Context disambiguation using deep neural networks to enable sentence rewriting without words of interest

Abstract

A computer-implemented process for updating an electronic document includes the following operations. Preprocessing is performed on an electronic document using a preprocessor to generate a computer data structure. A computer data structure is evaluated using a word sense disambiguation (WSD) engine and a deep neural network to determine a context of a sentence within an electronic document. Based on the context, it is determined that a word within the sentence is a word of interest. The sentence is rewritten using a mitigation engine and a large language model to generate a revised sentence that does not include the word of interest. Determining that the revised sentence does not include any other word of interest; and updating the electronic document to include the revised sentence.

Description

Background Technology

[0001] This invention relates to electronic document processing using natural language processing, and more specifically, to using deep neural networks for context disambiguation when processing electronic documents.

[0002] Natural Language Processing (NLP) is a field of computer science and linguistics that involves the interaction between computers and human (natural) language. As a field of computer science, NLP began as a branch of artificial intelligence. Modern NLP algorithms are based on machine learning (ML) and include statistical methods and neural networks. As used in this paper, an "NLP agent" is a specialized computer system that includes hardware and software utilizing NLP algorithms configured to process electronic documents by performing natural language processing and analyzing natural language data extracted from the documents.

[0003] Detecting keywords within a specific context is a common and often challenging requirement for NLP agents. While human readers may have little difficulty correctly interpreting the meaning of a particular word in a given context, semantic ambiguity can pose problems for many NLP agents. Therefore, techniques commonly known as word sense disambiguation (WSD) have been developed to enhance NLP algorithms. As used herein, the term "meaning" refers to the specific meaning of a word in a given context. For example, the word "bass" could refer to a type of fish or a type of sound / instrument. These different meanings are the different "meanings" of the word.

[0004] Understanding the meaning of words in context can have many benefits during electronic document processing. For example, an electronic document containing certain words in certain contexts (e.g., legal and business) may generate unintended exposure to the entities associated with that document. As another example, in situations where a document undergoes considerable public exposure (e.g., in the social media domain), certain words in certain contexts may attract unwanted attention. There are numerous use cases where identifying risky words in context (i.e., “words of interest”) can be useful in activities such as tagging / reporting, risk reduction, and blocking / censoring profanity.

[0005] In tagging / reporting, keywords of interest can be identified for the creator and / or other users of the electronic document. When the risk decreases, the keywords of interest can be removed or the content altered. In blocking / censorship, the electronic document itself or specific portions of it can be blocked or otherwise made unavailable for consumption by other users.

[0006] An example of a word of interest is the term "partnership." In an informal context, "partnership" can refer to cooperation or shared entertainment between two entities. However, in a business / legal context, the term "partnership" can refer to the existence / establishment of a specific formal business relationship. Therefore, the use of the term "partner" in a business / legal context, as it is used informally, can have undesirable implications.

[0007] An example of labeling is when an IT service provider might want to understand how many of its active contracts relate to cloud service provision and be able to label instances of contracts currently in progress. In this case, the phrase "third party" and privacy practices governing the use of non-vendor cloud services or other services, including content clients that can provide…, and the “cloud content” design slides that will be provided by cloud solution platform service providers and vendors and should include…, would be labeled as associated with cloud service provision. On the other hand, the phrase “depending on the altitude of the troposphere, the year, and the vertical temperature structure, these clouds can consist of…” would not be labeled as associated with cloud service provision, despite including the word “cloud.”

[0008] There are several known methods for performing WSD using NLP. However, many of these methods rely on perceptual annotations used to train the NLP agent, which are essentially curated knowledge bases of sentences and words. However, collecting perceptual annotations to generate a curated knowledge base is a labor-intensive task. Another problem with many machine learning-based NLP agents is the lack of interpretability for specific results. The term "interpretability," as used in the context of machine learning, refers to the ability to interpret the results of the underlying machine learning model in a way that is understandable to humans. For example, if a particular term (e.g., "any") is labeled as a word of interest, it would be desirable to understand why the term "any" is labeled as a word of interest, especially when attempting to reduce the use of that term.

[0009] Other problems with current methods for implementing WSD include the inability to provide a generalizable approach across multiple domains. For example, a specific WSD-enabled NLP agent might be needed for a legal context, while a different WDS-enabled NLP agent might be needed for a business context. Similarly, some current methods are not easily implemented for multilingual applications. Therefore, there is a need for an improved system for providing WDS-enabled NLP agents that addresses these shortcomings of current methods. Summary of the Invention

[0010] A computer-implemented process for updating an electronic document includes the following operations: Preprocessing the electronic document using a preprocessor to generate a computer data structure; evaluating the computer data structure using a Word Sense Disambiguation (WSD) engine and a deep neural network to determine the context of sentences within the electronic document; determining, based on the context, words within the sentences that are words of interest; rewriting the sentences using a mitigation engine and a large language model to generate a revised sentence that does not include the words of interest; determining that the revised sentence does not include any other words of interest; and updating the electronic document to include the revised sentence.

[0011] In other respects, the process involves a deep neural network with an adaptive Skip-gram (AdaGram) model. Additionally, a graphical user interface can be presented to the user and configured to prompt the user to generate revised sentences. Furthermore, modification templates can be used to generate revised sentences. In some respects, the modification templates are automatically selected by the mitigation engine. In other respects, the graphical user interface is configured to provide word-level explanations of the sentences. After the initial revised sentence is determined to meet the similarity assessment or after the initial revised sentence is determined to contain other words of interest, a second modification template can be used to generate revised sentences.

[0012] A computer hardware system for updating an electronic document includes a hardware processor configured to perform the following executable operations: Preprocessing the electronic document using a preprocessor to generate a computer data structure; evaluating the computer data structure using a word sense disambiguation (WSD) engine and a deep neural network to determine the context of sentences within the electronic document; determining, based on the context, that words within the sentences are words of interest; rewriting the sentences using a mitigation engine and a large language model to generate a revised sentence that does not include the words of interest; determining that the revised sentence does not include any other words of interest; and updating the electronic document to include the revised sentence.

[0013] In other aspects, the system includes a deep neural network with an adaptive Skip-gram (AdaGram) model. Additionally, a graphical user interface can be presented to the user and configured to prompt the user to generate revised sentences. Furthermore, modified templates can be used to generate revised sentences. In some aspects, the modified templates are automatically selected by the mitigation engine. In other aspects, the graphical user interface is configured to provide word-level explanations of the sentences. After the initial revised sentence is determined to meet the similarity assessment or after the initial revised sentence is determined to contain other words of interest, a second modified template can be used to generate revised sentences.

[0014] A computer program product includes a computer-readable storage medium storing program code for updating an electronic document. When executed by a computer hardware system, the program code causes the computer hardware system to perform the following operations: Preprocessing is performed on the electronic document using a preprocessor to generate a computer data structure. The computer data structure is evaluated using a Word Sense Disambiguation (WSD) engine and a deep neural network to determine the context of sentences within the electronic document. Based on the context, words within the sentences are determined to be words of interest. The sentences are rewritten using a mitigation engine and a large language model to generate a revised sentence that does not include the words of interest. It is determined that the revised sentence does not include any other words of interest; and the electronic document is updated to include the revised sentence.

[0015] In other aspects, the computer program product includes a deep neural network with an adaptive Skip-gram (AdaGram) model. Additionally, a graphical user interface can be presented to the user and configured to prompt the user to generate revised sentences. Furthermore, modification templates can be used to generate revised sentences. In some aspects, the modification template is automatically selected by the mitigation engine. In other aspects, the graphical user interface is configured to provide word-level explanations of the sentences. After the initial revised sentence is determined to meet the similarity assessment or after the initial revised sentence is determined to contain other words of interest, a second modification template can be used to generate revised sentences.

[0016] The summary portion of this invention is provided merely to illustrate certain concepts and not to identify any key or essential features of the claimed subject matter. Other features of the invention will become apparent from the accompanying drawings and the detailed description below. Attached Figure Description

[0017] Figure 1 This is a block diagram of a deep context disambiguation system according to a certain aspect of the present invention.

[0018] Figure 2 It is to further demonstrate Figure 1 A block diagram of a deep context disambiguation system.

[0019] Figure 3 This illustrates the use according to a certain aspect of the invention. Figure 2 A block diagram of an example method for the architecture.

[0020] Figure 4 The detection algorithm according to various aspects of the present invention is shown.

[0021] Figure 5A -C illustrates the similarity measures according to various aspects of the invention and the different AdaGram outputs for different sentences.

[0022] Figure 6A-B illustrates a mitigation algorithm according to various aspects of the present invention.

[0023] Figure 7 This shows the implementation Figure 2 A block diagram of an example computer hardware system for a deep context disambiguation system. Detailed Implementation

[0024] refer to Figure 1-3 The diagram illustrates architecture 200 and method 300, respectively, and describes training an NLP agent as part of a deep context disambiguation system 220 for use in processing, for example, one or more electronic documents 215 generated by users 205A-C on client devices 210A-C. Although the deep context disambiguation system 220 is shown as including components 225-260, one or more of these components 225-660 may be external to the deep context disambiguation system 220. For example, a deep neural network 245, a large language model 255, and a knowledge base 345 may reside in the cloud, as per [previous context]. Figure 7 A more detailed description.

[0025] Knowledge base 345 includes words of interest. As used herein, the phrase "word of interest" is defined as a word (or phrase) that has previously been identified as risky in a particular context. Thus, knowledge base 345 can map specific words of interest to specific contexts. The method of performing this mapping is not limited to a particular approach. For example, whitelists and / or blacklists can be used. For example, a whitelist can include those contexts that do not make a specific word a "word of interest." Optionally, a blacklist can include those contexts that make a specific word a "word of interest." Regardless of the approach, knowledge base 345 includes information that allows word sense disambiguation engine 240 to determine whether a specific word is a "word of interest" based on its use in a particular context.

[0026] Although not limited to this method, preprocessor 230 is used to preprocess electronic document 225 to generate computer data structures. Word sense disambiguation (WSD) engine 240 and deep neural network 245 are then used to evaluate the computer data structures to determine the context of sentences within electronic document 225. Based on the context, words within the sentences are identified as words of interest. Sentences are rewritten using mitigation engine 250 and large language model 225 to generate revised sentences that do not include the words of interest. It is determined that the revised sentences do not include any other words of interest; and electronic document 225 is updated to include the revised sentences.

[0027] In 310, the operation begins with the deep context disambiguation system 220 receiving a document 215 from the client device 210 associated with the user 205. Although not limited to this method, the deep context disambiguation system 220 may receive the document 215 via network 202 through interface 225 (e.g., API).

[0028] In step 320, once document 215 has been received, preprocessor 230 performs preprocessing on document 215. Preprocessing may include, but is not limited to, removing useless information and formatting / tagout text content. Document preprocessing is well-known in natural language processing, and the deep context disambiguation system 220 is not limited to a specific preprocessor 230 capable of such implementation. In some aspects, preprocessing also includes defining sentence boundaries (e.g., delineating a sentence from another sentence) and filtering noise. The output of preprocessing operation 320 is a computer data structure, and although not limited to this manner, the computer data structure can specifically identify each sentence within document 215 and the tokenized context content associated with each sentence.

[0029] At 330, the computer data structures generated by preprocessor 230 are provided to WSD engine 240 to provide word sense disambiguation (i.e., determining the meaning of words in a sentence of document 215 within a given context). While not limited to this approach, WSD engine 240 may employ a deep neural network 245 that uses one or more detection algorithms to perform word sense disambiguation. Although shown separately from WSD engine 240, some or all of the deep neural network 245 may be included within WSD engine 240. See below. Figure 4 AC and 5A-B describe a detection algorithm used in a deep neural network 245. This serves as an example and reference. Figure 4 A and 5A-B can be modeled using the adaptive Skip-gram (AdaGram) model 335.

[0030] In 340, the WSD engine 240 uses the contextual output of the knowledge base 345 and the deep neural network 245 to identify words of interest. As previously mentioned, the knowledge base 345 includes information that allows the WSD engine 240 to determine whether a particular word is a "word of interest" based on its use in a specific context. The word itself is the output of the preprocessor 230, and the context is the output of the WSD engine 240 combined with the deep neural network 245.

[0031] While not limited to this method, identifying words of interest can include evaluating only those words already identified as potentially interesting words in the knowledge base 345, based on the words provided by preprocessing 230 (i.e., there exists at least one context in which the word can be evaluated as an interesting word). In this case, the WSG engine 240 can be configured to determine the context only for those words identified as potentially interesting words. This aspect improves processing efficiency by not evaluating contexts that were not previously identified as potentially interesting words.

[0032] Once potential words of interest are identified, the WSD engine 240 and deep neural network 245 analyze the potential words of interest and associated sentences to determine the specific context of the sentence and the meaning of the potential words of interest within that specific context. In step 350, based on the specific context of the identified sentence and information contained in the knowledge base 345 associated with the evaluated words of interest, the sentence is identified and output by the WSD engine 240 to the mitigation engine 250.

[0033] In 360, for each specific sentence recognized by the WSD engine 240, a mitigation algorithm including prompts is executed. (See reference...) Figure 6A -B discusses examples of mitigation algorithms.

[0034] In step 370, each of the identified sentences is rewritten. While not limited to this approach, generative AI can be used, employing a large language model 255, to rewrite the identified sentences. Illustrative examples of generative AI using a large language model 225 include BLOOM (BigScience, a large open scientific access multilingual language model) and PaLM (Pathways language model).

[0035] In section 380, similarity engine 260 is used to perform similarity analysis between the original and rewritten sentences. Although shown separately from mitigation engine 250, some or all of the similarity engine 260 may be included within mitigation engine 250. Similarity engine 260 is used to compare the contextual meanings of two sentences input into it to generate a similarity metric. Many different methods for determining sentence similarity are known, and similarity engine 260 is not limited to a particular method. One such method is BERT (Bidirectional Encoder Representation from Transformer), a deep learning model where each output element is connected to each input element, and weights between them are dynamically computed based on their connections.

[0036] Figure 5A An example of an algorithm for determining similarity is shown in the figure. Figure 5B and 5CThe example shows how AdaGram is used to evaluate the sentences shown, and the corresponding output shows the cases where the two sentences are not similar enough to each other.

[0037] Regardless of the method, the similarity measure outputs a similarity metric. This similarity metric is then evaluated against a predetermined value. If the generated sentence is not sufficiently similar to the original sentence (e.g., the similarity metric fails the evaluation), the newly generated sentence is considered unacceptable, and process 300 returns to prompt 360 to eventually repeat the process of generating a new sentence in operation 370. However, if the generated sentence is sufficiently similar to the original sentence (e.g., the similarity metric passes the evaluation), process 300 proceeds to operation 390.

[0038] In step 390, an evaluation is performed to determine whether the generated sentence contains the word of interest. This operation is similar to that performed in step 340, which identifies sentences containing previously identified words of interest contained in knowledge base 345. If the newly generated sentence contains the word of interest, it is considered unacceptable, and process 300 returns to prompt 360 to ultimately repeat the process of generating a new sentence in step 370. However, if the newly generated sentence does not contain the word of interest, the process proceeds to step 395.

[0039] As described above, operations 350-390 are performed on each identified sentence output in 350. The process only proceeds to 395 after each sentence has been rewritten (in operation 370) and successfully evaluated in operation 380. However, in some respects, user 205 may have the option to override the evaluation in 380 or 390. This can be performed, for example, by presenting a prompt in client device 210 asking the user whether they accept the revised sentence, even though the revised sentence is not semantically sufficiently similar to the original sentence (i.e., operation 380) and / or still contains risky words (i.e., operation 390). If the user accepts the generated statement, processing 300 does not return to prompt operation 360.

[0040] In step 395, the update of electronic document 215 is finalized. This operation may include, for example, modifying electronic document 215 by replacing the identified sentence in step 350 with the generated sentence generated in step 370. The updated electronic document 215 can then be forwarded to client device 210. Additionally or alternatively, the final determination of the update of electronic document 215 may include publishing and / or storing electronic document 215. Different detection algorithms that can be used with disambiguation engine 240 and deep neural network 245 are disclosed. As mentioned above, disambiguation engine 240 and deep neural network 245 are not limited to a specific detection algorithm.

[0041] refer to Figure 4This paper demonstrates a detection algorithm using the Adaptive Skip-gram (AdaGram) model 335. The AdaGram model 335 is a nonparametric Bayesian extension of Skip-gram that is capable of automatically learning the required number of representations for all words with the desired semantic resolution. The advantages of using the AdaGram model 335 include fewer error locations, self-supervised training (i.e., no annotations are required when training the model 335), flexibility (i.e., the model 335 can be domain- and / or language-specific), and the ability to learn the depth of meaning (i.e., the model 335 can learn small, subtle semantic differences, which is important for word sense disambiguation).

[0042] refer to Figure 6A -B indicates the mitigation algorithm used to perform prompting operation 360. In this mitigation algorithm, the mitigation engine 250 can provide prompts to the client device 210 for the user 205 to automatically correct words of interest. If accepted, the mitigation engine 250 can perform automatic operations, such as those described in operation 370, to rewrite the sentence. Prompts can then be provided to the user 205 via the client device 210 to accept these changes.

[0043] In addition, such as Figure 6B As shown, the mitigation engine 250 may include multiple modification templates 605, which can then be used by the large language model 255 in operation 370 to rewrite sentences. In some aspects, the mitigation engine 250 may automatically select a specific modification template 605 to be used in operation 370 based on information contained within the knowledge base 345. As previously described, different modification templates 605 may be used in operation 370 when process 300 returns to prompt operation 360 (e.g., from operation 380 or operation 390).

[0044] Similarly, Figure 6B As shown, the mitigation engine 250 provides a graphical user interface 610 that allows user 205 to select a specific AI-generated large language model 255 to be used during operation 370 and to adjust its associated parameters. The graphical user interface 610 may also provide word-level explanations of why a particular word is considered a word of interest. This may include providing information including, but not limited to, (i) the word, (ii) the specific context, and (iii) the generated new sentence.

[0045] As defined herein, the term "in response to" means to respond to or react to an action or event. Therefore, if a second action is performed "in response to" a first action, there is a causal relationship between the occurrence of the first action and the occurrence of the second action, and the term "in response to" indicates such a causal relationship.

[0046] As defined in this article, the term “real-time” refers to the level of processing response perceived by the user or system that is sufficient to immediately complete a specific processing or decision to be made, or to enable the processor to keep up with certain external processes.

[0047] As defined in this article, the term “automatically” means without user intervention.

[0048] refer to Figure 7 The computing environment 700 includes examples of environments for executing at least some of the computer code involved in performing the methods of the present invention, such as code blocks 750 for implementing the operation of the deep context disambiguation system 220 and client devices 210A-C. The computing environment 700 includes, for example, a computer 701, a wide area network (WAN) 702, an end-user equipment (EUD) 703, a remote server 704, a public cloud 705, and a private cloud 706. In some aspects, the computer 701 includes a processor set 710 (including processing circuitry 720 and a cache 721), a communication structure 711, volatile memory 712, persistent storage 713 (including an operating system 722 and method code blocks 750), a peripheral device set 714 (including a user interface (UI), a device set 723, a storage device 724, and an Internet of Things (IoT) sensor set 725), and a network module 715. The remote server 704 includes a remote database 730. Public cloud 705 includes gateway 740, cloud orchestration module 741, host physical machine set 742, virtual machine set 743, and container set 744.

[0049] Computer 701 can take the form of a desktop computer, laptop computer, tablet computer, smartphone, smartwatch or other wearable computer, mainframe computer, quantum computer, or any other form of computer or mobile device now known or to be developed in the future capable of running programs, accessing networks, or querying databases such as remote database 730. As is well known in the field of computer technology, and depending on the technology, the performance of a computer-implemented method can be distributed among multiple computers and / or multiple locations. However, to simplify this presentation of computing environment 700, the detailed discussion focuses on a single computer, specifically computer 701. Computer 701 may or may not be located in the cloud, even if it is not in... Figure 7 The cloud in the image is shown unless it can be definitively indicated to any extent.

[0050] Processor assembly 710 includes one or more computer processors of any type now known or to be developed in the future. As defined herein, the term "processor" means at least one piece of hardware circuitry (e.g., an integrated circuit) configured to execute instructions contained in program code. Examples of processors include, but are not limited to, central processing units (CPUs), array processors, vector processors, digital signal processors (DSPs), field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), application-specific integrated circuits (ASICs), programmable logic circuits, and controllers. Processing circuitry 720 may be distributed across multiple packages, such as multiple cooperating integrated circuit chips. Processing circuitry 720 may implement multiple processor threads and / or multiple processor cores. Cache 721 is memory located within the processor chip package and is typically used for data or code that should be readily accessible by the threads or cores running on processor assembly 710. Cache memory is typically organized into multiple levels based on its relative proximity to the processing circuitry. Alternatively, some or all of the cache in the processor assembly may be located "off-chip". In some computing environments, processor assembly 710 may be designed to work with qubits and perform quantum computing.

[0051] Computer-readable program instructions are typically loaded onto computer 701 to cause the processor set 710 of computer 701 to perform a series of operational steps to implement a computer-implemented method, such that the instructions thus executed instantiate the method specified in the flowcharts and / or descriptive descriptions of the computer-implemented method included in this document (collectively, the "method of the invention"). These computer-readable program instructions are stored in various types of computer-readable storage media, such as cache 721 and other storage media discussed below. The program instructions and associated data are accessed by the processor set 710 to control and direct the execution of the method of the invention. In computing environment 700, at least some of the instructions for performing the method of the invention may be stored in code blocks 750 in permanent storage device 713.

[0052] Computer Program Product Embodiment (“CPP Embodiment” or “CPP”) is a term used in this disclosure to describe any collection of one or more storage media (also referred to as “media”) collectively included in a collection of one or more storage devices, the collection of one or more storage devices collectively including machine-readable code corresponding to instructions and / or data for performing computer operations specified in a given CPP claim. A “storage device” is any tangible hardware device that can hold and store instructions used by a computer processor. Without limitation, a computer-readable storage medium can be an electronic storage medium, a magnetic storage medium, an optical storage medium, an electromagnetic storage medium, a semiconductor storage medium, a mechanical storage medium, or any suitable combination of the foregoing. Some known types of storage devices that include these media include: magnetic disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), compact disc read-only memory (CD-ROM), digital versatile disc (DVD), memory sticks, floppy disks, mechanical encoding devices (such as punch cards or pits / platforms formed in the main surface of the disk), or any suitable combination of the foregoing.

[0053] Computer-readable storage media, as used in this disclosure, should not be construed as storing transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides, optical pulses through fiber optic cables, electrical signals transmitted through wires, and / or other transmission media. As those skilled in the art will understand, data is typically moved at certain incidental points in time during the normal operation of the storage device, such as during access, defragmentation, or garbage collection; however, this does not make the storage device transient, because the data is not transient when it is stored.

[0054] Communication structure 711 is a signal transmission path that allows various components of computer 701 to communicate with each other. Typically, communication structure 711 consists of switches and conductive paths, such as switches and conductive paths that form buses, bridges, physical input / output ports, etc. Other types of signal communication paths can be used in communication structure 711, such as fiber optic communication paths and / or wireless communication paths.

[0055] Volatile memory 712 is any type of volatile memory now known or to be developed in the future. Examples include dynamic random access memory (RAM) or static RAM. Typically, volatile memory 712 is characterized by random access, but this is not necessary unless explicitly indicated. In computer 701, volatile memory 712 is located in a single package and is internal to computer 701. Alternatively, volatile memory 712 may be distributed across multiple packages and / or located external to computer 701.

[0056] The persistent storage device 713 is any form of non-volatile memory for a computer, now known or to be developed in the future. The non-volatility of the persistent storage device 713 means that the stored data is retained regardless of whether power is supplied to the computer 701 and / or directly to the persistent storage device 713. The persistent storage device 713 may be a read-only memory (ROM), but typically at least a portion of the persistent storage device 713 allows for data writing, data deletion, and data rewriting. Some familiar forms of persistent storage devices 713 include hard disks and solid-state storage devices. The operating system 722 may take several forms, such as various known proprietary operating systems or operating systems employing an open-source portable operating system interface type with a kernel. The code included in code block 750 generally includes at least some of the computer code involved in performing the methods of the present invention.

[0057] The peripheral device set 714 includes the peripheral device set of computer 701. Data communication connections between peripheral devices and other components of computer 701 can be implemented in various ways, such as Bluetooth connections, near field communication (NFC) connections, connections made by cables (such as Universal Serial Bus (USB) type cables), plug-in connections (e.g., secure digital card (SD) cards), connections made through local area communication networks, and even connections made through wide area networks such as the Internet.

[0058] In various aspects, the UI device set 723 may include components such as displays, speakers, microphones, wearable devices (such as goggles and smartwatches), keyboards, mice, printers, touchpads, game controllers, and haptic devices. Storage device 724 is an external storage device, such as an external hard drive, or a pluggable storage device, such as an SD card. Storage device 724 may be persistent and / or volatile. In some aspects, storage device 724 may take the form of a quantum computing storage device for storing data in the form of qubits. In aspects where computer 701 requires a large amount of storage (e.g., where computer 701 locally stores and manages a large database), the storage device 724 may be provided by a peripheral storage device designed for storing very large amounts of data, such as a storage area network (SAN) shared by multiple geographically distributed computers. The Internet of Things (IoT) sensor set 725 consists of sensors that can be used in IoT applications. For example, one sensor may be a thermometer, while another sensor may be a motion detector.

[0059] Network module 715 is a collection of computer software, hardware, and firmware that allows computer 701 to communicate with other computers via a wide area network (WAN) 702. Network module 715 may include hardware such as a modem or Wi-Fi transceiver, software for packetizing and / or depacketizing data transmitted over the communication network, and / or web browser software for transmitting data over the Internet. In some aspects, the network control and network forwarding functions of network module 715 are performed on the same physical hardware device. In other aspects (e.g., using software-defined networking (SDN), the control and forwarding functions of network module 715 are performed on physically separate devices, such that the control function manages several different network hardware devices. Computer-readable program instructions for performing the methods of the present invention can typically be downloaded to computer 701 from an external computer or external storage device via a network adapter card or network interface included in network module 715.

[0060] A WAN 702 is any wide area network (e.g., the Internet) capable of transmitting computer data over non-local distances using any technology known now or developed in the future for transmitting computer data. In some respects, a WAN 702 can be replaced and / or supplemented by a local area network (LAN) designed to transmit data between devices located in a local area, such as a Wi-Fi network. WAN 702 and / or LAN typically include computer hardware such as copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and edge servers.

[0061] End User Equipment (EUD) 703 is any computer system used and controlled by an end user (e.g., a customer of the enterprise operating computer 701) and can take any of the forms discussed above in conjunction with computer 701. EUD 703 typically receives useful and helpful data from the operation of computer 701. For example, assuming computer 701 is designed to provide recommendations to the end user, these recommendations are typically transmitted from network module 715 of computer 701 to EUD 703 via WAN 702. Thus, EUD 703 can display or otherwise present the recommendations to the end user. In some respects, EUD 703 can be a client device, such as a thin client, heavy client, mainframe, desktop computer, etc.

[0062] As defined herein, the term "client device" refers to a data processing system that requests shared services from a server and with which a user interacts directly. Examples of client devices include, but are not limited to, workstations, desktop computers, computer terminals, mobile computers, laptop computers, netbooks, tablet computers, smartphones, personal digital assistants, smartwatches, smart glasses, gaming devices, set-top boxes, and smart TVs. Network infrastructure such as routers, firewalls, switches, and access points are not client devices because the term "client device" is defined herein. As defined herein, the term "user" means person (i.e., human).

[0063] Remote server 704 is any computer system that provides at least some data and / or functionality to computer 701. Remote server 704 can be controlled and used by the same entity operating computer 701. Remote server 704 represents a machine that collects and stores useful and useful data for use by other computers such as computer 701. For example, if computer 701 is designed and programmed to provide recommendations based on historical data, that historical data can be provided to computer 701 from a remote database 730 of remote server 704. As defined herein, the term "server" refers to a data processing system configured to share services with one or more other data processing systems.

[0064] Public cloud 705 is any computer system that can be used by multiple entities, providing on-demand availability of computer system resources and / or other computing capabilities (especially data storage (cloud storage) and computing power) without the need for direct, active management by users. Cloud computing typically leverages resource sharing to achieve scalability and cost-effectiveness. Direct and active management of the computing resources of public cloud 705 is performed by the computer hardware and / or software of cloud orchestration module 741. The computing resources provided by public cloud 705 are typically implemented by virtual computing environments running on various computers constituting host physical machine set 742, which is the entirety of physical computers in public cloud 705 and / or available to that public cloud.

[0065] Virtual Computing Environments (VCEs) typically take the form of virtual machines from virtual machine set 743 and / or containers from container set 744. It should be understood that these VCEs can be stored as images and can be transferred between various physical machine hosts as images or after the VCEs are instantiated. The cloud coordination module 741 manages the transfer and storage of images, deploys new instantiations of VCEs, and manages active instantiations of VCE deployments. Gateway 740 is a collection of computer software, hardware, and firmware that allows the public cloud 705 to communicate via WAN 702.

[0066] VCEs can be stored as "images," and new active instances of VCEs can be instantiated from these images. Two common types of VCEs are virtual machines and containers. Containers are VCEs that utilize operating system-level virtualization. This refers to an operating system feature where the kernel allows multiple isolated user-space instances, called containers, to exist. From the perspective of the programs running within them, these isolated user-space instances typically behave like actual computers. Computer programs running on a regular operating system can utilize all the resources of that computer, such as connected devices, files and folders, network shares, CPU power, and quantifiable hardware capabilities. However, programs running within containers can only use the contents of the container and the devices allocated to the container; this is a characteristic known as containerization.

[0067] Private cloud 706 is similar to public cloud 705, except that computing resources can only be used by a single enterprise. While private cloud 706 is described as communicating with WAN 702, in other respects, private cloud 706 can be completely disconnected from the Internet (e.g., WAN 702) and accessed only through a local / private network. Hybrid cloud is a combination of multiple clouds of different types (e.g., private, community, or public cloud types) typically implemented by different vendors. Each of the multiple clouds remains a separate and discrete entity, but the larger hybrid cloud architecture is bound together by standardization or proprietary technology that enables coordination, management, and / or data / application portability between the multiple component clouds. In this respect, both public cloud 705 and private cloud 706 are part of a larger hybrid cloud.

[0068] Various aspects of this disclosure are described by narrative text, flowcharts, block diagrams of computer systems, and / or block diagrams of machine logic included in embodiments of a computer program product (CPP). Regarding any flowchart, depending on the technology involved, operations may be performed in a different order than that shown in a given flowchart. For example, again according to the technology involved, two operations shown in consecutive flowchart blocks may be performed in reverse order, as a single integrated step, simultaneously, or in a manner that at least partially overlaps in time.

[0069] As another example, two boxes shown consecutively may actually be implemented as a single step, executed simultaneously, substantially simultaneously, with partial or complete time overlap, or these boxes may sometimes be executed in reverse order, depending on the functionality involved. It will also be noted that each box in the block diagram and / or flowchart illustration, and combinations of boxes in the block diagram and / or flowchart illustration, may be implemented by a dedicated hardware-based system that performs the specified function or action or executes a combination of dedicated hardware and computer instructions. Each box in a flowchart or block diagram may represent a module, segment, or portion of instructions, which includes one or more executable instructions for implementing the specified logical function.

[0070] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the terms “includes,” “including,” “comprises,” and / or “comprising,” when used in this disclosure, specify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0071] In this disclosure, references to “an embodiment,” “an embodiment,” “an arrangement,” “an aspect,” “an aspect,” or similar language mean that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment described within this disclosure. Therefore, the phrases “an embodiment,” “an embodiment,” “an arrangement,” “an aspect,” “an aspect,” and similar language appearing in this disclosure may, but not necessarily all, refer to the same embodiment.

[0072] As used herein, the term "a plurality of" is defined as two or more. As used herein, the term "another" is defined as at least a second or more. Unless otherwise indicated, as used herein, the term "coupled" is defined as a connection, whether directly without any intermediate element or indirectly with one or more intermediate elements. Two elements may also be mechanically, electrically, or communicatively connected via a communication channel, path, network, or system. As used herein, the term "and / or" refers to and covers any and all possible combinations of one or more of the associated listed items. It will also be understood that although the terms first, second, etc., may be used herein to describe various elements, these elements should not be limited by these terms, as they are used only to distinguish one element from another unless otherwise stated or indicated by the context.

[0073] The term "if" can be interpreted as meaning "when," "once," "in response to determination," or "in response to detection," depending on the context. Similarly, the phrases "if determination" or "if [the condition or event] is detected" can be interpreted as meaning "once determination," "in response to determination," "once [the condition or event] is detected," or "in response to detection," depending on the context. As used herein, the terms "if," "when," "once," "in response," etc., should not be interpreted as indicating that a particular operation is optional. Rather, the use of these terms indicates that a particular operation is conditional. For example, and by assumption, the language "once B performs operation A" does not indicate that operation A is optional. Rather, the language indicates that operation A is conditional upon B occurring.

[0074] The foregoing description is merely an example of embodiments of the invention, along with variations and substitutions. While this disclosure concludes with claims defining novel features, it is believed that the various features described herein will be better understood by considering the description in conjunction with the accompanying drawings. Processes, machines, manufactures, and any variations thereof described in this disclosure are provided for illustrative purposes. Any specific structural and functional details described should not be construed as limiting, but merely as the basis for the claims and as a representative basis for teaching those skilled in the art to adopt the features described in virtually any suitable detailed structure. Furthermore, the terminology and phrases used in this disclosure are not intended to be limiting, but rather to provide an understandable description of the described features.

Claims

1. A computer-implemented method for updating electronic documents using a preprocessor, a word sense disambiguation WSD engine, and a mitigation engine, comprising: The electronic document is preprocessed using the preprocessor to generate a computer data structure; The context of sentences within the electronic document is determined by evaluating the computer data structure using the WSD engine and deep neural network. Based on the context, the words in the sentence are determined to be words of interest; The sentence is rewritten using the mitigation engine and the large language model to generate a revised sentence that does not include the word of interest; It was determined that the revised sentence did not contain any other words of interest; as well as Update the electronic document to include the revised sentence.

2. The method according to claim 1, wherein, The deep neural network includes the adaptive Skip-gram, or AdaGram model.

3. The method according to any one of the preceding claims, wherein, A graphical user interface is presented to the user, and The graphical user interface is configured to prompt the user to generate the revised sentence.

4. The method according to any one of the preceding claims, wherein, The revised sentence was generated using a modification template.

5. The method according to claim 4, wherein, The modified template is automatically selected by the mitigation engine.

6. The method according to claim 4, wherein, A graphical user interface is presented to the user, and The graphical user interface is configured to provide word-level explanations of sentences.

7. The method according to claim 4, wherein, After the initial revised sentence is determined to meet the similarity assessment, a second modified template is used to generate the revised sentence.

8. The method according to claim 4, wherein After the initial revised sentence is identified as containing other words of interest, a second modification template is used to generate the revised sentence.

9. A computer hardware system for updating electronic documents, comprising: A hardware processor, including a preprocessor, a word sense disambiguation WSD engine, and a mitigation engine, is configured to perform the following executable operations: The electronic document is preprocessed using the preprocessor to generate a computer data structure; The context of sentences within the electronic document is determined by evaluating the computer data structure using the WSD engine and deep neural network. Based on the context, the words in the sentence are determined to be words of interest; The sentence is rewritten using the mitigation engine and the large language model to generate a revised sentence that does not include the word of interest; It was determined that the revised sentence did not contain any other words of interest; as well as Update the electronic document to include the revised sentence.

10. The system according to claim 9, wherein, The deep neural network includes the adaptive Skip-gram, or AdaGram model.

11. The system according to any one of claims 9 to 10, wherein, A graphical user interface is presented to the user, and The graphical user interface is configured to prompt the user to generate the revised sentence.

12. The system according to any one of claims 9 to 11, wherein, The revised sentence was generated using a modification template.

13. The system according to claim 12, wherein, The modified template is automatically selected by the mitigation engine.

14. The system according to claim 12, wherein, A graphical user interface is presented to the user, and The graphical user interface is configured to provide word-level explanations of sentences.

15. The system according to claim 12, wherein, After the initial revised sentence is determined to meet the similarity assessment, a second modified template is used to generate the revised sentence.

16. The system according to claim 12, wherein, After the initial revised sentence is identified as containing other words of interest, a second modification template is used to generate the revised sentence.

17. A computer program product comprising: A computer-readable storage medium containing program code for updating electronic documents. When the program code is executed by a computer hardware system including a preprocessor, a word sense disambiguation WSD engine, and a mitigation engine, the computer hardware system performs the following: The electronic document is preprocessed using the preprocessor to generate a computer data structure; The context of sentences within the electronic document is determined by evaluating the computer data structure using the WSD engine and deep neural network. Based on the context, the words in the sentence are determined to be words of interest; The sentence is rewritten using the mitigation engine and the large language model to generate a revised sentence that does not include the word of interest; It was determined that the revised sentence did not contain any other words of interest; as well as Update the electronic document to include the revised sentence.

18. The computer program product according to claim 17, wherein, The deep neural network includes the adaptive Skip-gram, or AdaGram model.

19. The computer program product according to any one of claims 17 to 18, wherein, The revised sentence was generated using a modification template.

20. The computer program product according to any one of claims 17 to 19, wherein, After the initial revised sentence is determined to meet the similarity assessment or contain other words of interest, a second modification template is used to generate the revised sentence.

Images