Information extraction methods, devices, equipment, and media based on RPA and AI

Abstract

This invention discloses an information extraction method, apparatus, device, and medium based on RPA and AI. The method includes: S1, determining the type information of the file content to be identified and the target content to be extracted based on a preset identifier; S2, if the type information exists in the file content to be identified, determining the position information of the target content in the file content to be identified based on the preset information extraction model; S3, extracting the target content from the file content to be identified based on the position information; wherein, the preset information extraction model is obtained by training the file content and the type information of the target content to be extracted as training samples to establish the association between the target content and its position in the file content. By adopting the above technical solution, the adaptability of the information extraction model is improved, and the efficiency of information extraction in different fields is increased.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of robot application technology, and more specifically, to an information extraction method, apparatus, device and medium based on RPA and AI. Background Technology

[0002] RPA (Robotic Process Automation) uses specific "robot software" to simulate human operations on a computer and automatically execute process tasks according to rules.

[0003] AI (Artificial Intelligence) is a new technical science that studies and develops theories, methods, technologies, and application systems for simulating, extending, and expanding human intelligence.

[0004] RPA boasts unique advantages: low-code and non-intrusive. Low-code means RPA can be operated without requiring high IT skills; even business personnel without programming knowledge can develop processes. Non-intrusive means RPA can simulate human operations without requiring software systems to expose interfaces. However, traditional RPA has certain limitations: it can only be based on fixed rules, and its application scenarios are restricted. With the continuous development of AI technology, the deep integration of RPA and AI overcomes the limitations of traditional RPA. RPA + AI = Handwork + Headwork, significantly changing the value of labor.

[0005] In the process of RPA (Resource Processing Automation), a large number of documents requiring information extraction and integration are typically encountered. Currently, when extracting information from documents across different domains, pre-trained models are generally used. Before using this model, a small-scale sample annotation is usually performed on the specific information extraction task, and then the model is trained on the labeled data using a sequence labeling task. However, when faced with a completely new information extraction task, if the type of information to be extracted is entirely different from that extracted in the previous training task, the previously trained model cannot be directly applied to the new task because the type information for the new task has not been used for model training. Therefore, data annotation and model training on the new task are required again. This process is very labor-intensive and inefficient in terms of information extraction. Summary of the Invention

[0006] This invention provides an information extraction method, apparatus, device, and medium based on RPA and AI, which improves the efficiency of information extraction in different fields by enhancing the adaptability of the information extraction model.

[0007] In a first aspect, this invention provides an information extraction method based on RPA and AI, including:

[0008] S1. Determine the type information of the file content to be identified and the target content to be extracted based on the preset identifier;

[0009] S2. If the type information exists in the content of the file to be identified, then based on the preset information extraction model, determine the position information of the target content in the content of the file to be identified;

[0010] S3. Extract the target content from the file content to be identified based on the location information;

[0011] The preset information extraction model is obtained by training the file content and the type information of the target content to be extracted as training samples, so as to establish the association between the target content and its position in the file content.

[0012] Optionally, S2 specifically includes:

[0013] S21. If the type information exists in the content of the file to be identified, then based on the preset information extraction model, the file content and the type information are converted into feature vectors containing semantic information and contextual information respectively.

[0014] S22. Determine the location information of the target content in the file content to be identified based on the feature vector.

[0015] Optionally, S22 specifically includes:

[0016] S221. The feature vector is converted into a first feature vector and a second feature vector in different vector spaces using a preset encoding method, wherein the first feature vector corresponds to the file content and the second feature vector corresponds to the type information; the preset encoding method includes linear mapping and convolution operation;

[0017] S222. Perform a product operation on the first feature vector and the second feature vector to determine the position information of the target content in the file content to be identified based on the product result.

[0018] Optionally, the preset information extraction model is trained in the following manner:

[0019] Sequentially obtain the sample file content containing identification information and the type information of the sample target content, wherein the sample target content is the content to be extracted;

[0020] The initial information extraction model is trained using the content of the sample file and the type information of the sample target content;

[0021] During training, the predicted position of the target content of the sample is determined in the content of the sample file. When the loss function between the predicted position and the labeled real position converges, an information extraction model is obtained. This information extraction model associates the target content of the sample with its position information in the content of the sample file.

[0022] Optionally, the initial information extraction model is a pre-trained model in Natural Language Processing (NLP).

[0023] Optionally, the content of the document to be identified is obtained by performing optical character recognition (OCR) processing on the document to be identified.

[0024] Secondly, embodiments of the present invention also provide an information extraction device based on RPA and AI, the device comprising:

[0025] The identification module is configured to: determine the type information of the file content to be identified and the target content to be extracted based on preset identifiers;

[0026] The location information determination module is configured to: if the type information exists in the content of the file to be identified, determine the location information of the target content in the content of the file to be identified based on the preset information extraction model;

[0027] The content extraction module is configured to extract the target content from the content of the file to be identified based on the location information;

[0028] The preset information extraction model is obtained by training the file content and the type information of the target content to be extracted as training samples, so as to establish the association between the target content and its position in the file content.

[0029] Optionally, the location information determination module includes:

[0030] The first conversion unit is configured to, if the type information exists in the file content to be identified, convert the file content and the type information into feature vectors containing semantic information and contextual information respectively based on the preset information extraction model.

[0031] The location determination unit is configured to determine the location information of the target content within the file content to be identified based on the feature vector.

[0032] Optionally, the position determination unit specifically includes:

[0033] The second conversion subunit is configured to convert the feature vector into a first feature vector and a second feature vector in different vector spaces using a preset encoding method, wherein the first feature vector corresponds to the file content and the second feature vector corresponds to the type information; the preset encoding method includes linear mapping and convolution operation;

[0034] The location determination unit is configured to perform a product operation on a first feature vector and a second feature vector to determine the location information of the target content in the file content to be identified based on the product result.

[0035] Optionally, the preset information extraction model is trained in the following manner:

[0036] Sequentially obtain the sample file content containing identification information and the type information of the sample target content, wherein the sample target content is the content to be extracted;

[0037] The initial information extraction model is trained using the content of the sample file and the type information of the sample target content;

[0038] During training, the predicted position of the target content of the sample is determined in the content of the sample file. When the loss function between the predicted position and the labeled real position converges, an information extraction model is obtained. This information extraction model associates the target content of the sample with its position information in the content of the sample file.

[0039] Optionally, the initial information extraction model is a pre-trained model in Natural Language Processing (NLP).

[0040] Optionally, the content of the document to be identified is obtained by performing optical character recognition (OCR) processing on the document to be identified.

[0041] Thirdly, embodiments of the present invention also provide a computing device, comprising:

[0042] Memory containing executable program code;

[0043] A processor coupled to the memory;

[0044] The processor calls the executable program code stored in the memory to execute the information extraction method based on RPA and AI provided in any embodiment of the present invention.

[0045] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the information extraction method based on RPA and AI provided in any embodiment of the present invention.

[0046] The technical solution provided in this invention, during the training process of a preset information extraction model, uses the concatenated information of the file content and the type information of the target content to be extracted as training samples input into the model for training. This allows the information extraction model to establish the association between the target content and its position within the file content. During information extraction, after acquiring the file information to be extracted, the type information of the file content to be identified and the target content to be extracted can be determined based on a preset identifier. If the type information of the target content exists in the file content to be identified, the position information of the target content within the file content to be identified can be determined based on the preset information extraction model, and thus the target content can be extracted from the file content according to this position information. Especially for files from different application domains, the type information of the file content to be identified and the target content to be extracted can be input into the aforementioned preset information extraction model. Using this model, the position information of the target content within the file content to be identified can be determined, thereby enabling the extraction of the target content. Compared to the prior art method of continuously training the model using file content from different domains as samples to extract file content from different domains, the technical solution provided in this invention improves the adaptability of the information extraction model. For different application areas and different information extraction tasks, there is no need to re-label samples or retrain the model. The information extraction task can be completed by inputting the file content and the target content to be extracted into the information extraction model, saving a lot of time and effort and improving the efficiency of information extraction.

[0047] The innovative aspects of this invention include:

[0048] 1. During the training process of the preset information extraction model, by using the information obtained by concatenating the file content and the type information of the target content to be extracted as training samples to input into the model for training, the information extraction model can establish the correlation between the target content and its position in the file content, thereby improving the adaptability of the information extraction model and thus improving the efficiency of information extraction. This is one of the innovations of this invention.

[0049] 2. The feature vectors output by the preset encoding model are converted into first feature vectors and second feature vectors in different vector spaces using a preset encoding method. The first feature vector corresponds to the file content, and the second feature vector corresponds to the type information. This setting can distinguish the type information of the file content and the target content to be extracted, which helps to improve the accuracy of the preset information extraction model in predicting the position of the target content in the file content. This is one of the innovations of this invention. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1a A flowchart illustrating a training method for an information extraction model provided in Embodiment 1 of the present invention;

[0052] Figure 1b This is a schematic diagram of the processing procedure of the information extraction model provided in Embodiment 1 of the present invention;

[0053] Figure 2a This is a flowchart of an information extraction method based on RPA and AI provided in Embodiment 2 of the present invention;

[0054] Figure 2b This is a screenshot of the process for extracting contract content on an automated service platform, as provided in Embodiment 2 of the present invention.

[0055] Figure 3 This is a structural block diagram of an information extraction device based on RPA and AI provided in Embodiment 3 of the present invention;

[0056] Figure 4 This is a schematic diagram of the structure of a computing device provided in Embodiment 4 of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] It should be noted that the terms "comprising" and "having," and any variations thereof, in the embodiments and drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0059] In this invention, the pre-trained model is a model that has already been trained using a dataset. The advantages of a pre-trained model are lower training costs, faster convergence speed when used with downstream tasks, and improved model performance, especially for tasks with scarce training data. In other words, the pre-training method can be considered as allowing the model to learn from a better initial state, thereby achieving better performance.

[0060] In this invention, the loss function is a function that maps the values ​​of a random event or its related random variables to non-negative real numbers to represent the "risk" or "loss" of that random event. In applications, the loss function is often used as a learning criterion in relation to optimization problems; that is, the model is solved and evaluated by minimizing the loss function. In the technical solution provided by this invention, the loss function between the true labeled location and the predicted location of the information extraction model is minimized, making the prediction result of the information extraction model closer to the true labeled result.

[0061] This invention discloses an information extraction method, apparatus, device, and medium based on RPA and AI. These will be described in detail below.

[0062] Example 1

[0063] Robotic Process Automation (RPA) uses specialized "robot software" to simulate human actions on a computer, automatically executing tasks according to rules. RPA robots can intelligently understand existing enterprise applications through a user interface, automating routine, rule-based operations such as automatically repetitively reading emails, retrieving Office components, manipulating databases, web pages, and client software, collecting data, performing tedious calculations, generating large batches of documents and reports, and completing monotonous document checks. This significantly reduces labor costs, effectively improves existing office efficiency, and completes tasks accurately, stably, and quickly.

[0064] AI (Artificial Intelligence) is a new technical science that studies, develops, and applies theories, methods, technologies, and application systems to simulate, extend, and expand human intelligence. In this embodiment of the invention, OCR (Optical Character Recognition) technology from AI is specifically used to recognize the document content, and an encoder model from NLP (Natural Language Processing) is used to encode the type information of the document content and the target content to be extracted, resulting in a feature vector containing semantic information.

[0065] Currently, RPA typically uses pre-trained models for information extraction during document content extraction. When faced with a completely new information extraction task, if the type of content to be extracted is entirely different from that extracted in previous training tasks, data labeling and model training must be performed again on the new task. This process is very labor-intensive and does not utilize the structural similarities between tasks. The information extraction method proposed in this invention trains a universal information extraction model for different tasks. This model automatically extracts the corresponding content based on the type of information to be extracted, achieving adaptability across different tasks and greatly improving the efficiency of model application. Figure 1a This is a flowchart of a training method for an information extraction model provided in Embodiment 1 of the present invention. Figure 1b This is a schematic diagram illustrating the processing steps of the information extraction model provided in Embodiment 1 of the present invention. The following is in conjunction with... Figure 1b right Figure 1a The process involves detailed explanations of each step.

[0066] S110. Sequentially obtain the sample file content containing identification information and the type information of the sample target content.

[0067] The sample files can be documents from different fields, such as contracts, resumes, bidding information, equipment information, and personnel statistics. The target content of the sample files is the content to be extracted from the sample files. The sample file content can be content obtained after OCR recognition processing.

[0068] In this embodiment, the sample file content containing the identification information and the type information of the sample target content are obtained sequentially. This is equivalent to concatenating the two parts of information, namely, if the string length of the sample file content is C and the string length of the type information is Q, then after obtaining these two parts of content sequentially, the length of the sample string is L = C + Q.

[0069] In this embodiment, when the type information of the sample file content and the sample target content is input into the initial information extraction model, these two parts of information can be separated by a preset identifier.

[0070] For example, if the input order of these two parts is fixed, the two parts can be separated using a preset separator. For instance, as shown below... Figure 1b As shown, if the two parts of the input model are pre-defined as the file content to be identified and the type information of the target content to be extracted, then it is only necessary to separate the two parts using the delimiter "SEP".

[0071] For example, when the input order is not fixed, for two parts of information input into the model sequentially, a start identifier can be added at the beginning of the first input to indicate the start of that part, and an end identifier can be added after the end of that part to indicate the end of that part, i.e., the start of the other part. For example, as... Figure 1b As shown, a start identifier can be added before "Party A" in the file content to be identified to indicate the beginning of the file content to be identified, and an identifier "SEP" can be added after the end of the file content to be identified to indicate the beginning of the type information of the target content to be extracted.

[0072] It is important to note that if there are multiple types of the target content to be extracted, a separator should be added before each type, such as... Figure 1b As shown, "SEP" is added before both "Party A" and "Date" to distinguish different types of information. Compared to existing technologies that only use file content for model training, this embodiment uses both the file content and the type information of the target content to be extracted as training samples. That is, the information obtained by concatenating the type information with the sample file content is input into the model for training. This setup leverages the structural similarities of various tasks. For information extraction tasks in different domains, such as resumes or contracts, the type information is generally followed by specific information content, such as: Party A: Company X; Name: Li; Date: September 2020. This configuration improves the model's adaptability, allowing it to be applied to information extraction tasks in different domains.

[0073] S120. Train the initial information extraction model using the type information of the sample file content and the sample target content.

[0074] In this embodiment, the preset initial information extraction model can be a pre-trained model in NLP, preferably BERT (Bidirectional Encoder Representations from Transformers) or XLNET (a general autoregressive pre-trained model).

[0075] During training, pre-trained encoding models can be used to encode the type information of sample file content and sample target content, resulting in computer-recognizable feature vectors containing semantic and contextual information, such as... Figure 1bThe E_C and E_Q shown are used to train the preset initial information extraction model to obtain the predicted position of the target content in the sample file content. For example, the start position prediction S_1 and end position prediction E_1 for the first type "Party A", and the start position prediction S_2 and end position prediction E_2 for the second type "Date".

[0076] To improve the accuracy of information extraction in the information extraction model, a pre-defined encoding method can be used to constrain the feature vectors corresponding to the sample file content and the feature vectors corresponding to the type information to different vector spaces. This results in a first feature vector corresponding to the sample file content and a second feature vector corresponding to the type information. For example... Figure 1b In this model, the content is encoded to obtain the first feature vector W_C, and the type is encoded to obtain the second feature vector W_Q. This configuration better distinguishes between the content of sample files with different semantics and the type information of the target content, thereby helping to improve the accuracy of the information extraction model in predicting the location of the target content. The preset encoding method can be linear mapping or convolution operations, etc., and this implementation does not specify a particular limitation.

[0077] After obtaining the first feature vector corresponding to the sample file content and the second feature vector corresponding to the type information, cross-coding can be used. For example, the first and second feature vectors can be multiplied, and the predicted position of the target content in the sample file content can be predicted based on the product result. For example, Figure 1b In the first type, “Party A”, the start position prediction is S_1 and the end position prediction is E_1, and in the second type, “Date”, the start position prediction is S_2 and the end position prediction is E_2.

[0078] S130. During the training process, the predicted position of the target content of the sample is determined in the sample file content. When the loss function between the predicted position and the labeled real position converges, the information extraction model is obtained.

[0079] The convergence of the loss function indicates that the prediction result of the information extraction model is close to the labeled true location.

[0080] In this embodiment, during the training process of the preset information extraction model, the concatenated content of the file content and the type information of the target content to be extracted is used as a training sample and input into the model for training. This allows the information extraction model to establish the association between the target content and its position within the file content. Compared to the existing technology that uses file content from different domains as samples to continuously train the model in order to extract file content from different domains, the technical solution provided by this embodiment improves the adaptability of the information extraction model. For different application domains and different information extraction tasks, there is no need to re-label samples or retrain the model. By simply inputting the file content and the target content to be extracted into the information extraction model, the information extraction task can be completed, saving a significant amount of time and effort and improving the efficiency of information extraction.

[0081] The following section will provide a detailed introduction to the specific application process of the information extraction model.

[0082] Example 2

[0083] Figure 2a This is a flowchart illustrating an information extraction method based on RPA and AI, provided in Embodiment 2 of the present invention. This method can be applied to application scenarios such as resumes, contracts, bidding information, and equipment information. It can be executed by an RPA and AI-based information extraction device, which can be implemented through software and / or hardware. The method provided in this embodiment can be applied to automated service platforms, such as UiBot Mage, which provides AI capabilities for RPA robots. Figure 2b This is a screenshot of the contract content extraction process built on an automated service platform according to Embodiment 2 of the present invention. Below, in conjunction with... Figure 2b right Figure 2a The information extraction methods based on RPA and AI are described in detail below:

[0084] S210. Determine the type information of the file content to be identified and the target content to be extracted based on the preset identifier.

[0085] The content of the file to be recognized is obtained by performing OCR processing on the file.

[0086] The preset identifier, along with the type information of the file content to be identified and the target content to be extracted, needs to be input into the preset information extraction model so that the two parts of the content can be distinguished based on the preset identifier.

[0087] In this embodiment, an RPA robot can be used to first capture the content of the file to be identified, and then the captured file content and the type information of the target content to be extracted are concatenated. That is, the file content and type information are sequentially input into a preset information extraction model. Specifically, such as... Figure 2b As shown, an RPA robot can be used to capture the contract to be identified, and then the contract can be OCR-recognized. The recognition result and the type information of the target content to be extracted are then concatenated and input into the preset information extraction model.

[0088] S220. If type information exists in the content of the file to be identified, the location information of the target content in the content of the file to be identified is determined based on the preset information extraction model.

[0089] The preset information extraction model is obtained by training the system with the concatenated information of the file content and the type information of the target content to be extracted as training samples, in order to establish the association between the target content and its position in the file content. The training process of this preset information extraction model can be referred to the above embodiment, and will not be repeated here.

[0090] In this embodiment, the preset information extraction model, during the location information determination process, can convert the file content and type information into feature vectors containing semantic and contextual information, respectively. This configuration is to convert the type information of the file content to be identified and the target content to be extracted into language that the computer can recognize. Based on the feature vectors, the location information of the target content within the file content to be identified can be predicted. The process of converting the type information of the input file content and the target content to be extracted into feature vectors can be performed by a pre-trained encoding model, preferably a BERT model or an XLNET model.

[0091] In this embodiment, to improve the accuracy of location prediction by the preset information extraction model, a preset encoding method can be used to convert the feature vectors output by the pre-trained encoding model into first and second feature vectors in different vector spaces. The first feature vector corresponds to the file content, and the second feature vector corresponds to type information. This setup can distinguish the type information of the file content and the target content to be extracted, thereby improving the accuracy of the preset information extraction model in predicting the location of the target content within the file content. The preset encoding method can be linear mapping or convolution operations, etc.

[0092] In this embodiment, when the preset information extraction model determines the location information of the target content in the file content based on the first feature vector and the second feature vector, it can use cross-coding. For example, the first feature vector and the second feature vector can be multiplied, and the location information of the target content in the file content to be identified can be determined based on the product result.

[0093] S230. Extract the target content from the content of the file to be identified based on the location information.

[0094] Furthermore, such as Figure 2bAs shown, after extracting the target content, RPA can enter the target content into a table. If it detects that no target content has been extracted, it returns to the contract retrieval operation.

[0095] In this embodiment, after obtaining the file information to be extracted, the type information of the file content to be identified and the target content to be extracted can be determined according to a preset identifier. If the type information of the target content exists in the file content to be identified, the position information of the target content in the file content to be identified can be determined based on the preset information extraction model, and thus the target content can be extracted from the file content according to the position information. In particular, for files in different application fields, the type information of the file content to be identified and the target content to be extracted can be input into the above-mentioned preset information extraction model. Using this model, the position information of the target content in the file content to be identified can be determined, thereby enabling the extraction of the target content.

[0096] Example 3

[0097] Figure 3 This is a structural block diagram of an information extraction device based on RPA and AI provided in Embodiment 3 of the present invention, as shown below. Figure 3 As shown, the device includes: an identification module 310, a location information determination module 320, and a content extraction module 330; wherein,

[0098] The identification module 310 is configured to: determine the type information of the file content to be identified and the target content to be extracted based on a preset identifier;

[0099] The location information determination module 320 is configured to: if the type information exists in the content of the file to be identified, determine the location information of the target content in the content of the file to be identified based on the preset information extraction model;

[0100] Content extraction module 330 is configured to: extract the target content from the content of the file to be identified based on the location information;

[0101] The preset information extraction model is obtained by training the file content and the type information of the target content to be extracted as training samples, so as to establish the association between the target content and its position in the file content.

[0102] Optionally, the location information determination module includes:

[0103] The first conversion unit is configured to, if the type information exists in the file content to be identified, convert the file content and the type information into feature vectors containing semantic information and contextual information respectively based on the preset information extraction model.

[0104] The location determination unit is configured to determine the location information of the target content within the file content to be identified based on the feature vector.

[0105] Optionally, the position determination unit specifically includes:

[0106] The second conversion subunit is configured to convert the feature vector into a first feature vector and a second feature vector in different vector spaces using a preset encoding method, wherein the first feature vector corresponds to the file content and the second feature vector corresponds to the type information; the preset encoding method includes linear mapping and convolution operation;

[0107] The location determination unit is configured to perform a product operation on a first feature vector and a second feature vector to determine the location information of the target content in the file content to be identified based on the product result.

[0108] Optionally, the preset information extraction model is trained in the following manner:

[0109] Sequentially obtain the sample file content containing identification information and the type information of the sample target content, wherein the sample target content is the content to be extracted;

[0110] The initial information extraction model is trained using the content of the sample file and the type information of the sample target content;

[0111] During training, the predicted position of the target content of the sample is determined in the content of the sample file. When the loss function between the predicted position and the labeled real position converges, an information extraction model is obtained. This information extraction model associates the target content of the sample with its position information in the content of the sample file.

[0112] Optionally, the initial information extraction model is a pre-trained model in Natural Language Processing (NLP).

[0113] Optionally, the content of the document to be identified is obtained by performing optical character recognition (OCR) processing on the document to be identified.

[0114] The information extraction device based on RPA and AI provided in this embodiment of the invention can execute the information extraction method based on RPA and AI provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in the above embodiments can be found in the information extraction method based on RPA and AI provided in any embodiment of the invention.

[0115] Example 4

[0116] Please see Figure 4 , Figure 4This is a schematic diagram of the structure of a computing device provided in Embodiment 4 of the present invention. Figure 4 As shown, the computing device may include:

[0117] Memory 701 storing executable program code;

[0118] Processor 702 coupled to memory 701;

[0119] The processor 702 calls the executable program code stored in the memory 701 to execute the information extraction method based on RPA and AI provided in any embodiment of the present invention.

[0120] This invention discloses a computer-readable storage medium storing a computer program that enables a computer to execute the information extraction method based on RPA and AI provided in any embodiment of this invention.

[0121] In various embodiments of the present invention, it should be understood that the sequence number of each process does not necessarily imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0122] In the embodiments provided by this invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0123] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0124] If the integrated units described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several requests to cause a computer device (which can be a personal computer, server, or network device, specifically a processor in the computer device) to execute some or all of the steps of the methods described in the various embodiments of the present invention.

[0125] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0126] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0127] Those skilled in the art will understand that the modules in the apparatus of the embodiments can be distributed in the apparatus of the embodiments as described in the embodiments, or they can be located in one or more devices different from this embodiment with corresponding changes. The modules of the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An information extraction method based on RPA and AI, characterized in that, include: S1. Determine the type information of the file content to be identified and the target content to be extracted based on the preset identifier; S2. If the type information exists in the content of the file to be identified, then the location information of the target content in the content of the file to be identified is determined based on the preset information extraction model. S3. Extract the target content from the file content to be identified based on the location information; The preset information extraction model is obtained by training the file content and the type information of the target content to be extracted as training samples to establish the relationship between the target content and its position in the file content. Specifically, S2 includes: S21. If the type information exists in the content of the file to be identified, then based on the preset information extraction model, the file content and the type information are converted into feature vectors containing semantic information and contextual information respectively. S22. Determine the position information of the target content in the file content to be identified based on the feature vector; Specifically, S22 includes: S221. The feature vector is converted into a first feature vector and a second feature vector in different vector spaces using a preset encoding method, wherein the first feature vector corresponds to the file content and the second feature vector corresponds to the type information; the preset encoding method includes linear mapping and convolution operation; S222. Perform a product operation on the first feature vector and the second feature vector to determine the position information of the target content in the file content to be identified based on the product result.

2. The method according to claim 1, characterized in that, The preset information extraction model is trained in the following manner: Sequentially obtain the sample file content containing identification information and the type information of the sample target content, wherein the sample target content is the content to be extracted; The initial information extraction model is trained using the content of the sample file and the type information of the sample target content; During training, the predicted position of the target content of the sample is determined in the content of the sample file. When the loss function between the predicted position and the labeled real position converges, an information extraction model is obtained. This information extraction model associates the target content of the sample with its position information in the content of the sample file.

3. The method according to claim 2, characterized in that, The initial information extraction model is a pre-trained model in Natural Language Processing (NLP).

4. The method according to claim 1, characterized in that, The content of the document to be identified is obtained by performing optical character recognition (OCR) processing on the document.

5. An information extraction device based on RPA and AI, characterized in that, include: The identification module is configured to: determine the type information of the file content to be identified and the target content to be extracted based on preset identifiers; The location information determination module is configured to: if the type information exists in the content of the file to be identified, determine the location information of the target content in the content of the file to be identified based on a preset information extraction model; The content extraction module is configured to extract the target content from the content of the file to be identified based on the location information; The preset information extraction model is obtained by training the file content and the type information of the target content to be extracted as training samples to establish the relationship between the target content and its position in the file content. The location information determination module includes: The first conversion unit is configured to convert the file content and the type information into feature vectors containing semantic information based on the preset information extraction model if the type information exists in the file content to be identified. The location information extraction unit is configured to determine the location information of the target content in the file content to be identified based on the first feature vector corresponding to the file content and the second feature vector corresponding to the type information; The location information extraction unit specifically includes: The second conversion subunit is configured to convert the feature vector into a first feature vector and a second feature vector in different vector spaces using a preset encoding method, wherein the first feature vector corresponds to the file content and the second feature vector corresponds to the type information; the preset encoding method includes linear mapping and convolution operation; The location determination subunit is configured to perform a product operation on a first feature vector and a second feature vector to determine the location information of the target content in the file content to be identified based on the product result.

6. A computing device, characterized in that, include: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the information extraction method based on RPA and AI as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the information extraction method based on RPA and AI as described in any one of claims 1-4.

Images