Text generation method, apparatus and system, computing device, and readable storage medium

Abstract

Provided are a text generation method, apparatus and system, a computing device, and a readable storage medium. The method comprises: acquiring target data to be analyzed, and on the basis of the target data, generating a text generation template on which a text needs to be based; on the basis of each piece of configuration information in the text generation template, determining a plurality of tasks to be executed for the target data; performing topological sorting on the plurality of tasks on the basis of a dependency relationship between the plurality of tasks, and determining, from among the plurality of tasks, target tasks supporting parallel execution; and executing the target tasks in parallel for the target data, and integrating the execution results of the target tasks to generate a text conforming to the text generation template. The method can improve the text generation efficiency.

Description

Text generation method, device and system, computing device, readable storage medium

[0001] The present application claims priority from the Chinese patent application No. 202410625183.X entitled "Text generation method, device and system, computing device, readable storage medium" filed on May 20, 2024 with the China Patent Office, the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0002] Embodiments of the present specification relate to the field of computer technology, in particular to a text generation method. Meanwhile, the present specification also relates to a text generation device, a text generation system, a computing device, a computer readable storage medium, and a computer program product. BACKGROUND

[0003] With the development of computer technology, many various tasks in people's life and work can be executed by computers.

[0004] For example, some basic data required for writing can be analyzed by computers to generate a text that meets the writing requirements. Usually, a corresponding text generation template is set for the text to be generated, which can include multiple pieces of configuration information indicating the theme of the text to be generated and the content to be included, etc. The computer can analyze the basic data based on the text generation template to generate a text that meets the text generation template.

[0005] However, the current text generation efficiency still needs to be improved. SUMMARY

[0006] Embodiments of the present specification provide a text generation method, which can improve the text generation efficiency. One or more embodiments of the present specification also relate to a text generation device, a text generation system, a computing device, a computer readable storage medium, and a computer program product.

[0007] According to an aspect of embodiments of the present specification, a text generation method is provided, the method comprising:

[0008] obtaining target data to be analyzed, and a text generation template required for generating a text based on the target data;

[0009] determining, based on each piece of configuration information in the text generation template, multiple tasks to be executed for the target data;

[0010] topologically sorting the multiple tasks based on a dependency relationship of the multiple tasks, and determining target tasks supporting parallel execution in the multiple tasks;

[0011] perform the target tasks in parallel on the target data, and integrate execution results of the target tasks to generate the text conforming to the text generation template.

[0012] According to another aspect of the embodiments of the present disclosure, a text generation apparatus is provided, comprising:

[0013] an obtaining module configured to obtain target data to be analyzed, and a text generation template to be followed to generate a text based on the target data;

[0014] a first determining module configured to determine, based on each piece of configuration information in the text generation template, a plurality of tasks to be performed on the target data;

[0015] a second determining module configured to perform topological sorting on the plurality of tasks based on a dependency relationship of the plurality of tasks, and determine target tasks in the plurality of tasks that support parallel execution;

[0016] a text generation module configured to perform the target tasks in parallel on the target data, and integrate execution results of the target tasks to generate the text conforming to the text generation template.

[0017] According to still another aspect of the embodiments of the present disclosure, a text generation system is provided, comprising a terminal and a server;

[0018] the terminal is configured to send, to the server, a text generation instruction for target data and a text generation template to be followed to generate a text based on the target data;

[0019] the server is configured to perform the above method in response to the text generation instruction, generate the text conforming to the text generation template, and send the generated text to the terminal.

[0020] According to still another aspect of the embodiments of the present disclosure, a computing device is provided, comprising a memory and a processor;

[0021] the memory is configured to store computer programs / instructions, and the processor is configured to execute the computer programs / instructions, wherein the programs / instructions, when executed by the processor, implement the steps of the above method.

[0022] According to still another aspect of the embodiments of the present disclosure, a computer readable storage medium is provided, which stores computer programs / instructions, wherein the computer programs / instructions, when executed by a processor, implement the steps of the above method.

[0023] According to still another aspect of the embodiments of the present disclosure, a computer program product is provided, comprising computer programs / instructions, wherein the computer programs / instructions, when executed by a processor, implement the steps of the above method.

[0024] In one embodiment of the present specification, target data to be analyzed and a text generation template required for generating a text can be acquired, and in the process of generating a text based on the target data to be analyzed according to the text generation template, a plurality of tasks to be executed on the target data can be determined based on each piece of configuration information in the text generation template, and the tasks to be executed are topologically sorted based on their dependency relationship to determine target tasks supporting parallel execution and perform the target tasks in parallel, and then the execution results of the tasks are integrated to generate a text conforming to the text generation template. In this way, the process of generating a patterned text (such as paradigm writing) can be more flexible, and by reasonably performing the target tasks in the text generation process in parallel, the text generation efficiency can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0025] FIG. 1 is a simple flowchart of generating a text based on a template according to the related art;

[0026] FIG. 2 is a structural schematic diagram of a text generation system according to an embodiment of the present specification;

[0027] FIG. 3 is a flowchart of a text generation method according to an embodiment of the present specification;

[0028] FIG. 4 is a schematic diagram of a result of topologically sorting a plurality of tasks according to an embodiment of the present specification;

[0029] FIG. 5 is a schematic diagram of a result of topologically sorting a plurality of subtasks in a target task according to an embodiment of the present specification;

[0030] FIG. 6 is a structural schematic diagram of a text generation apparatus according to an embodiment of the present specification;

[0031] FIG. 7 is a structural block diagram of a computing device according to an embodiment of the present specification. DETAILED DESCRIPTION

[0032] In the following description, a lot of specific details are set forth in order to facilitate a thorough understanding of the present specification. However, the present specification can be practiced in many different ways beyond the specific embodiments described herein, and it is understood that similar modifications can be made by one skilled in the art without departing from the spirit of the present specification, and therefore the present specification is not limited to the specific implementations disclosed below.

[0033] The terminology used in this disclosure of one or more embodiments is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure of one or more embodiments. As used in this disclosure and the appended claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items. The term "at least one" as used in this disclosure refers to "one or more" and "multiple" refers to "two or more." The term "comprising" is an open term, which is intended to mean "including but not limited to" and thus specifies the presence of stated features, integers, steps, operations, elements, or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or groups thereof.

[0034] It will be understood that, although the terms "first," "second," etc. can be used herein to describe various information, these terms are not intended to denote a particular order or priority among the information. These terms are used to distinguish one from another. For example, a first can be termed a second, and, similarly, a second can be termed a first, without departing from the scope of this disclosure of one or more embodiments. Depending on the context, the word "if' as used herein can be interpreted to mean "when" or "in response to determining".

[0035] In addition, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this disclosure of one or more embodiments are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant standards and requirements, and provide corresponding operation portal for user to choose authorization or rejection.

[0036] In this disclosure of one or more embodiments, a large model refers to a deep learning model with a large number of model parameters, usually containing hundreds of millions, tens of billions, hundreds of billions, thousands of billions or even tens of billions of model parameters. The large model can also be called a foundation model. Through large-scale unlabeled corpus pre-training of the large model, a pre-trained model with hundreds of millions of parameters is output. Such a model can adapt to a wide range of downstream tasks, and the model has good generalization ability, such as large language model (LLM) and multi-modal pre-training model.

[0037] In practical applications, a large model can be applied to different tasks by fine-tuning the pre-trained model with a small amount of samples. The large model can be widely applied in natural language processing (NLP) and computer vision, and can be applied to computer vision tasks such as visual question answering (VQA), image captioning (IC), image generation, and natural language processing tasks such as text-based sentiment classification, text summarization, and machine translation. The main application scenarios of the large model include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, and the like. The model involved in the embodiments of the present specification includes but is not limited to the large model described above, and can be any type of machine learning model. For example, it can be a convolutional neural network (CNN) model, a Transformer model, or the like.

[0038] With the development of computer technology and information digitization, various data is increasing, and the requirement for data efficiency is also increasing. Many tasks originally performed by humans can be performed by computers to reduce labor costs and improve task execution efficiency. For example, in many fields, existing data needs to be analyzed, and the analysis results for the data are written into a text according to a set template. This process can be referred to as paradigm writing. For example, in the financial field, financial institutions need to analyze their financial reports to obtain a review text of the financial reports. The types of content that need to be included in the review text can be indicated by a set template. For example, the template can indicate that the review text needs to include the financial institution to which the review is directed, the authenticity and completeness of the financial report, and the revenue trend of the financial institution.

[0039] Currently, natural language models are increasingly widely applied. A natural language model can be applied to a writing scenario to generate text that meets writing requirements. For example, the natural language model can include a large language model (LLM). Large language models have shown strong capabilities in natural language processing, but still face certain challenges in practical applications. For example, large language models can have hallucination phenomena, their knowledge updates are relatively slow, and the information transparency is low during output of a reply. To address these issues, retrieval-augmented generation (RAG) technology is often used for improvement. RAG refers to retrieving relevant information from an external knowledge base before using an LLM to answer a question, so that the LLM answers the question in combination with the relevant information and data that the LLM would have analyzed to answer the question originally. Using RAG can significantly improve the accuracy of answers output by an LLM, reduce hallucination phenomena of the LLM, and particularly for knowledge-intensive tasks, the effect is more significant. An LLM can use RAG to display the source of the output answer, so as to facilitate verification of the accuracy of the answer by a user and increase trust in the output of the model. Using RAG also promotes knowledge updates and introduction of specific domain knowledge, so that an LLM can effectively combine its parameterized knowledge and non-parameterized external knowledge base, and ensure the reliability of the LLM.

[0040] A paradigm writing can also be implemented using an LLM based on RAG. For example, for target data that needs to be analyzed for paradigm writing, an LLM can be used to query an external data source to obtain relevant information of the target data, and then the LLM answers a corresponding question (query) or generates text. In this way, it is ensured that the answer output by the LLM is based on the retrieved information, which can significantly improve the accuracy and relevance of the writing content. In addition, the output of the LLM can be converted into text output that meets the requirements of the paradigm by using a pre-defined template to indicate a text structure and generation rules.

[0041] FIG. 1 is a simple flowchart of generating text based on a template according to related technologies. As shown in FIG. 1, a text generation device can first obtain a template to be followed for generating text, and obtain natural language text generated based on configuration information in the template. For example, the text to be generated is a financial report review text, and the template is a template for writing a financial report review. Each line of configuration information in the template can correspond to content that needs to be included in the generated text. The text generation device can sequentially rewrite query statements (queries) for each line of configuration information in the template, and map each line of configuration information to a series of query statements. The query statement is a retrieval condition corresponding to a natural language model.

[0042] The text generation apparatus can perform information recall from a material library based on the obtained query sentence to obtain a plurality of information associated with the query sentence. The information recall can be performed by a vector recall method or a BM25 (Best Matching 25) algorithm. The material library can include the financial report to be commented and historical data related to the financial report, such as annual reports and announcements of the same company as the financial report. The text generation apparatus can rank the recalled results, such as ranking according to the matching degree with the query sentence, to obtain a plurality of ranked materials.

[0043] After obtaining the template, the text generation apparatus can further trigger a calling instruction of the large language model to call the large language model to generate text based on the plurality of ranked materials after obtaining the plurality of ranked materials, to obtain a piece of comment text corresponding to each query sentence. The text generation apparatus can further post-process each obtained piece of comment text to obtain a final comment text. For example, the text generation apparatus can check the comment text, or determine whether the comment text meets the requirements and regenerate the comment text if the comment text does not meet the requirements, or perform certain processing to remove possible problems of the large language model itself.

[0044] However, the above process of generating text in the related art presents high linearity and rigidity, resulting in low text generation efficiency, i.e., low paradigm writing efficiency, and lacks certain systematization. The present specification provides a text generation method using a different text generation process from the related art, which can improve the efficiency and quality of text generation. The present specification also relates to a text generation apparatus, a text generation system, a computing device, a computer-readable storage medium, and a computer program product. The text generation apparatus and the computing device can be a server device or a terminal device with strong computing capability.

[0045] FIG. 2 is a structural schematic diagram of a text generation system according to an embodiment of the present specification. As shown in FIG. 2, the text generation system 10 can include a server 101 and a terminal 102, and the terminal 102 can establish a communication connection with the server 101. For example, the server 101 can be a cloud server or a server cluster, and the terminal 102 can be a smartphone, a desktop computer, a notebook computer, a tablet computer, or a smart wearable device.

[0046] The text generation method provided in the specification can be applied to the service end 101. The terminal 102 can interact with the service end 101, such as sending data to the service end 101 or obtaining data from the service end 101. For example, the terminal 102 can send target data to be analyzed and a text generation template to which the generated text needs to conform to the service end 101, and the service end 101 can process the received target data, such as executing the text generation method provided in the embodiments of the specification, to obtain the text conforming to the text generation template, and feed back the generated text to the terminal 102. The terminal 102 can send a text generation request to the service end 101, which can carry target data and a text generation template. The text generation template can exist in the form of a document, or the terminal 102 can directly send the content in the template to the service end 101 in the form of a dialogue.

[0047] In some embodiments, if the computing power of the terminal 102 is strong, the terminal 102 can also not be connected with the service end 101, and the terminal 102 directly executes the text generation method provided in the embodiments of the specification.

[0048] FIG. 3 is a flowchart of a text generation method provided in an embodiment of the specification. In the following, the method is applied to a text generation device, and the text generation device is taken as an example of a service end. As shown in FIG. 3, the method can include the following steps:

[0049] Step 302, obtaining target data to be analyzed, and a text generation template needed to generate a text based on the target data.

[0050] In the embodiments of the specification, the text generation device can analyze certain target data to generate a text needed based on the target data. The target data can be specified by a terminal, and a user can send target data to be analyzed to the text generation device through the terminal. For example, the target data can be a financial report, and the text to be generated based on the target data can be a review text of the financial report. The target data can also be data in other fields, such as related data in the execution process of a project of an enterprise, and the text to be generated can be a summary text of the execution process of the project, etc.

[0051] The text generation device can generate a text based on the target data in accordance with a certain text generation template. The text generation template can be used to specify the content needed to be included in the text to be generated. For example, the text generation template can include the theme of the text, the evaluation index, and the question to be answered, etc. The text generation template can also be specified by the terminal, such as being sent together with the target data by the terminal. Alternatively, the text generation template can also be stored in the text generation device, and the terminal can only indicate the type of the text to be generated, and the text generation device determines the corresponding text generation template based on the type of the text.

[0052] Optionally, the terminal can also not send the target data and the text generation template to the text generation apparatus directly, but send indication information of the target data to the text generation apparatus, and the text generation apparatus can acquire the target data and the text generation template from other data storage devices based on the indication information.

[0053] In step 304, based on each piece of configuration information in the text generation template, a plurality of tasks to be executed for the target data are determined.

[0054] Different pieces of configuration information in the text generation template can correspond to different contents in the text to be generated. For example, each piece of configuration information can occupy a row of the text generation template, which can be a basic unit of the text generation template, and correspond to a basic piece of content in the finally generated text.

[0055] The text generation apparatus can execute at least one task (Tasks) for each piece of configuration information to generate the text corresponding to the configuration information. Accordingly, the corresponding at least one task can be determined based on each piece of configuration information, and the at least one task can all be tasks to be executed for the target data.

[0056] In the embodiments of the present specification, the text generation apparatus can also need to execute tasks unrelated to the text generation template to generate the required text. The text generation apparatus can also determine these tasks as tasks to be executed for the target data. For example, for the generation process of various texts, the text generation apparatus can need to execute certain fixed tasks, such as data formatting tasks, index creation tasks, etc., and the text generation apparatus can also determine these tasks as tasks to be executed for the target data.

[0057] In step 306, based on the dependency relationship of the plurality of tasks, the plurality of tasks are topologically sorted, and target tasks supporting parallel execution in the plurality of tasks are determined.

[0058] Among the tasks to be executed for the target data, there can be a dependency relationship between some tasks and no dependency relationship between some tasks, which is the relationship of execution order. For example, the execution of task A needs to use the execution result of task B, then task A and task B have a dependency relationship, and task A depends on task B.

[0059] After determining the tasks to be executed for the target data, the text generation apparatus can also analyze the dependency relationship of the tasks, and then topologically sort the tasks based on the dependency relationship, and generate a directed acyclic graph (DAG) for the tasks. Based on the result of the topological sorting, the execution process of the tasks can be directly determined, and target tasks supporting parallel execution without a sequence limitation can be determined.

[0060] For example, the task generation apparatus can perform retrieval on the target data after performing a formatting process on the target data, the formatting process task and the retrieval task having a sequential order. The formatting processes on different data can not have a sequential execution order, supporting parallel execution.

[0061] Step 308, performing the target tasks on the target data in parallel, and integrating the execution results of the target tasks to generate a text conforming to the text generation template.

[0062] After the text generation apparatus performs topological sorting on each task to determine the execution order of each task, it can execute each task on the target data according to the order. For tasks with a sequential execution order, they are executed sequentially according to the order, and for target tasks that support parallel execution, they are executed in parallel to improve task execution efficiency. For each task, the task generation apparatus makes the task start execution after the execution of the task it depends on is completed.

[0063] Since the execution results of the parallel target tasks are independent, and a complete text needs to be generated finally, the text generation apparatus can also integrate the execution results of the target tasks according to the relationship between the target tasks and the relationship between the target tasks and the generation of the overall text, to generate a text conforming to the text generation template.

[0064] In some embodiments, the step of obtaining a complete text in the text generation process only depends on a plurality of target tasks executed in parallel, and the execution results of the target tasks are integrated according to the set integration method, so that the final generated text is obtained. In other embodiments, after the plurality of target tasks are executed in parallel, other tasks need to be executed, and then the execution results of the plurality of target tasks are integrated and further processed based on the execution results until the step of generating a complete text is executed to obtain a text conforming to the text generation template. In the text generation process, there can be multiple groups of target tasks, each group of target tasks can include multiple target tasks, the multiple target tasks in the same group can be executed in parallel, and the target tasks in different groups can have a dependency relationship or can also be executed in parallel.

[0065] The text generation method of the embodiments of the present specification is equivalent to splitting the paradigm writing process into multiple independent tasks, and performing parallel processing on the target tasks that support parallel processing according to the dependency relationship between the tasks, which can effectively optimize the processing flow and improve the efficiency of paradigm writing.

[0066] In some embodiments, the parallel execution of the target tasks on the target data in step 308 includes: invoking a plurality of coroutines to parallel execute the target tasks on the target data by using a coroutine pool. The coroutines run on threads, and when a coroutine execution is completed, another coroutine can be selected to run on the current thread. For example, a coroutine can be invoked for each target task, and the thread on which the coroutine runs is used to execute the target task. The use of coroutines can effectively control the concurrent execution of multiple tasks. The embodiments of the present specification can also use the coroutine pool to limit and reuse the coroutines, thereby avoiding unlimited creation of coroutines and resource waste.

[0067] The use of coroutines does not increase the number of threads, but runs multiple coroutines on the basis of threads through time-sharing multiplexing, and the switching of coroutines is completed in the user mode, and the switching cost is much lower than the cost of switching from the user mode to the kernel mode.

[0068] In some embodiments, the maximum number of concurrent tasks can be set, and accordingly, the parallel execution of the target tasks on the target data in step 308 can include: based on the set number of concurrent tasks, controlling the parallel execution of the target tasks less than or equal to the number of concurrent tasks on the target data.

[0069] For example, if the number of target tasks that can be executed in parallel is greater than the set number of concurrent tasks m, the text generation device can first execute m target tasks in parallel, and then execute the remaining target tasks. Alternatively, the text generation device can also divide the target tasks into a plurality of groups, the number of target tasks in each group is relatively small and less than or equal to m, and then execute the plurality of groups of target tasks in sequence, and make the target tasks in each group executed in parallel.

[0070] In the embodiments of the present specification, the natural language model (such as a large language model LLM) can be used to generate text, and accordingly the efficiency of paradigm writing based on LLM can be improved. For example, in step 306, the natural language model can be determined to execute the target tasks in parallel, and in step 308, the natural language model can be invoked to execute one or more tasks on the target data.

[0071] In summary, the text generation method provided by the embodiments of the present specification can obtain target data to be analyzed and a text generation template required for generating a text, and in the process of generating a text based on the target data to be analyzed according to the text generation template, a plurality of tasks to be executed on the target data can be determined based on each piece of configuration information in the text generation template, and the target tasks supporting parallel execution are determined based on the dependency relationship of the tasks to be executed and are executed in parallel, and the execution results of each task are integrated to generate a text conforming to the text generation template. In this way, the process of generating a patterned text (such as paradigm writing) can have higher flexibility, and by reasonably executing the target tasks in the text generation process in parallel, the text generation efficiency can be improved.

[0072] In some embodiments, the step 304 of determining a plurality of tasks to be executed on the target data based on each piece of configuration information in the text generation template comprises:

[0073] determining, based on each piece of configuration information in the text generation template, at least one data query task corresponding to each piece of configuration information, and an integration task of query results of each data query task;

[0074] determining, for each data source in the target data, a data formatting task and at least one index construction task corresponding to each data source;

[0075] obtaining, based on the data query task, the integration task, the data formatting task, and the index construction task, a plurality of tasks to be executed on the target data.

[0076] In the embodiments of the present specification, the tasks to be executed in the text generation process can include a data formatting (Format Data) task and an index construction (Build Index) task on the target data, and can also include a data query task on the target data and an integration (combination) task of query results of each data query task. The integration task can be a task in the post-processing process in the text generation.

[0077] Each piece of configuration information in the text generation template can correspond to at least one data query task, which can be a task of querying required data from the target data and related data (such as from the above-mentioned material library). The data query task can be a task executed by using a natural language model. The text generation device can integrate the execution results of each data query task to generate the required text, and accordingly can determine an integration task dependent on the data query task.

[0078] The data in the target data to be analyzed can be complex, such as data including data from multiple data sources, and the data formats of different data sources can be different. The data from different data sources described in the embodiments of the present specification can correspond to data collected from different devices, or can correspond to data collected from different functional modules of the same device, and the content, type or form of the data from different data sources is different. In order to facilitate analysis (such as query) of the target data, the text generation device can perform format processing on the target data before performing the data query task, so as to convert the target data into data in a unified format. Correspondingly, the process of the format processing can correspond to a data formatting task. The text generation device can perform format processing on different data sources respectively, and correspondingly, each data source can correspond to a data formatting task.

[0079] After the text generation device performs format processing on the data of different data sources, it can also construct an index for the data of each data source, so as to quickly determine the required data for analysis based on the index in the subsequent process. Correspondingly, an index construction task can be set for each data source. Optionally, only one index construction task can be established for one data source; or different types of data can be included in one data source, and different index construction tasks can also be established for different types of data to directly determine the corresponding type of data based on the index.

[0080] The data formatting task, the index construction task, the data query task and the integration task for the data of the same data source have a sequential dependency relationship, so these tasks for the unified data source do not support parallel execution, but these tasks for different data sources can be executed in parallel. As described above, the step 306 of topologically sorting the multiple tasks based on the dependency relationship of the multiple tasks to determine the target tasks supporting parallel execution can include: topologically sorting the multiple tasks based on the dependency relationship of the multiple tasks to determine the data formatting task, the index construction task and the data query task for different data sources, which belong to the target tasks supporting parallel execution.

[0081] FIG. 4 is a schematic diagram of results of topological sorting of a plurality of tasks according to an embodiment of the present specification. As shown in FIG. 4, the target data can include data of two data sources, such as data of one data source being some financial data and data of another data source being some other description information. Data formatting tasks can be performed on the two kinds of data to obtain the financial data in the standard format and the pdf parsing file shown in FIG. 4. Index construction tasks can be performed on the two kinds of data respectively. FIG. 4 takes an example of performing an index construction task on the financial data to obtain the financial data index, performing two different index construction tasks on the block parsing file to obtain the index of the table data and the index of the text data respectively. Alternatively, the text generation apparatus can further obtain other indexes other than the target data, such as the index of external data related to the target data.

[0082] The text generation apparatus can invoke a plurality of data query instances to perform data query tasks based on the indexes of the respective data. FIG. 4 takes an example of the plurality of data query instances corresponding to a financial data query task, a question query task and a feature query task respectively. The data to which different query tasks are directed can be intersected or overlapped. As shown in FIG. 4, the financial data query task can perform data query based on the financial data index and the table data index, the question query task can perform data query based on the text data index and the external data index, and the feature query task can perform data query based on the table data index, the text data index and the external data index. After the plurality of data query tasks, post-processing can be performed, such as integrating the query results of the respective data query tasks.

[0083] For example, as can be known from FIG. 4, the data formatting tasks and the index construction tasks for the two kinds of data can be performed in parallel, and the respective data query tasks can be performed in parallel after the index construction tasks are completed. The task of obtaining the external data index can also be performed in parallel with the data formatting tasks and the index construction tasks.

[0084] In some embodiments, the step 306 of topologically sorting the plurality of tasks based on the dependency relationship of the plurality of tasks to determine the target tasks supporting parallel execution can include: topologically sorting the plurality of tasks based on the dependency relationship of the plurality of tasks to determine that different index construction tasks for the same data source belong to the target tasks supporting parallel execution. For example, as shown in FIG. 4, the two index construction tasks performed on the block parsing file belong to the target tasks supporting parallel execution. After the block parsing file is obtained by performing the data formatting processing, the two index construction tasks can be performed in parallel.

[0085] In the embodiments of the present specification, each task can support mutual nesting. For example, any task in the text generation process can include multiple sub-tasks, which can also be executed in parallel similar to the execution mode of the aforementioned tasks. Taking the target task as an example, the text generation device can perform topological sorting on the multiple sub-tasks included in the target task based on the dependency relationship of the multiple sub-tasks, and determine target sub-tasks supporting parallel execution among the multiple sub-tasks. Further, the target sub-tasks can be executed in parallel, and the execution results of each target sub-task are integrated to obtain the execution result of the target task.

[0086] For example, the target task is a data query task, which can be a to-be-executed task determined based on the configuration information in the text generation template. The data query task can also be a query task of other set types of data, such as a task of directly setting data that meets certain characteristic conditions. The target task can include a query statement rewriting sub-task, a data recall sub-task, an analysis sub-task, and a post-processing sub-task that need to be executed in sequence.

[0087] The process of executing the target task for the target data in step 308 can include:

[0088] Step s2, for each configuration information in the text generation template, executing a query statement rewriting sub-task to obtain prompt information of the natural language model.

[0089] Each configuration information in the text generation template can indicate some content required to be included in the to-be-generated text. The text generation device can map the configuration information to prompt information required to be based on for generating the text by the natural language model. For example, the mapping is achieved by executing the query statement rewriting sub-task.

[0090] Step s4, executing a recall sub-task of related data of the prompt information for the target data.

[0091] For any prompt information, the text generation device can recall the related data of the prompt information from the material library to obtain multiple materials. The related information can be determined based on the matching degree with the prompt information. For the recall sub-task, reference can be made to the related description of information recall in FIG. 1, which is not repeated here. For example, in the case where the target data includes multiple data sources, the recall sub-task of the related data of the prompt information can be executed in parallel for different data sources.

[0092] Optionally, the text generation device can perform recall of related data based on data indexes, and the recall sub-task can be executed in parallel for different data indexes.

[0093] Step s6, performing an analysis subtask on the plurality of relevant data in parallel using a natural language model to obtain a relevance degree of each relevant data to the prompt information.

[0094] After recalling the plurality of relevant data of the prompt information, the text generation apparatus can analyze each relevant data using a natural language model to generate a text corresponding to the prompt information based on the relevant data. This process can refer to the introduction of generating a review text in FIG. 1. The text generation apparatus can also determine the accuracy of the analysis result (such as the generated text) based on the relevant data to determine the relevance degree of the relevant data to the prompt information. The relevance degree can be used as the score of the relevant data.

[0095] Step s8, performing a post-processing subtask based on each relevant data and its relevance degree to the prompt information to obtain a result text corresponding to the prompt information, and determining the execution result of the data query task based on the result text.

[0096] The text generation apparatus can perform a post-processing subtask, such as integrating the analysis results of each relevant data, or further processing or adjusting the analysis results to obtain a result text corresponding to the prompt information.

[0097] FIG. 5 is a schematic diagram of a result of topologically sorting a plurality of subtasks in a target task according to an embodiment of the present specification, and FIG. 5 illustrates the case where the target task corresponds to the question query task in FIG. 4. As shown in FIG. 5, the text generation apparatus rewrites the configuration information to obtain the corresponding prompt information. Then, the recall subtask can be performed based on the text data index and the external data index respectively to recall the prompt information from the text data and the external data, such as recalling file 1 and file 2 from the text data and recalling file 3 and file 4 from the external data. The analysis subtask can be performed on each file using the large language model LLM service to obtain an analysis result, and each file can be scored based on the analysis result. Then, the post-processing can be performed based on the analysis results of each file to obtain a result text generated by the target task.

[0098] For the target task shown in FIG. 5, the process of performing the recall subtask based on the text data index and the external data index can be performed in parallel, and the process of calling the LLM service to analyze each file recalled can also be performed in parallel.

[0099] In the embodiments of the present specification, the text generation apparatus can set a retry mechanism, and the apparatus can re-execute any task that fails. The above step s8 can include:

[0100] performing a post-processing subtask based on each relevant data and its relevance degree to the prompt information;

[0101] In a case where the association degree between the processing result of the post-processing subtask and the prompt information does not meet the association degree condition, the recall subtask, the analysis subtask, and the post-processing subtask are re-executed.

[0102] In a case where the association degree between the processing result of the post-processing subtask and the prompt information meets the association degree condition, the processing result is determined as the result text corresponding to the prompt information.

[0103] The post-processing subtask can refer to the foregoing description about the post-processing in FIG. 1. For example, the text generation apparatus can integrate the analysis results of the related data with the association degree greater than the threshold value to obtain the processing result of the post-processing subtask. For another example, the text generation apparatus can further adjust the analysis results of the related data with the association degree greater than the threshold value to obtain the processing result of the post-processing subtask.

[0104] The text generation apparatus can further determine whether the processing result of the post-processing subtask meets the text generation requirement corresponding to the prompt information, for example, by determining whether the association degree between the processing result and the prompt information meets the association degree condition (for example, whether the association degree is greater than the threshold value) to determine whether the text generation requirement is met. In a case where the association degree condition is not met, the recall subtask and the analysis subtask can be re-executed based on the prompt information. In a case where the association degree condition is met, the processing result of the post-processing can be determined as the result text corresponding to the prompt information, that is, the execution result of the target task. As shown in FIG. 5, in a case where the post-processing result is invalid, retry can be performed to return to the rewritten process.

[0105] Optionally, the retry mechanism can further set the number of retries and the retry time interval. The text generation apparatus can re-execute a task in a case where the number of retries of the task is less than a set number threshold value, and the time interval from the last execution of the task is greater than a set retry time interval. For example, the recall subtask and the analysis subtask are re-executed.

[0106] In the embodiments of the present specification, the text generation apparatus can invoke a DAG engine to implement the foregoing steps 306 and 308. The DAG engine can be defined with a Runnable interface, a Task Base class, and a DAG class, and the Task Base class and the DAG class correspond to executable entities.

[0107] Runnable interface defines the basic structure of all executable entities. Runnable interface serves as an abstract base class to ensure different types of execution entities follow a uniform behavior pattern. The base task class inherits from Runnable interface and can provide the basic implementation of task execution and additional functions (such as retry mechanism). The DAG class belongs to the core of the DAG engine, and its class extends the Runnable interface, so that it can not only serve as an independent task execution unit, but also as a task collection. The DAG class supports nesting, allowing a DAG to be added as a subtask to a larger DAG. This nesting function provides great flexibility and strong organization ability for building complex and multi-level task execution processes.

[0108] For example, in the text generation process, the text generation apparatus can first define instances of subclasses of the base task class, determine the specific task logic implemented by each subclass instance. It can also create a DAG instance and add each subtask to be executed to the DAG instance to implement DAG nesting. Then, the asynchronous execution method of the DAG class can be called to perform topological sorting of tasks and manage the execution order and concurrency of each subtask in the DAG instance. Through the nesting capability of the DAG, a complex task structure can be constructed to optimize the organization and execution of the text generation process.

[0109] In the embodiments of the present specification, the text generation apparatus can divide the entire paradigm writing process into several base task instances and DAG instances. The base task instances can include data formatting task instances, index creation task instances, and post-processing task instances. The DAG instances can include data query task instances.

[0110] In some embodiments, for the above interfaces and classes, corresponding attributes and methods to be executed can be defined for them, which need to be followed when calling the corresponding instances to execute tasks. For example, the attributes of the Runnable interface can include: identifier (name, used to identify and track the corresponding instance), execution result (output), dependent task list (dependencies), other tasks that depend on this task (dependents), number of unresolved dependencies (dependency_count), and current state (status, including states such as not started, running, completed, and failed).

[0111] The methods corresponding to the Runnable interface can include: initializing the basic attributes, such as can be represented by the function initialize ( ); adding dependent tasks, such as can be represented by the function add_dependencies (tasks) ; checking the task preparation condition, such as can be represented by the function is_ready ( ); and defining the asynchronous execution logic, such as can be represented by the function async_run ( ).

[0112] The attributes of the basic task class can include: the task name, the task input parameter, the task output result, the list of dependent tasks, the list of other tasks dependent on this task, the task retry number (max_retries) and the retry interval (retry_delay), and the current task state. The methods corresponding to the basic task class can include: adding a set of dependent tasks (add_dependencies), checking whether the task is ready for execution (is_ready), asynchronously executing the task (including processing the retry logic and updating the dependent task state), and setting the specific task logic to be implemented by the subclass (execute).

[0113] The attributes of the DAG class can include: the maximum number of concurrent task execution (max_workers), the list of all tasks in the DAG, and the attribute indicating whether the tasks have been topologically sorted (Sorted). The methods corresponding to the DAG class can include: adding a single task (add_task) or a list of tasks (add_tasks) to the DAG, and asynchronously executing all the child tasks in the DAG.

[0114] In the embodiments of the present specification, a new text generation mode is adopted, the generation link of paradigm writing is optimized, the technical gap is filled, and the text generation process is more efficient and controllable. Alternatively, the text generation device can also use other task scheduling algorithms or parallel processing frameworks other than the DAG engine to optimize the text generation process.

[0115] To sum up, in the text generation method provided by the embodiments of the present specification, the target data to be analyzed and the text generation template required for generating the text can be obtained. In the process of generating the text based on the target data to be analyzed according to the text generation template, a plurality of tasks to be executed for the target data can be determined based on each piece of configuration information in the text generation template, and the target tasks supporting parallel execution are determined based on the topological sorting of the dependent relationship of the tasks to be executed and are executed in parallel, and then the execution results of each task are integrated to generate the text conforming to the text generation template. In this way, the process of generating patterned text (such as paradigm writing) can be more flexible, and the text generation efficiency can be improved by reasonably executing the target tasks in the text generation process in parallel.

[0116] Corresponding to the above method embodiments, this specification also provides embodiments of a text generation device, which can be a server. Figure 6 is a schematic diagram of the structure of a text generation device provided in an embodiment of this specification. As shown in Figure 6, the text generation device includes:

[0117] The acquisition module 602 is used to acquire the target data to be analyzed, as well as the text generation template required to generate text based on the target data;

[0118] The first determining module 604 is used to determine multiple tasks to be executed for the target data based on each configuration information in the text generation template.

[0119] The second determining module 606 is used to perform topological sorting on the multiple tasks based on their dependencies, and to determine the target task among the multiple tasks that supports parallel execution.

[0120] The text generation module 608 is used to execute target tasks in parallel on target data, integrate the execution results of each target task, and generate text that conforms to the text generation template.

[0121] Optionally, the first determining module 604 is used for:

[0122] Based on each configuration information in the text generation template, determine at least one data query task corresponding to each configuration information, as well as an integration task for the query results of each data query task;

[0123] For each data source in the target data, determine the data formatting task and at least one index building task corresponding to each data source;

[0124] Based on data query tasks, integration tasks, data formatting tasks, and index building tasks, multiple tasks to be executed for the target data are obtained.

[0125] Optionally, data formatting tasks, index building tasks, data querying tasks, and integration tasks for the same data source have sequential dependencies; the second determining module 606 is used for:

[0126] Based on the dependencies between these multiple tasks, a topological sort is performed on these tasks to determine the data formatting task, index building task, and data query task for different data sources, which are the target tasks that support parallel execution.

[0127] Optionally, the second determining module 606 is used for:

[0128] Based on the dependencies between these multiple tasks, a topological sort is performed on these tasks to determine the different index building tasks for the same data source, which are the target tasks that support parallel execution.

[0129] Optionally, the target task includes a data query task, and the text generation module 608 is configured to:

[0130] For each piece of configuration information in the text generation template, perform a query statement rewriting subtask to obtain prompt information of the natural language model;

[0131] Perform a relevant data recall subtask of the prompt information on the target data;

[0132] Perform an analysis subtask on the plurality of relevant data in parallel using the natural language model to obtain an association degree of each relevant data and the prompt information;

[0133] Perform a post-processing subtask based on each relevant data and the association degree of the relevant data and the prompt information to obtain a result text corresponding to the prompt information, and determine an execution result of the data query task based on the result text.

[0134] Optionally, the text generation module 608 is configured to:

[0135] In a case where the target data includes a plurality of data sources, perform the relevant data recall subtask of the prompt information on different data sources in parallel.

[0136] Optionally, the text generation module 608 is configured to:

[0137] Perform the post-processing subtask based on each relevant data and the association degree of the relevant data and the prompt information;

[0138] In a case where the processing result of the post-processing subtask and the association degree of the prompt information do not meet the association degree condition, re-perform the recall subtask, the analysis subtask, and the post-processing subtask;

[0139] In a case where the processing result of the post-processing subtask and the association degree of the prompt information meet the association degree condition, determine the processing result as the result text corresponding to the prompt information.

[0140] Optionally, the text generation module 608 is configured to:

[0141] For the target data, control target tasks less than or equal to a set number of concurrent tasks to be executed in parallel based on the number of concurrent tasks.

[0142] Optionally, the text generation module 608 is configured to:

[0143] For the target data, call a plurality of coroutines to execute the target task in parallel using a coroutine pool.

[0144] In summary, the text generation apparatus provided by the embodiments of the present specification can acquire target data to be analyzed and a text generation template required for generating a text, determine a plurality of tasks to be executed on the target data based on the configuration information in the text generation template, perform topology sorting on the tasks based on the dependency relationship between the tasks, determine target tasks that can be executed in parallel, and execute the target tasks in parallel, and then integrate the execution results of the tasks to generate a text that conforms to the text generation template. In this way, the process of generating a patterned text (e.g., paradigm writing) can be more flexible, and the efficiency of text generation can be improved by reasonably executing the target tasks in parallel.

[0145] Each of the embodiments in the present specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, the text generation apparatus is basically similar to the text generation method, and thus the description is relatively simple, and the relevant parts can be referred to the description of the text generation method.

[0146] FIG. 7 is a structural block diagram of a computing device according to an embodiment of the present specification. The components of the computing device 700 include, but are not limited to, a memory 710 and a processor 720. The processor 720 is connected to the memory 710 through a bus 730, and a database 750 is used to save data.

[0147] The computing device 700 also includes an access device 740 that enables the computing device 700 to communicate via one or more networks 760. Examples of such networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or combinations of such networks, such as the Internet. The access device 740 can include one or more of any type of network interface (for example, a network interface card (NIC)) such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC).

[0148] In one embodiment of the present specification, the above-mentioned components of the computing device 700 and other components not shown in FIG. 7 can also be connected to each other, for example, through a bus. It should be understood that the computing device structure block diagram shown in FIG. 7 is only for the purpose of example, and is not a limitation on the scope of the present specification. Other components can be added or replaced as needed by those skilled in the art.

[0149] The computing device 700 can be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (for example, a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (for example, a smartphone), a wearable computing device (for example, a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 700 can also be a mobile or stationary server.

[0150] The processor 720 is configured to execute computer program / instructions that implement the steps of the text generation method described above when the computer program / instructions are executed by the processor.

[0151] For the computing device embodiment, since it is basically similar to the text generation method embodiment, the description is relatively simple, and the relevant part can be seen in the part of the text generation method embodiment.

[0152] One embodiment of the present specification also provides a computer readable storage medium storing computer instructions, which, when executed by a processor, implement the steps of the above-mentioned text generation method. The computer instructions include computer program codes, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable storage medium can include any entity or device capable of carrying the computer program codes, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable storage medium does not include electrical carrier signals and telecommunication signals.

[0153] One embodiment of the present specification also provides a computer program product, which includes computer programs / instructions, when the computer programs / instructions are executed in a processor, implement the steps of the above-mentioned text generation method.

[0154] For the computer readable storage medium embodiment and the computer program product embodiment, since they are basically similar to the text generation method embodiment, the description is relatively simple, and the relevant part can be seen in the part of the text generation method embodiment.

[0155] The above describes specific embodiments of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the acts or steps recited in the claims can be performed in a different order than those in the embodiments and still achieve desirable results. In addition, the processes depicted in the figures do not necessarily require the particular order shown or sequential order to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous or possible.

[0156] It is noted that the foregoing describes a number of implementations of various embodiments of the application. Other embodiments, however, are within the scope of the following claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve desirable results. Additionally, some of the processes depicted in the accompanying figures can be performed in an order other than that depicted. In certain embodiments, multitasking and parallel processing can be advantageous. Also, the variations of these embodiments are within the scope of the application. Moreover, those skilled in the art will appreciate that the implementations described herein are merely exemplary in nature, and that these implementations can be embodied in various forms other than the specific embodiments described herein. Additionally, the various operations as described in the embodiments can be performed by specific electrodes or groups of electrodes, or by a combination of electrodes and other components. Accordingly, any and all modifications, variations or equivalent arrangements are also within the scope of the application.

[0157] In the above embodiments, the description of each embodiment focuses on different aspects. The parts not described in detail in one embodiment can be referred to the relevant description of other embodiments.

[0158] The preferred embodiments of the present application disclosed above are only used to clarify the present application. The alternative embodiments do not describe all the details and limit the application to the specific embodiments described. Obviously, according to the content of the embodiments of the present application, many modifications and changes can be made. The present application selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of the present application, so that those skilled in the art can well understand and use the present application.

Claims

1. A text generation method characterized by, The method comprises: obtaining target data to be analyzed, and generating a text generation template required for text generation based on the target data; determining a plurality of tasks to be executed for the target data based on each piece of configuration information in the text generation template; topologically sorting the plurality of tasks based on the dependency relationship of the plurality of tasks, and determining target tasks supporting parallel execution in the plurality of tasks; parallelly executing the target tasks for the target data, and integrating execution results of each target task to generate text conforming to the text generation template.

2. The method of claim 1, wherein, The method comprises: determining at least one data query task corresponding to each piece of configuration information in the text generation template based on each piece of configuration information in the text generation template, and determining an integration task for a query result of each data query task; determining a data formatting task and at least one index construction task corresponding to each data source in the target data for each data source; obtaining the plurality of tasks to be executed for the target data based on the data query task, the integration task, the data formatting task and the index construction task.

3. The method of claim 2, wherein, The data formatting task, the index construction task, the data query task and the integration task for the same data source have a sequential dependency relationship; the method comprises: topologically sorting the plurality of tasks based on the dependency relationship of the plurality of tasks, and determining the data formatting task, the index construction task and the data query task for different data sources as target tasks supporting parallel execution.

4. The method of claim 2, wherein, The method comprises: topologically sorting the plurality of tasks based on the dependency relationship of the plurality of tasks, and determining different index construction tasks for the same data source as target tasks supporting parallel execution.

5. The method according to any one of claims 1 to 4, characterized in that, The target task comprises a data query task, and the method comprises: executing a query statement rewriting subtask for each piece of configuration information in the text generation template to obtain prompt information of a natural language model; executing a recall subtask of related data of the prompt information for the target data; parallelly executing an analysis subtask for the plurality of related data by using the natural language model to obtain an association degree of each related data and the prompt information; executing a post-processing subtask based on each related data and the association degree of the related data and the prompt information to obtain a result text corresponding to the prompt information, and determining an execution result of the data query task based on the result text.

6. The method of claim 5, wherein, The method comprises: in a case where the target data comprises a plurality of data sources, parallelly executing the recall subtask of related data of the prompt information for different data sources.

7. The method of claim 5, wherein, The post-processing subtask is executed based on each related data and the relevance of the prompt information, and a result text corresponding to the prompt information is obtained, including: The post-processing subtask is executed based on each related data and the relevance of the prompt information; In the case where the relevance of the processing result of the post-processing subtask and the prompt information does not meet the relevance condition, the recall subtask, the analysis subtask and the post-processing subtask are re-executed; In the case where the relevance of the processing result of the post-processing subtask and the prompt information meets the relevance condition, the processing result is determined as the result text corresponding to the prompt information.

8. The method according to any one of claims 1 to 4, characterized in that, The target task is executed in parallel for the target data, including: For the target data, based on the set number of concurrent tasks, the target task is controlled to be executed in parallel in an amount less than or equal to the number of concurrent tasks.

9. The method according to any one of claims 1 to 4, characterized in that, The target task is executed in parallel for the target data, including: For the target data, a plurality of coroutines are called to execute the target task in parallel using a coroutine pool.

10. A text generating apparatus characterized by comprising: The text generation device includes: An acquisition module configured to acquire target data to be analyzed and a text generation template required for generating a text based on the target data; A first determination module configured to determine a plurality of tasks to be executed for the target data based on each piece of configuration information in the text generation template; A second determination module configured to topologically sort the plurality of tasks based on a dependency relationship of the plurality of tasks, and determine a target task supporting parallel execution in the plurality of tasks; A text generation module configured to execute the target task in parallel for the target data, integrate execution results of each target task, and generate a text conforming to the text generation template.

11. A text generation system, characterized by, It includes a terminal and a server; The terminal is configured to send a text generation instruction for target data and a text generation template required for generating a text based on the target data to the server; The server is configured to execute the method of any one of claims 1 to 9 in response to the text generation instruction, generate a text conforming to the text generation template, and send the generated text to the terminal.

12. A computing device, comprising: It includes: A memory and a processor; The memory is configured to store computer programs / instructions, and the processor is configured to execute the computer programs / instructions, and the programs / instructions are executed by the processor to implement the method of any one of claims 1 to 9.

13. A computer-readable storage medium, characterized in that, The computer programs / instructions are stored, and the computer programs / instructions are executed by the processor to implement the method of any one of claims 1 to 9.

14. A computer program product, characterised in that, The computer program / instruction, when executed by the processor, implements the method of any one of claims 1 to 9.

Images