RPA data processing method and system based on artificial intelligence

Abstract

The invention relates to the technical field of data operation process automation, in particular to an RPA data processing method and system based on artificial intelligence. The method comprises the steps of obtaining a data operation path from a path database according to a name of a digital service platform and a task type of a to-be-executed task, operating the digital service platform based on the data operation path, and obtaining original data; if the acquisition of the data operation path fails, automatically planning the data operation path through a pre-trained path planning model; meanwhile, the data operation path is stored in a path database; a task target of the task to be executed is analyzed, a data structure is constructed, and the data structure comprises a plurality of variables; preprocessing the original data based on the data structure; and inputting the preprocessed data structure into a to-be-executed task, and performing data analysis. By introducing a path planning model to simulate an interface operation process, intelligent combination of the artificial intelligence model and the RPA is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of data operation process automation, and in particular to an artificial intelligence-based RPA data processing method and system. Background Art

[0002] RPA technology refers to the use of software robots or artificial intelligence to automatically execute repetitive business processes, aiming to improve work efficiency and reduce human errors. RPA automation programs can perform various operations on the digital service platforms and websites, including clicking, input, verification, etc., all of which are controlled by programs. The core of the RPA automation program is process automation, which can automatically perform a series of operations according to the set rules and processes. However, the use of RPA automation programs currently still usually requires manual setting of the execution path. And if the digital service platform or web page is updated, the execution path needs to be manually reset.

[0003] With the rapid development of my country's economy, artificial intelligence and digital technologies are increasingly being used in enterprise management. The rational use of artificial intelligence and digital technologies can promote the timeliness and professionalism of management work, help enterprises save management time, and improve the efficiency of management work. At present, the LLM large language model has been introduced in the application of RPA to improve cognitive automation, so that it can not only perform tasks according to rules, but also process unstructured data and better adapt to the context, realizing a more intelligent, more adaptable and more efficient automation process solution, and expanding the boundaries of RPA automation. However, there is still a problem that the data operation path of RPA needs to be manually preset. Summary of the invention

[0004] The objective of the present invention is to address the problem that when RPA is currently used to automatically execute tasks, the data operation path still relies on human pre-setting, making it difficult to expand and transplant the application scenarios of RPA to other third-party applications. This application introduces an artificial intelligence model to simulate the process of human identification and operation of interface elements, and infers and plans the path of the interface elements that need to be operated to complete the tasks to be executed, which is conducive to increasing the adaptability and scalability of the data management method and system provided by this application.

[0005] In order to achieve the above-mentioned invention object, the present invention provides the following technical solutions: An artificial intelligence-based RPA data processing method, characterized by comprising: Acquire a data operation path from a path database according to the name of the digital service platform and the task type of the task to be executed, and if the data operation path is successfully acquired, operate the digital service platform based on the data operation path to acquire original data; If the data operation path fails to be obtained, the data operation path is automatically planned by a pre-trained path planning model; and the data operation path is stored in the path database; the input of the path planning model includes the task to be executed, and the interface elements of the digital service platform are feature extracted based on the task to be executed, and the data operation path to be operated is output; The task objectives of the task to be executed are analyzed to construct a data structure, wherein the data structure includes a plurality of variables; the original data is preprocessed based on the data structure; and the data structure obtained after the preprocessing is input into the task to be executed to perform data analysis.

[0006] As a preferred technical solution of the present application, a path database is established, and the method for constructing the path database includes: combining or fusing the task type, the program type and the update version of the digital service platform to obtain a combined feature; using the combined feature as a keyword and the data operation path as a corresponding value to construct.

[0007] As a preferred technical solution of the present application, the data operation path includes an interface element path and a data node path. The interface element path is a combination of the names of the interface elements arranged in an interactive order; the data node path is an index combination of the original data arranged in a hierarchical structure of the data.

[0008] As a preferred technical solution of the present application, the input of the path planning model also includes task configuration options, and the task configuration options include the task to be executed.

[0009] As a preferred technical solution of the present application, the automatic planning of the data operation path by the path planning model includes: Receive the task configuration options and extract task target features, receive the interface elements of the last operation to extract context features of path planning; Generate a first picture according to the interface of the digital service platform, identify the interface element currently operated in the first picture based on the task target feature and the context feature, and combine multiple interface elements in the operation order to obtain the interface element path; The original data is obtained through the interface element path, the index in the original data is identified based on the task target characteristics, and multiple indexes are combined according to the hierarchical structure to obtain the data node path; the interface element path and the data node path are combined to obtain the data operation path.

[0010] As a preferred technical solution of the present application, the path planning model is established based on a long short-term memory neural network; the long short-term memory neural network receives the task configuration options, the current interface and the interface elements of the last operation, and outputs the interface elements to be operated in the current interface.

[0011] As a preferred technical solution of the present application, the long short-term memory neural network extracts task target features based on the task configuration options through an input gate, and extracts context features based on the current interface and the interface elements of the last operation; deletes interface element features in the current interface that are irrelevant and / or redundant to the task to be performed through a forget gate; outputs the planned interface elements to be operated through an output gate; and outputs the task target features and the context features at the same time.

[0012] As a preferred technical solution of the present application, the identifying the interface element currently operated in the first picture includes: Identify content display text, asynchronously load data, and scroll content from the first image; The interface elements in the first picture are identified based on the layout features of the interface and the appearance features of the interface elements; the content is displayed as the present, and the interface elements corresponding to the asynchronously loaded data and the scrolling content are deleted to obtain interactive elements.

[0013] As a preferred technical solution of the present application, the extracting of the text features from the appearance features of the interactive element also includes: obtaining function prompt text from a preset icon mapping library according to the icon features of the interactive element, and fusing the function prompt text with the text features of the interactive element to obtain the new text features.

[0014] The present application also provides an artificial intelligence-based RPA data processing system, including: A data acquisition module, used to acquire a data operation path from a path database according to the name of the digital service platform and the task type of the task to be executed, and if the data operation path is successfully acquired, operate the digital service platform based on the data operation path and output original data; A path planning module, when the data operation path fails to be obtained, is used to automatically plan the data operation path through a pre-trained path planning model; and store the data operation path in the path database; the input of the path planning model includes tasks to be executed, and based on the tasks to be executed, features of the interface elements of the digital service platform are extracted, and the data operation path to be operated is output; The data processing module is used to analyze the task objectives of the task to be executed and construct a data structure, wherein the data structure includes multiple variables; pre-process the original data based on the data structure; input the data structure obtained after pre-processing into the task to be executed for data analysis.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The introduction of artificial intelligence models simulates the process of human recognition and operation of interface elements, and infers and plans the paths of interface elements that need to be operated to complete the tasks to be performed, which is conducive to increasing the adaptability and scalability of the data management method and system provided by this application. The combination of artificial intelligence models and RPA realizes a more intelligent, adaptable and efficient automation process solution, and expands the boundaries of RPA automation. The method of introducing pre-trained models not only increases the ability of cognitive automation for RPA, but also enables intelligent decision-making, processing of unstructured data, and better adaptation to context.

[0016] 2. In order to solve the problem that different functions in the digital service platform correspond to different data operation paths, the task type and task theme of the task to be executed are preset, and the path planning model extracts the task target features, which can be used for feature comparison with the functions in the digital service platform to achieve the correspondence between the task to be executed and the data operation path, providing parameter optimization direction for the path planning model. By considering the contextual features formed by the previous operation, the system can more intelligently predict and plan the current operation, improve the relevance and accuracy of path planning, and reduce path recognition errors.

[0017] 3. There are interface elements in the interface that are only used for data display, which interfere with the identification of interactive interface elements and also increase the amount of calculation for path planning. To address these issues, by identifying content display text, asynchronously loaded data, and scrolling content, and removing the corresponding interface elements, the interference data input into the path planning model is reduced, which is beneficial to improving the accuracy of the path planning model in identifying interactive elements based on the tasks to be performed, so that the system can understand the user's intentions more accurately and perform tasks more efficiently. The improvement in the accuracy of identifying interactive elements further improves the accuracy of the data operation path, making the data management method provided in this application applicable to a variety of interface layouts, with strong versatility and adaptability.

[0018] 4. Interactive elements only include icon information, which makes it difficult to directly extract the corresponding functional information. At the same time, different interface design styles lead to changes in the icon information of the same function, which further increases the difficulty of identification. To address these problems, an icon mapping library is built based on the icon features of interactive elements and the corresponding functional prompt texts, and the icon information is directly mapped to the preset functional prompt text, which is conducive to the unified expression of icon information, and at the same time, it is conducive to enhancing the functional description of interactive elements including icon information and text information, and improving the accuracy of data operation paths.

[0019] 5. The interface and functions of the digital service platform have a hierarchical structure, which makes the data operation path have contextual relationships. In order to capture this operational contextual relationship, it is preferred to build a path planning model based on the LSTM long short-term memory neural network. By retaining the contextual features of different interfaces, the operated path features are formed, so that the selection range of subsequent interface elements continues to narrow, which is conducive to improving the accuracy of path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flowchart of a data processing method; Figure 2 A flowchart for planning data operation paths; Figure 3 A structural diagram of the data processing system. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them.

[0022] Therefore, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the invention claimed for protection, but merely represents some embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features and technical solutions in the embodiments may be combined with each other.

[0024] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0025] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the invention product is usually placed when in use, or the orientation or positional relationship commonly understood by those skilled in the art. Such terms are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0026] Example 1: See Figure 1-Figure 2 As shown, this embodiment provides an artificial intelligence-based RPA data processing method, including: Use the RPA automation program to obtain the data operation path from the path database according to the name of the digital service platform and the task type of the task to be executed. If the data operation path is successfully obtained, operate the digital service platform based on the data operation path to obtain the original data.

[0027] RPA automation program is a software-based technology that can simulate human operations on computers to automate repetitive and tedious work. For example, RPA automation program can perform the operations of the digital service platform, including clicking, inputting and data crawling, etc. These operations are all controlled by RPA program. The core of RPA automation program is process automation, which can automatically perform a series of operations according to set rules and processes.

[0028] The digital service platform includes the digital service platform in the personal PC or mobile terminal that provides the task to be performed, or a webpage program provided by the website. The task type is set by the user according to the goal of the task to be performed, and can be selected through preset options, or the user enters the task keyword. The RPA program automatically implements the operation process of the interface element through the data operation path to obtain the original data required for the task to be performed.

[0029] The data operation path includes an interface element path and a data node path. Before obtaining the original data, there may be a situation where intermediate data is obtained, such as database classification and functional classification. Therefore, when the user directly uses the digital service platform to obtain and view the original data, it is usually necessary to perform at least one interactive operation of the interface, such as selection and / or click. At the same time, the digital service platform can use XML, whose basic data organization method is a tag-text content structure; or use JSON data format, whose basic data organization method is a key-value structure, and obtain the corresponding value by using the tag or key as an index. Therefore, the data operation path includes a name combination and an index combination of the interface elements that need to be clicked, and the index combination includes a tag combination and / or a key combination. The interface element path is obtained by arranging the name combination in the interactive order; the data node path is obtained by arranging the index combination according to the hierarchical structure of the data.

[0030] The digital service platform has a version update cycle. During a version update cycle, its interactive interface remains unchanged, that is, the data operation path also remains unchanged. Therefore, in this embodiment, a path database is also established to save the data operation path. The method for constructing the path database includes: combining or fusing the task type, the program type and the update version of the digital service platform to obtain a combined feature; using the combined feature as a keyword and the data operation path as a corresponding value to construct.

[0031] When obtaining the data operation path, it is first obtained from the path database. If the data operation path fails to be obtained, it means that the task to be executed is executed for the first time, and the user is prompted to set the data operation path in advance. The method for setting the data operation path includes: constructing and training a path planning model, and automatically planning the data operation path through the path planning model; the input of the path planning model includes the task to be executed, and all the interface elements are feature extracted based on the task to be executed, and the output includes the interface elements to be operated.

[0032] At present, there are practices of introducing LLM large language models into RPA. For example, the chatbots provided in the APP of telecommunications operators can obtain the task requirements of users through dialogue with users, and provide options for possible next operations based on the task requirements. After the user selects, the chatbots obtain data and perform tasks according to the preset data operation path. However, this chatbot provides services based on its own APP. Both the APP and the chatbot are developed by the same software service provider, and it is difficult to transplant them to other APPs without intrusion. By combining RPA with artificial intelligence models, RPA has the ability to automatically plan data operation paths based on the tasks to be performed provided by users, so it can be extended to third-party applications, improving the automation application scenarios of RPA.

[0033] Analyze the task objectives of the task to be executed, construct a data structure required for data analysis, and construct a plurality of variables in the data structure; pre-process the original data based on the data structure. For example, obtain an index with the same name as the variable in the data structure, and extract the value corresponding to the index from the original data as the variable value. Since the RPA automatic robot belongs to the third-party digital service platform, there may be differences in the definition of the data analysis node and the interactive service. If the variable name does not correspond to the same index, the corresponding value is obtained as the variable value through the pre-established mapping relationship between the variable name and the index. If the variable value is not filled with the corresponding value, an alarm prompt is required to the user to verify whether the data operation path is correct or whether the original data has defects, and the variable value is set to null and / or the user manually sets the fill value.

[0034] The data structure obtained after preprocessing is input into the task to be executed to perform data statistics, comparative analysis and classification. For example, for project plan data, data dashboards at all levels are established to perform classified data statistics and analysis by project name, plan category, plan name, etc., to form various chart forms.

[0035] The introduction of artificial intelligence models simulates the process of human recognition and operation of interface elements, and infers and plans the paths of the interface elements that need to be operated to complete the tasks to be performed, which is conducive to increasing the adaptability and scalability of the data management method and system provided in this application. The combination of artificial intelligence models and RPA realizes a more intelligent, more adaptable, and more efficient automation process solution, expanding the boundaries of RPA automation. The method of introducing pre-trained models not only adds cognitive automation capabilities to RPA, but also enables intelligent decision-making, processing of unstructured data, and better adaptation to context.

[0036] In actual applications, different digital service platforms have different functions. At the same time, for the same function, different digital service platforms may also have different execution steps and / or execution paths. And the traditional method of locating the interface elements is to obtain the interface elements through the interface element coordinates or labels. The digital service platform may change its interface with the version update. At the same time, different device sizes have different resolutions and different coordinate axis lengths, and different device systems may also use rectangular coordinate systems with different origins. These factors lead to the problem of low adaptability in the method of implementing the task to be executed through a fixed data operation path. If the interface design is changed or a new device size is introduced when the digital service platform is updated, the data operation path needs to be reset, which increases the complexity of the operation process.

[0037] Therefore, as a preferred technical solution provided by this embodiment, the input of the path planning model also includes task configuration options, which include but are not limited to the task type and task subject of the task to be executed, the program type and update version of the digital service platform, and the screen size. The task type is used to indicate the business field to which the task to be executed belongs, including but not limited to data entry, email processing, report generation, and order processing. The task subject is used to indicate the data usage of the task to be executed, and to explain the data analysis target of the task to be executed. For example, for data entry, the core goal may be to improve the execution speed of the business process, improve the accuracy of data processing, and reduce human errors; for email processing, the core goal may be to reduce the time consumption of manual operations. The program type is used to indicate the usage scenario of the digital service platform, that is, the application scenario of the RPA automatic robot, including but not limited to office, mailbox, plan review, and schedule. The update version and the screen size are used to indicate whether the data operation path is re-captured through the path planning model. On the premise that the update version and the screen size remain unchanged, the data operation path is obtained from the path database.

[0038] The automatically planning the data operation path by using the path planning model includes: The task configuration options are received to extract the task target features, and the interface elements of the last operation are received to extract the context features of the path planning. The task configuration options can be provided through interface options, or imported into the digital service platform through configuration files, and features are extracted from text information through word embedding and / or sentence embedding to obtain the vector representation corresponding to the task target features or the context features.

[0039] A first image is generated according to the interface of the digital service platform, the interface element currently operated in the first image is identified based on the task target feature and the context feature, and the interface element path is obtained by combining multiple interface elements in the order of operation. The task target feature is used to guide the optimization direction of the path planning model, and the neuron parameters are adjusted in combination with the loss function so that the difference between the identified interface element and the correctly operated interface element meets the preset error range; the context feature includes the inferred interface hierarchical structure of the interface element, which is used to narrow the functional scope corresponding to the interface element.

[0040] The original data is obtained through the interface element path, the index in the original data is identified based on the task target feature, and multiple indexes are combined according to the hierarchical structure to obtain the data node path. The interface element path and the data node path are combined to obtain the data operation path.

[0041] In order to solve the problem that different functions in the digital service platform correspond to different data operation paths, the task type and task theme of the task to be executed are preset, and the path planning model extracts the task target features, which can be used for feature comparison with the functions in the digital service platform to achieve the correspondence between the task to be executed and the data operation path, providing parameter optimization direction for the path planning model. By considering the contextual features formed by the previous operation, the system can more intelligently predict and plan the current operation, improve the relevance and accuracy of path planning, and reduce path recognition errors.

[0042] The interface elements in the digital service platform usually use text information and / or icons to provide function prompts. For the text information, it can be recognized by OCR technology, and then converted into a text vector using natural language processing technologies such as bag-of-words model and / or word embedding model. For icon information, it can be extracted by the outline features and color features of the icon. The outline features are obtained using the boundary feature method or edge detection algorithm, and the color features are obtained by extracting the RGB value of the image corresponding to the position of the outline features.

[0043] Further, the identifying the interface element currently operated in the first picture includes: Identify content display text from the first image. Identify all text information from the first image through OCR technology to obtain multiple independent texts. Text recognition APIs existing in the prior art, such as Baidu Text Recognition API, can identify text paragraphs based on the blank distance between texts or whether there are blank lines. Count the number of texts in the independent text, and mark the text paragraphs whose text number is greater than a preset text number threshold as information displayed to the user as content display text. The content display text is usually not used for interactive interaction.

[0044] Identify asynchronously loaded data from the first image. The asynchronously loaded data is data loaded through concurrent threads, and is used to reduce the blocking time of interface loading, including network data, data with too large local bytes, such as ultra-high-definition images, large EXCEL tables, etc. The identification of the asynchronously loaded data includes: presetting an asynchronous loading time, marking data exceeding the asynchronous loading time as the asynchronously loaded data; or obtaining the first sub-image when the interface control loading is completed and the second sub-image when the interface data loading is completed, and marking the content that does not exist in the first sub-image but exists in the second image as the asynchronously loaded data. The asynchronously loaded data is usually not used for interactive interaction.

[0045] Identify scrolling content from the first image, the scrolling content is used to display data or images in a loop. The identifying scrolling content includes: obtaining multiple first images with the same time interval, comparing and finding the difference between the multiple first images, and marking the difference as the scrolling content. The scrolling content is usually not used for interactive interaction.

[0046] The interface elements in the first picture are identified based on the layout features of the interface and the appearance features of the interface elements; the content is displayed as present, and the interface elements corresponding to the asynchronously loaded data and the scrolling content are deleted to obtain interactive elements. The layout features include but are not limited to the spacing distance between the interface and the interface elements, the spacing distance between the interface elements, and the background color features of the interface. The appearance features include but are not limited to geometric features and background color features of the interface elements. The text features include the number of texts, the number of consecutive lines of text, the text color, and the text transparency. The icon features include color, outline, and geometric size. In order to avoid interactive interference between different interface elements, at least there will be no overlap between the interface elements. Therefore, the size and position of the interface elements can be determined using the layout features, the geometric features, and the background color features of the interface elements.

[0047] Different types of the interactive elements are distinguished by the text features and the icon features. For example, the icon of the input box has an obvious border and includes prompt text information with low transparency; a button may include function prompt text information and / or icon information to indicate the button function, and its icon is also designed according to its function; a network link has only text information but no icon information, and the amount of text information is greater than the amount of function prompt text information of the button; the transparency of the function prompt text information of the network link and the button is greater than that of the input box.

[0048] The text feature is extracted from the appearance feature of the interactive element, and the cosine similarity is calculated with the task target feature to determine whether the vectors of the text feature and the task target feature are in the same direction, that is, to determine whether the function of the interactive element satisfies the task to be performed. A similarity threshold is preset, and the interactive element whose cosine similarity is less than the similarity threshold is added to the data operation path.

[0049] There are interface elements in the interface that are only used for data display, which interfere with the identification of interactive interface elements and also increase the amount of calculation for path planning. To address these issues, by identifying content display text, asynchronously loaded data, and scrolling content, and removing the corresponding interface elements, the interference data input into the path planning model is reduced, which is beneficial to improving the accuracy of the path planning model in identifying interactive elements based on the tasks to be performed, so that the system can understand the user's intentions more accurately and perform tasks more efficiently. The improvement in the accuracy of identifying interactive elements further improves the accuracy of the data operation path, making the data management method provided in this application applicable to a variety of interface layouts, with strong versatility and adaptability.

[0050] Different digital service platforms have different interface design styles. Therefore, the representation of text information and / or icon information of the same function is also different. Therefore, an icon mapping library is also preset in this embodiment to map the icon information into unified text information. For the interactive element that only includes icon information, it is difficult to directly obtain the corresponding function description. It is necessary to map the icon information into the corresponding function prompt text to realize the function identification of the interactive element. For the interactive element that includes icon information and text information, the icon information can also be converted into a preset function prompt text, which is conducive to enhancing the function description of the interactive element. Further, the extracting the text feature from the appearance feature of the interactive element also includes: obtaining the icon feature of the interactive element, using the icon feature as an index, and obtaining the function prompt text from the icon mapping library. The function prompt text is converted into a corresponding prompt vector through a natural language processing technology such as a bag of words model or a word embedding model, and the prompt vector is fused with the text feature by concatenation or vector addition or vector multiplication to obtain a new text feature. The icon mapping library is constructed based on the icon of the interactive element and the corresponding function prompt text.

[0051] Interactive elements only include icon information, which makes it difficult to directly extract the corresponding functional information. At the same time, different interface design styles lead to changes in the icon information of the same function, which further increases the difficulty of identification. To address these problems, an icon mapping library is constructed based on the icon features of interactive elements and the corresponding functional prompt texts, and the icon information is directly mapped to the preset functional prompt text, which is conducive to the unified expression of icon information, and at the same time, it is conducive to enhancing the functional description of interactive elements including icon information and text information, and improving the accuracy of data operation paths.

[0052] Embodiment 2: Based on Embodiment 1, this embodiment introduces a pre-trained LSTM neural network or convolutional neural network to realize the construction of the path planning model based on the task to be performed by the user, derives and automatically operates the interface elements that need to be operated at each step of the digital service platform, and achieves the goal of the task to be performed.

[0053] In this embodiment, an implementation of the path planning model can be established based on a convolutional neural network, and the structure of the convolutional neural network includes: 1 convolution layer, 1 maximum pooling layer, 1 residual network layer, 1 activation function layer, 1 regularization layer and 1 fully connected layer. After obtaining pictures of all interfaces of the digital service platform, the convolutional neural network receives the task configuration options and pictures, and then performs data compression through the convolution layer and the maximum pooling layer, and performs feature extraction by the residual network layer and sets weights for different features. The generalization ability of the model is improved by the activation function layer and the regularization layer, and finally the data operation path is output by the fully connected layer. The residual network layer includes 2 parallel feature extraction layers. Each of the feature extraction layers includes 1 convolution kernel, 1 residual layer and 1 attention mechanism layer. The feature extraction layer extracts features for the task to be executed and the digital service platform respectively, and sets corresponding weights through the attention mechanism layer, so that the path planning model pays more attention to the features of the task to be executed.

[0054] In this embodiment, preferably, the path planning model is established based on the LSTM long short-term memory neural network, and the interface elements that need to be operated are gradually inferred through the slow execution characteristics of the LSTM, simulating the process of user manual operation, which is conducive to capturing the contextual dependencies in the data operation path. The core components of LSTM include input gate, forget gate and output gate.

[0055] The long short-term memory neural network receives the task configuration options, the current interface, and the interface elements of the last operation, and extracts the task target features through the input gate based on the task to be performed in the task configuration options. The operation interface of the digital service platform has a hierarchical structure, and different hierarchical structures have different functional divisions. For example, in the financial integration platform, the tax declaration interface can include personal income tax and value-added tax, which correspond to different specific detail interfaces. In the detail interface, the probability of clicking the interface element "declaration" is high. Therefore, the interface elements of the last operation have a certain correlation with the interface that needs to be operated in the current interface. The path planning model extracts this correlation through the input gate to form the context feature. The task target feature is conducive to optimizing the recognition direction of the path planning model, and the context features provided by the interface elements are used to further narrow the selection range of the interface elements in the current interface and improve the accuracy of the data operation path.

[0056] The path planning model deletes interface element features in the current interface that are irrelevant and / or redundant to the task to be performed through a forget gate. For example, if the task to be performed is personal income tax declaration, then the interface element of value-added tax in the tax declaration interface is an irrelevant feature and can be deleted through a forget gate to reduce the interference of interface elements irrelevant to the task to be performed and improve the accuracy of the data operation path.

[0057] The path planning model outputs the interface elements to be operated planned in the current interface through the output gate; at the same time, it also continues to output the task target features and the context features as inputs to subsequent neural units. The interfaces and functions of the digital service platform have a hierarchical structure, so that the data operation path has a contextual relationship. In order to capture this operational contextual relationship, it is preferred to construct a path planning model based on the LSTM long short-term memory neural network. By retaining the contextual features of different interfaces, the operated path features are formed, so that the selection range of subsequent interface elements continues to narrow, which is conducive to improving the accuracy of path planning.

[0058] Embodiment 3: This embodiment provides an artificial intelligence-based RPA data processing system, device, and a storage medium for storing the data processing system.

[0059] For example Figure 3 As shown, an artificial intelligence-based RPA data processing system includes: A data acquisition module, used to acquire a data operation path from a path database according to the name of the digital service platform and the task type of the task to be executed, and if the data operation path is successfully acquired, operate the digital service platform based on the data operation path and output original data; A path planning module, when the data operation path fails to be obtained, is used to automatically plan the data operation path through a pre-trained path planning model; and store the data operation path in the path database; the input of the path planning model includes tasks to be executed, and based on the tasks to be executed, features of the interface elements of the digital service platform are extracted, and the data operation path to be operated is output; The data processing module is used to analyze the task objectives of the task to be executed and construct a data structure, wherein the data structure includes multiple variables; pre-process the original data based on the data structure; input the data structure obtained after pre-processing into the task to be executed for data analysis.

[0060] An artificial intelligence-based RPA data processing device, the device comprising: at least one processor capable of processing various data of a digital service platform; at least one memory for storing various data and at least one program acquired and related to a task to be executed; when the at least one program is executed by the at least one processor, the at least one processor implements the artificial intelligence-based RPA data processing method, the processor and the memory can be connected by a bus or other means for storing at least one program; the memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof, when the at least one program is executed by the at least one processor, the at least one processor implements the artificial intelligence-based RPA data processing method.

[0061] A computer storage medium stores a processor-executable program, which is used to implement the artificial intelligence-based RPA data processing method when executed by the processor.

[0062] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0063] The above embodiments are only used to illustrate the present invention and are not intended to limit the technical solutions described in the present invention. Although the present invention has been described in detail with reference to the above embodiments, the present invention is not limited to the above specific implementation methods. Therefore, any modification or equivalent replacement of the present invention; and all technical solutions and improvements thereof that do not depart from the spirit and scope of the invention are included in the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based RPA data processing method, characterized in that: include: Acquire a data operation path from a path database according to the name of the digital service platform and the task type of the task to be executed, and if the data operation path is successfully acquired, operate the digital service platform based on the data operation path to acquire original data; If the data operation path fails to be obtained, the data operation path is automatically planned by a pre-trained path planning model; and the data operation path is stored in the path database; the input of the path planning model includes the task to be executed, and the interface elements of the digital service platform are feature extracted based on the task to be executed, and the data operation path to be operated is output; Analyze the task objectives of the task to be executed and construct a data structure, wherein the data structure includes a plurality of variables; pre-process the original data based on the data structure; The data structure obtained after preprocessing is input into the task to be executed to perform data analysis.

2. The RPA data processing method based on artificial intelligence according to claim 1, characterized in that: A path database is established, and the construction method of the path database includes: combining or fusing the task type, the program type and the update version of the digital service platform to obtain a combined feature; using the combined feature as a keyword and the data operation path as a corresponding value to construct.

3. The RPA data processing method based on artificial intelligence according to claim 1 is characterized in that: The data operation path includes an interface element path and a data node path. The interface element path is a combination of the names of the interface elements arranged in an interactive order; the data node path is a combination of the indexes of the original data arranged in a hierarchical structure of the data.

4. The RPA data processing method based on artificial intelligence according to claim 3 is characterized in that: The input of the path planning model also includes task configuration options, and the task configuration options include the task to be performed.

5. The RPA data processing method based on artificial intelligence according to any one of claim 4, characterized in that: The automatically planning the data operation path by using the path planning model includes: Receive the task configuration options and extract task target features, receive the interface elements of the last operation to extract context features of path planning; Generate a first picture according to the interface of the digital service platform, identify the interface element currently operated in the first picture based on the task target feature and the context feature, and combine multiple interface elements in the operation order to obtain the interface element path; The original data is obtained through the interface element path, the index in the original data is identified based on the task target characteristics, and multiple indexes are combined according to the hierarchical structure to obtain the data node path; the interface element path and the data node path are combined to obtain the data operation path.

6. The RPA data processing method based on artificial intelligence according to claim 4 is characterized in that: The path planning model is established based on a long short-term memory neural network; the long short-term memory neural network receives the task configuration options, the current interface and the interface elements of the last operation, and outputs the interface elements to be operated in the current interface.

7. The RPA data processing method based on artificial intelligence according to claim 6 is characterized in that: The long short-term memory neural network extracts task target features based on the task configuration options through an input gate, and extracts context features based on the current interface and the interface elements of the last operation; Deleting interface element features in the current interface that are irrelevant to the task to be performed and / or redundant through a forget gate; The planned interface elements to be operated are output through the output gate; at the same time, the task target features and the context features are output.

8. The RPA data processing method based on artificial intelligence according to claim 5, characterized in that: The identifying the interface element currently being operated in the first picture includes: Identify content display text, asynchronously load data, and scroll content from the first image; The interface elements in the first picture are identified based on the layout features of the interface and the appearance features of the interface elements; the content is displayed as the present, and the interface elements corresponding to the asynchronously loaded data and the scrolling content are deleted to obtain interactive elements.

9. The RPA data processing method based on artificial intelligence according to claim 8, characterized in that: The extracting the text feature from the appearance feature of the interactive element also includes: obtaining function prompt text from a preset icon mapping library according to the icon feature of the interactive element, and fusing the function prompt text with the text feature of the interactive element to obtain the new text feature.

10. An artificial intelligence-based RPA data processing system, characterized in that: include: A data acquisition module, used to acquire a data operation path from a path database according to the name of the digital service platform and the task type of the task to be executed, and if the data operation path is successfully acquired, operate the digital service platform based on the data operation path and output original data; A path planning module, when the data operation path fails to be obtained, is used to automatically plan the data operation path through a pre-trained path planning model; and store the data operation path in the path database; the input of the path planning model includes tasks to be executed, and based on the tasks to be executed, features of the interface elements of the digital service platform are extracted, and the data operation path to be operated is output; A data processing module, used to analyze the task target of the task to be executed, construct a data structure, wherein the data structure includes a plurality of variables; and pre-process the raw data based on the data structure; The data structure obtained after preprocessing is input into the task to be executed to perform data analysis.

Images