeling task can be manually extracted from the original text (the original text in this case can be a conference transcript). The labeling task description is a text that provides a detailed description of the labeling task. This text may include information such as the task content, background, objectives, and requirements.

[0086] Step 420: Based on the large language model, apply the description text of each annotation task to polish each annotation task to obtain a first transcription task corresponding to each annotation task; and / or, Based on a large language model and the position of each description text in the original text, the task extraction requirements for each description text are determined; based on the task extraction requirements for each description text, prompt text for each description text is constructed; based on the prompt text for each description text, multiple candidate tasks are extracted from the description text of each annotation task, and information completion and logical integration are performed on each annotation task based on the multiple candidate tasks to obtain the second transcription task corresponding to each annotation task. The above-mentioned transcription tasks include the first transcription task and / or the second transcription task.

[0087] Specifically, the prompt text provides specific instructions or guidance for task extraction. It helps identify and guide key information during the extraction process. Prompt text guides task extraction, clarifying the focus and scope of extraction, enabling the model to more accurately find relevant information.

[0088] When constructing prompt text, you can first understand the task extraction requirements for each description text. These requirements define the type of task information to be extracted in a specific context. For example, for text with a high probability of a to-do task, the prompt text can be more lenient, allowing for more relevant information to be extracted. For text with a low probability of a to-do task, the prompt text should be more stringent, ensuring that the information extracted is highly relevant to the to-do task.

[0089] Optionally, the description texts can be analyzed to extract keywords, key phrases, or key sentences, which form the basis of the prompt text. For example, words such as "meeting decision" and "task assignment" can be extracted from the description text as key information. The extracted key information is organized into structured prompt text, which can be a natural language description or a formatted template. For example, "Please extract the tasks discussed in the meeting" or "List the action items related to the meeting decision."

[0090] In addition, adding examples and contextual information to the prompt text can help better understand the extraction target. For example, the prompt text can include "In the past meeting, we identified the following to-do tasks:" to guide the system to focus on similar text.

[0091] After constructing the prompt text for each description text, the prompt text can be input into a large language model, which extracts corresponding candidate tasks from the description text of each annotation task based on the guidance in the prompt text.

[0092] Step 430: Determine the task thinking chain based on the description text of each annotation task.

[0093] Specifically, the task thinking chain refers to a structure that coherently represents the logical relationship and execution order between multiple tasks. It is used to characterize how tasks are connected, influence and depend on each other, ensuring the comprehensiveness and rationality of task execution.

[0094] Step 440: Based on the task thinking chain, integrate the labeling tasks and the transcription tasks to obtain multiple to-do tasks.

[0095] Specifically, considering that the task thinking chain can provide logical connections and sequential relationships between tasks, when integrating various labeling tasks and transcription tasks based on the task thinking chain, the transcription tasks can be used to supplement the information omissions and logical inconsistencies in the labeling tasks, and the information can be optimized and adjusted in combination with the expression of human natural language, so that the multiple to-do tasks obtained by integration can also be aligned with human cognition, thereby ensuring the comprehensiveness, orderliness and comprehensibility of the extracted to-do tasks.

[0096] Based on any of the above embodiments, Figure 5 This is the fifth flow chart of the method for generating to-do tasks provided by the present invention, such as Figure 5 As shown, the method includes the following steps: Step 510: obtain multiple labeling tasks and description texts of each labeling task; wherein, the description texts of each labeling task are obtained based on the following steps: obtain the initial description texts corresponding to each labeling task from the original text; when the initial description text of any labeling task contains description sentences of the remaining labeling tasks, obtain the remaining description sentences of the remaining labeling tasks from the original text; add the remaining description sentences to the initial description text of any labeling task to obtain candidate description texts; determine the description text of any labeling task from multiple target description texts, and the target description text is the candidate description text containing the description sentence of any labeling task.

[0097] Specifically, the initial description text includes the annotation sentences related to the annotation task and the context sentences of the annotation sentences that were manually annotated in the original text. However, considering that the initial description text may contain description sentences from other annotation tasks due to unclear expression, missing context or information omission when manually annotating the original text, the initial description text contains not only description sentences related to the annotation task, but also description sentences related to other annotation tasks.

[0098] Taking into account that when extracting to-do tasks later, most of the task extraction is based on the description text. If the description text contains incomplete sentences for describing the task, it may cause missing or incorrect information when extracting the task, thereby affecting the accuracy and completeness of the task. In this regard, the embodiment of the present invention obtains the remaining description sentences of the remaining annotation tasks from the original text when it is determined that the initial description text of any annotation task contains the description sentences of the remaining annotation tasks. The remaining description sentences and the description sentences of the remaining annotation tasks in the initial description text together constitute a complete description text, ensuring that the final candidate description text can fully reflect the information of all relevant tasks. In other words, the remaining description sentences are added to the initial description text of any annotation task, and the obtained candidate description text can fully describe the remaining annotation tasks.

[0099] Taking into account that the description sentences of any of the above-mentioned labeling tasks may also exist in the candidate description texts of other labeling tasks, the embodiment of the present invention determines the description text of any labeling task from multiple target description texts. Optionally, any target description text can be selected as the description text of any labeling task; or the correlation between each target description text and any labeling task can be determined based on the semantic information of each target description text. The greater the correlation, the stronger the relevance of the target description text to any labeling task, and the target description text with the maximum correlation can be used as the description text of any labeling task; or it can be determined whether the description sentence used to describe any labeling task in each target description text is located in the center position. If so, it indicates that the central theme of the target description text is most likely to be describing the labeling task, and the target description text can be used as the description text of any labeling task.

[0100] Step 520: Based on the description text of each annotation task, perform machine language transcription on each annotation task to obtain a transcription task corresponding to each annotation task.

[0101] Specifically, machine language transcription refers to converting annotation tasks expressed in human natural language into a machine language format that computers can understand and process, enabling machines to more accurately interpret task information. A transcription task is the task expression formed after the machine language transcription process.

[0102] Step 530: Determine the task thinking chain based on the description text of each annotation task.

[0103] Specifically, the task thinking chain refers to a structure that coherently represents the logical relationship and execution order between multiple tasks. It is used to characterize how tasks are connected, influence and depend on each other, ensuring the comprehensiveness and rationality of task execution.

[0104] Step 540: Based on the task thinking chain, integrate the labeling tasks and the transcription tasks to obtain multiple to-do tasks.

[0105] Specifically, considering that the task thinking chain can provide logical connections and sequential relationships between tasks, when integrating various labeling tasks and transcription tasks based on the task thinking chain, the transcription tasks can be used to supplement the information omissions and logical inconsistencies in the labeling tasks, and the information can be optimized and adjusted in combination with the expression of human natural language, so that the multiple to-do tasks obtained by integration can also be aligned with human cognition, thereby ensuring the comprehensiveness, orderliness and comprehensibility of the extracted to-do tasks.

[0106] Based on any of the above embodiments, Figure 6 This is the sixth flow chart of the method for generating to-do tasks provided by the present invention, such as Figure 6 As shown, the method includes the following steps: Step 610: Obtain multiple labeling tasks and description texts of each labeling task; wherein, the description texts of each labeling task are obtained based on the following steps: obtaining the initial description text corresponding to each labeling task from the original text; when the initial description text of any labeling task contains description sentences of the remaining labeling tasks, obtaining the remaining description sentences of the remaining labeling tasks from the original text; adding the remaining description sentences to the initial description text of any labeling task to obtain candidate description texts; based on the position of each target description text in the original text, the position of the description sentence describing any labeling task in each target description text in the original text, and the number of labeling tasks corresponding to each target description text, determining the description text of any labeling task from multiple target description texts.

[0107] Specifically, the position of each target description text in the original text refers to the specific location index of each target description text in the original text, including its starting and ending character positions in the text. By analyzing the position of description texts in the original text, we can determine their contextual relationships and cross-references, thereby more accurately understanding their meaning. If a description text is closely related to the theme or other important information of the original text, it is likely to be more relevant.

[0108] The position of the descriptive sentence describing any annotation task in each target description text refers to the specific location in the original text of the sentence specifically used to describe a particular annotation task. Analyzing this position can help identify the direct context of the descriptive sentence and thus assess its clarity and information completeness. Effective descriptive sentences generally occupy a more prominent position in the text they belong to, rather than being relegated to secondary or marginal parts.

[0109] The number of labeling tasks corresponding to each target description text refers to the number of labeling tasks described in each target description text. For example, if any target description text describes labeling task 1 and labeling task 2, the corresponding number is 2. The smaller the number, the greater the probability that the corresponding target description text is in the core description of the corresponding labeling task.

[0110] The embodiment of the present invention can select target description texts with superior positions and semantic relevance as the description texts for any labeling task based on the positions of the target description texts in the original text, the positions of the description sentences describing any labeling task in the target description texts in the original text, and the number of target description texts.

[0111] Step 620: Based on the description text of each labeling task, perform machine language transcription on each labeling task to obtain a transcription task corresponding to each labeling task.

[0112] Specifically, machine language transcription refers to converting annotation tasks expressed in human natural language into a machine language format that computers can understand and process, enabling machines to more accurately interpret task information. A transcription task is the task expression formed after the machine language transcription process.

[0113] Step 630: Determine the task thinking chain based on the description text of each annotation task.

[0114] Specifically, the task thinking chain refers to a structure that coherently represents the logical relationship and execution order between multiple tasks. It is used to characterize how tasks are connected, influence and depend on each other, ensuring the comprehensiveness and rationality of task execution.

[0115] Step 640: Based on the task thinking chain, integrate the labeling tasks and the transcription tasks to obtain multiple to-do tasks.

[0116] Specifically, considering that the task thinking chain can provide logical connections and sequential relationships between tasks, when integrating various labeling tasks and transcription tasks based on the task thinking chain, the transcription tasks can be used to supplement the information omissions and logical inconsistencies in the labeling tasks, and the information can be optimized and adjusted in combination with the expression of human natural language, so that the multiple to-do tasks obtained by integration can also be aligned with human cognition, thereby ensuring the comprehensiveness, orderliness and comprehensibility of the extracted to-do tasks.

[0117] Based on any of the above embodiments, Figure 7 This is the seventh flow chart of the method for generating to-do tasks provided by the present invention, such as Figure 7 As shown, the method includes the following steps: Step 710: obtain multiple labeling tasks and description texts of each labeling task; wherein, the description texts of each labeling task are obtained based on the following steps: obtain the labeling sentences of each labeling task in the original text; determine the context sentences of each labeling task from the original text based on the position of each labeling sentence in the original text; use the context sentences of each labeling task and the labeling sentences of each labeling task as the initial description texts of each labeling task; when the initial description text of any labeling task contains the description sentences of the remaining labeling tasks, obtain the remaining description sentences of the remaining labeling tasks from the original text; add the remaining description sentences to the initial description text of any labeling task to obtain candidate description texts; determine the description text of any labeling task from multiple target description texts.

[0118] Specifically, annotated sentences are manually annotated sentences in the original text that are relevant to the annotation task. These annotated sentences are typically key sentences that directly describe the corresponding annotation task. Contextual sentences help understand the background and purpose of the annotation task, thereby enabling more accurate extraction of to-do tasks. The position of each annotated sentence in the original text can include the starting and ending character positions of each annotated sentence in the original text.

[0119] Optionally, the context sentence may be a sentence that is extended forward by a number of characters starting from the start character of each marked sentence in the original text, and extended backward by a number of characters starting from the end character of each marked sentence.

[0120] Step 720: Based on the description text of each labeling task, perform machine language transcription on each labeling task to obtain a transcription task corresponding to each labeling task.

[0121] Specifically, machine language transcription refers to converting annotation tasks expressed in human natural language into a machine language format that computers can understand and process, enabling machines to more accurately interpret task information. A transcription task is the task expression formed after the machine language transcription process.

[0122] Step 730: Determine the task thinking chain based on the description text of each annotation task.

[0123] Specifically, the task thinking chain refers to a structure that coherently represents the logical relationship and execution order between multiple tasks. It is used to characterize how tasks are connected, influence and depend on each other, ensuring the comprehensiveness and rationality of task execution.

[0124] Step 740: Based on the task thinking chain, integrate the labeling tasks and the transcription tasks to obtain multiple to-do tasks.

[0125] Specifically, considering that the task thinking chain can provide logical connections and sequential relationships between tasks, when integrating various labeling tasks and transcription tasks based on the task thinking chain, the transcription tasks can be used to supplement the information omissions and logical inconsistencies in the labeling tasks, and the information can be optimized and adjusted in combination with the expression of human natural language, so that the multiple to-do tasks obtained by integration can also be aligned with human cognition, thereby ensuring the comprehensiveness, orderliness and comprehensibility of the extracted to-do tasks.

[0126] in, Figure 8 This is a flow chart of the method for obtaining the description text of each annotation task provided by the present invention. Figure 8 As shown in Figure 1, this method mainly obtains the description text of each annotation task by performing sliding window segmentation on the original text, specifically including: First, traverse each annotation task in the original text (such as the conference transcript) For di in D (where di represents the i-th annotation task and D represents the annotation task set): Step 1: Get the context window of di: 1.1 Get the ID of the annotation statement list of di in the original text, and get the maximum and minimum values ​​of the ID list: ID max =max(ID), ID min =min(ID).

[0127] 1.2 About ID min , from the original text equal to ID min -1, expand 800 words to the previous context and add it to the window as a context sentence; 1.3 ID min and ID max All the marked sentences between them are added to the window; 1.4 About ID max , from the original text equal to ID max +1, and expand 800 words to the following text. After adding it to the window as a context sentence, we get the context window of di, that is, the initial description text of di.

[0128] Step 2: Expand the context window of di: The obtained context window may contain description sentences of other annotation tasks dj, so it is necessary to further expand the context window and merge the remaining description sentences of other annotation tasks dj in the original text into the context window to avoid the situation where incomplete annotation tasks exist in the context window; 2.1 Traverse each labeling task For dj in D: Find the context window containing the dj description statement to form a to-do sub-collection ; 2.2 Traversing the To-Do Sub-Collection All To Do in For dj in : 2.2.1 If all sentences in the description statement list ID of dj cannot be completely contained in the context window of di, then the ID corresponding to dj min and ID max All the annotated sentences between are added to the context window of di; 2.2.2 Check all the labeling tasks involved in the context window of di, and see if their description statement lists are all included in the context window, and get the sliding window slice set of di , represents all the description sentences in the original text of the p-th annotation task; 2.2.3 Update the corresponding to-do sub-collection in the context window. Specifically, after the context window is expanded, the corresponding annotation tasks in the serial port will increase. Update the to-do sub-collection ; Step 3: Determine the optimal context window for each annotation task (the description text corresponding to each annotation task): After the context window is expanded, one context window may correspond to multiple labeling tasks; similarly, one labeling task may exist in multiple context windows; in order to maintain the one-to-one correspondence between labeling tasks and context windows, it is necessary to find the optimal context window for each labeling task di. The text contained in the optimal context window is the description text of the labeling task di.

[0129] Traverse the sliding window slice set For corresponding to each labeling task di in , Refers to the set of window slices containing the annotation task di, Represents the set of sliding window slices corresponding to the labeling task di.

[0130] 3.1 Calculate the following measurement index r: For each sliding window slice set , which is located in the original text , Indicates the starting character position, Indicates the ending character position.

[0131] The above measurement basis is that the farther the annotation task is from the edge of the context window, the closer it is to the center of the context window, and the fewer annotation tasks the context window corresponds to, the context window is the best window for the annotation task. , represents the number of labeling tasks described in the context window. Formula Indicates the position of the character in the original text.

[0132] 3.2 For all context windows that exist in the annotation task di , find the context window with the largest r value, which is the best context window for the labeling task di (the text in this window is the description text), and sort out the subset of labeling tasks corresponding to the context windows of all labeling tasks di ; Step 4: Output sliding window slice set , and the core to-do set corresponding to the sliding window . Represents the description text corresponding to the p-th annotation task, Represents the set of all annotation tasks described by the description text under the p-th annotation task.

[0133] After obtaining the description text of each annotation task, the sliding window slice set is substituted into the position of the {Conference Original Text Segmented Sliding Window} identifier in the prompt text, and the corresponding core to-do set of the sliding window is substituted into the position of the <Sliding Window Core To-Do> identifier in the prompt text, and the prompt text is input into the large language model, and the large language model polishes each annotation task to obtain the first transcribing task. Among them, an example of the prompt text is as follows: Please complete the incomplete meeting to-do with abbreviated expressions in <> in combination with the conference original text in {}. The sentence numbers of the relevant sentences in the meeting to-do content in <> and the conference original text in {{}} are given in (list serial numbers). Combine the context related to the to-do content to supplement the details such as the event background, key requirements of the matter, precautions, digital data / clear time information related to the event, etc. of the to-do. Date data that is not the execution time of the to-do does not need to add special formats. The modified and improved to-do should be output in a more complete, clear, smooth and natural sentence, and make sure the number of words is more than 30 and less than 70. Note: 1. It is prohibited to modify and must retain the information boxed with #Date and Time# in <>. I confirm that this part of the information is accurate. If there is no to-do date in the original to-do, it is prohibited to stereotypically generate the套用 format of "#Date and Time#" at the beginning of the sentence. 2. When modifying and improving the execution time of the to-do, the date must still be in the #Date and Time# format at the beginning of the sentence to ensure unified format for easy reading. 3. The to-do executor information marked with @Executor@ in <> is accurate. Please remove the special format and integrate it into the modified and improved to-do to form a smooth and natural expression. <Sliding Window Core To-Do> {Conference Original Text Segmented Sliding Window} The modified and improved to-do should be output in the form of a dictionary: [{{"Original To-Do 1": to-do sentence, "Modified To-Do 1": improved to-do sentence}}, {{"Original To-Do 2": to-do sentence, "Modified To-Do 2": improved to-do sentence}}].

[0134] In addition, Figure 9 is a schematic flow diagram of the second transcribing task generation method provided by the present invention. As Figure 9 shown, in the sliding window slice set the corresponding segmentation serial number , if It is considered to be in the first half of the original text; otherwise, it is in the second half. Based on the valid discrimination result, substitute the corresponding sliding window slice set into the <Conference Original Text Segmentation Sliding Window> identification position indicated by the first half prompt text or the second half prompt text, and call the large language model to obtain multiple candidate tasks.

[0135] Among them, an example of the first half prompt text is as follows: Please extract the tasks regarding the measures to be taken for further research issues involved in the <> conference original text and assign clear responsibilities for the meeting to-dos. Focus on what is clearly mentioned by one or more parties to be done next, and focus on the specific short-term phased arrangements of the conference project events. Note: 1. The to-dos related to the same matter in the meeting must be summarized into one to-do, and extract the number in the "(number)" of the <> conference original text related to the to-do, and indicate from which original text sentences the to-do is summarized, marked at the end of the to-do sentence as "[number1,number2,number3]" 2. Deeply understand the requirements for extracting to-dos given above. Please only extract the key to-dos that can be clearly discriminated. Other content that cannot be accurately discriminated as to-dos, such as suggestions and ideas discussed in the meeting, is prohibited from being extracted. The number of to-dos extracted does not exceed 2. 3. The format of the to-do must be kept unified. If the to-do involves a clear date arrangement, please mark it in the form of "#To-do Date#" at the beginning of the sentence, for example, in the form of "#Month Day#". If there is no clear date, it is prohibited to mechanically generate the format of "#Date Time#" at the beginning of the sentence. 3. Each to-do should be more than 30 words and less than 50 words. Please summarize and extract details such as the event background related to the meeting to-do, the person in charge of to-do execution, the personnel and institutions involved in the event, the precautions, and the numerical data / clear time information related to the event. Date data that is not the to-do execution time does not need to add a special format. 4. Please summarize and extract each meeting to-do in a complete, clear, and smooth natural sentence for easy reading and execution of the to-do. <Conference Original Text Segmentation Sliding Window> If there are no matters in the <> conference original text that can be accurately discriminated as to-dos, directly output "No to-do"; otherwise, output the extracted to-dos in the form of a list ["To-do 1", "To-do 2"].

[0136] An example of the second half prompt text is as follows: Please extract the meeting to-dos involved in the <> conference original text, focus on what is clearly mentioned by one or more parties to be done next, and focus on the short-term phased arrangements. Note: 1. For todos related to the same matter in the meeting, they must be grouped into one todo. Extract the number in the "(number)" in the meeting original text <> related to the todo, and indicate from which original text sentences the todo is summarized, marked at the end of the todo sentence as "[number1,number2,number3]" 2. Clearly define the requirements for extracting todos given above, extract the key todos that you are sure about, and prohibit misjudging some discussion content or suggestions as todos 3. The format of todos must be kept unified. If there is a clear date arrangement involved in the todo, mark it at the beginning of the sentence in the form of "#Todo Date#" for reference, such as in the form of "#Month Day#". If there is no clear date, do not stereotype and generate the format of "#Date Time#" at the beginning of the sentence 3. Each todo should have more than 30 words and less than 50 words. Please be sure to summarize and extract details such as the event background of the meeting todo, the person in charge of todo execution, the personnel and organizations involved in the event, precautions, digital data / clear time information related to the event, etc. Date data that is not the todo execution time does not need to add a special format 4. Please summarize and extract each meeting todo in a complete, clear, and smooth natural sentence for easy reading and execution of the todo <Meeting Original Text Segmentation Sliding Window> If there are no pending matters in the <> meeting original text, directly output "No todo"; otherwise, output the extracted todos in the form of a list ["Todo 1","Todo 2","Todo 3"].

[0137] In the above example, in the first half of the prompt text, the task extraction requirements are "Deeply understand the requirements for extracting todos given above. Please only extract the key todos that can be clearly judged. Other content that cannot be accurately judged as todos, such as suggestions and ideas discussed in the meeting, is prohibited from being extracted. The number of extracted todos does not exceed 2". In the second half of the prompt text, the task extraction requirements are "Clearly define the requirements for extracting todos given above, extract the key todos that you are sure about, and prohibit misjudging some discussion content or suggestions as todos". Obviously, it can be seen that the task extraction requirements in the first half of the prompt text are stricter than those in the second half of the prompt text

[0138] After obtaining the candidate tasks, substitute multiple candidate tasks and labeled tasks into the supplementary prompt text. The large language model will perform information supplementation and logical integration on each labeled task according to the supplementary prompt text to obtain the second transcription task

[0139] Among them, an example of the supplementary prompt text is as follows Please extract the meeting todos related to the brief todo outline given in <> from the rich and redundant meeting todos given in {}, and summarize and organize them into the complete todo items for supplementing details in <>. Note: 1. Since there are many duplicates in the meeting to-do list in {}, it is necessary to filter and remove them based on the main to-do list in <> 2.{} Please integrate the same to-do content into a complete, detailed, clear and effective to-do. The original numbers of the to-do related meetings must be effectively merged, and the merged result should be marked at the end of the to-do sentence with "[number1, number2, number3]". 3. After sorting, each to-do item must be no less than 40 words and no more than 80 words, including the background of the to-do event, the person in charge / department, precautions, numerical data related to the event / clear time information, etc. 4. The format of the to-do list must remain unified after sorting. The to-do date indicated at the beginning of the sentence in the form of #to-do date# must remain at the beginning of the sentence. If there is no clear date, it is prohibited to rigidly generate the "#date time#" format at the beginning of the sentence.

[0140] {Candidate Tasks} <Annotation Task> Please integrate each to-do item into a more complete, clear, and fluent natural sentence. Please arrange the integrated to-do items in ascending order according to the corresponding original text numbers and output them in the form of a list ["To-do 1", "To-do 2", "To-do 3"].

[0141] Next, the annotation task, the first transcription task, the second transcription task, and the description text are substituted into the generated prompt text, and the large language model generates the to-do tasks. The generated prompt text example is as follows: You are asked to come up with some new subtasks given a list of existing subtasks and another existing list of sibling tasks, and then generate a set of examples for each new subtask. Each example consists of an instruction, an input, and an output. The following are the requirements: 1. The skills required to perform subtasks belong to the skills required to perform the target task, and the former is a subset of the latter.

[0142] 2. The skills required to perform the same-level tasks are related to the skills required to perform the target task. The former and the latter have some overlap.

[0143] 3. Subtasks and sibling tasks should focus on common areas rather than specific areas.

[0144] 4. New subtasks are supplements to existing subtasks, and adding new subtasks is crucial to completing the target task.

[0145] 5. New subtasks should be distinct from existing subtasks and sibling tasks. The skills required for designing new subtasks should avoid overlapping with existing subtasks and sibling tasks.

[0146] 6. Instructions should be 1 to 2 sentences long. Imperative or interrogative sentences are permitted.

[0147] 7. Instructions should not contain specific examples and details.

[0148] 8. Try to avoid repeating verbs in each directive in your examples to maximize variety.

[0149] 9. Instructions should be capable of being completed by the GPT language model. For example, instructions should not require the assistant to create any visual or audio output. Another example is not asking the assistant to wake you up at 5 PM or set a reminder, as it cannot perform any actions.

[0150] 10. Create appropriate input based on the instructions in the example, but the input should not be responsive to the instructions. The input should involve real-world data and should not contain simple placeholders. The input should provide substantive content to make the instructions challenging, but should generally not exceed 200 words.

[0151] 11. Please note that some instructions do not require input. For example, when the instruction asks for some general information about the self-contained content, such as "What is the highest mountain in the world?" or "Please list 5 different fruits.", no specific context is required. In this case, we simply place "<no input> ".

[0152] 12. Generate appropriate output based on the input in the instructions and examples. Ensure that the output does not exceed 200 words.

[0153] Target task: [Meeting To-Do] Example of target tasks: ### instruction: enter: Output: ### Subtask Example ### instruction: enter: Output: reason: ### The existing subtask list of the target task is: [].

[0154] The list of existing tasks at the same level as the target task is: [].

[0155] The target task should be broken down into a total of 10 diverse and complementary subtasks, and there are currently 0 existing subtasks. Generate 3 new subtasks and their corresponding reasons, and then list 10 examples of the new subtasks.

[0156] The to-be-done task generating device provided by the present invention is described below. The to-be-done task generating device described below and the to-be-done task generating method described above can be referenced to each other.

[0157] Based on any of the above embodiments, Figure 10 This is a schematic diagram of the structure of the to-do task generation device provided by the present invention. Figure 10 As shown, the device includes: An acquiring unit 1010 is configured to acquire a plurality of labeling tasks and a description text of each labeling task; The transcription unit 1020 is used to perform machine language transcription on each annotation task based on the description text of each annotation task to obtain a transcription task corresponding to each annotation task; A determining unit 1030 is configured to determine a task thought chain based on the description text of each labeled task; The generating unit 1040 is used to integrate the labeling tasks and the transcription tasks based on the task thinking chain to obtain a plurality of to-do tasks.

[0158] Figure 11 Schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 11 As shown, the electronic device may include: a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1140. The processor 1110, the communication interface 1120, and the memory 1130 communicate with each other via the communication bus 1140. The processor 1110 may call logic instructions in the memory 1130 to execute the to-do task generation method.

[0159] Furthermore, the logic instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0160] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the to-do task generation method provided by the above methods.

[0161] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the to-do task generation method provided by the above methods.

[0162] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0163] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for generating to-do tasks, characterized in that: include: Get multiple annotation tasks and the description text of each annotation task; Based on the description text of each annotation task, each annotation task is transcribed into machine language to obtain the corresponding transcription task of each annotation task; Determine the task thinking chain based on the description text of each labeled task; Based on the task thinking chain, each labeling task and each transcription task are integrated to obtain a plurality of to-do tasks.

2. The method for generating pending tasks according to claim 1, wherein: The process of performing machine language transcription on each annotation task based on the description text of each annotation task to obtain a transcription task corresponding to each annotation task includes: Based on the large language model, the description text of each annotation task is applied to polish each annotation task to obtain the first transcription task corresponding to each annotation task; and / or, Based on a large language model, multiple candidate tasks are extracted from the description text of each annotation task, and information completion and logical integration are performed on each annotation task based on the multiple candidate tasks to obtain a second transcription task corresponding to each annotation task.

3. The method for generating pending tasks according to claim 2, wherein: The step of extracting multiple candidate tasks from the description text of each labeling task includes: Determine the task extraction requirements of each description text based on its position in the original text; Based on the task extraction requirements of each description text, multiple candidate tasks are extracted from the description text of each annotation task.

4. The method for generating pending tasks according to claim 3, wherein: The task extraction requirements based on each description text are used to extract multiple candidate tasks from the description text of each labeled task, including: Based on the task extraction requirements of each description text, construct the prompt text of each description text; Based on the hint text of each description text, multiple candidate tasks are extracted from the description text of each annotation task.

5. The method for generating to-do tasks according to any one of claims 1 to 4, characterized in that: The description text of each annotation task is obtained based on the following steps: Obtain the initial description text corresponding to each annotation task from the original text; When the initial description text of any labeling task contains description sentences of other labeling tasks, obtaining the remaining description sentences of the other labeling tasks from the original text; Adding the remaining description sentences to the initial description text of any one of the annotation tasks to obtain candidate description text; A description text of any one of the labeling tasks is determined from a plurality of target description texts, where the target description text is a candidate description text containing a description sentence of any one of the labeling tasks.

6. The method for generating to-do tasks according to claim 5, characterized in that: Determining the description text of any one of the labeling tasks from a plurality of target description texts includes: Based on the position of each target description text in the original text, the position of the description sentence describing any labeling task in each target description text in the original text, and the number of labeling tasks corresponding to each target description text, the description text of any labeling task is determined from the multiple target description texts.

7. The method for generating pending tasks according to claim 5, wherein: The step of obtaining the initial description text corresponding to each annotation task from the original text includes: Obtaining the annotated sentences in the original text for each annotation task; Determining context sentences for each annotation task from the original text based on the position of each annotation sentence in the original text; The context sentence of each annotation task and the annotation sentence of each annotation task are used as the initial description text of each annotation task.

8. A device for generating pending tasks, characterized in that: include: An acquisition unit, used to acquire multiple labeling tasks and description text of each labeling task; A transcription unit is used to perform machine language transcription on each annotation task based on the description text of each annotation task to obtain a transcription task corresponding to each annotation task; The determination unit is used to determine the task thinking chain based on the description text of each annotation task; The generating unit is used to integrate the labeling tasks and the transcription tasks based on the task thinking chain to obtain a plurality of to-do tasks.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method for generating to-do tasks as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for generating to-do tasks as claimed in any one of claims 1 to 7 is implemented.

Images