Methods and systems for generating replies to data packages using sentiment analysis
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-02-03
- Publication Date
- 2026-08-13
Smart Images
Figure DK2026060007_13082026_PF_FP_ABST
Abstract
Description
[0001] METHODS AND SYSTEMS FOR GENERATING REPLIES TO DATA PACKAGES USING SENTIMENT ANALYSIS
[0002] Technical Field
[0003] The present disclosure relates to systems and methods for automated response generation, and more particularly to generating replies to data packages using sentiment analysis and historical response data. The disclosure further relates to computer-implemented techniques for processing and analyzing communications between senders and recipients.
[0004]
[0005] Communication processing systems process increasing volumes of electronic messages across various channels as organizations manage customer interactions. The growing volume of messages creates operational challenges when relying on manual processing approaches. Organizations seek to maintain service levels while managing available resources efficiently. Manual processing can create workflow constraints as message volumes fluctuate throughout operating periods.
[0006] Traditional communication approaches depend on human operators to review incoming messages and prepare individual responses. This manual approach requires time to process each message and compose reply content. The output can vary between different operators based on their experience levels and domain knowledge. Processing times are affected by operator availability and workload distribution across teams handling communications.
[0007] Various systems have been developed to support communication processing operations. Different approaches allow organizations to handle common message types through different processing methods. However, these systems may face limitations in addressing specific communication requirements. Existing approaches encounter challenges with variations in communication formats and styles. Systems may not fully address the detailed aspects of specific messages.
[0008] Organizations maintain records of their communication activities including various message types and outcomes. These records typically include different forms of interactions between organizations and message senders. Current systems process
[0009] P24-060PCT1communications using different methodological approaches. Manual review of communications requires time investment from operators. The information contained in communication records presents opportunities for process improvements.
[0010] Summary
[0011] A method and system generate replies to data packages through a process of data analysis and reply customization. The process involves receiving data packages from senders and extracting data from these packages. The extracted data undergoes analysis to determine features of the data package, including sentiment data associated with the sender. Based on this analysis, the system determines reply instances from previously stored data packages in a database. The reply instances undergo adjustments according to the analysis results. The system then generates replies based on the adjusted reply instances and sends these replies to the senders.
[0012] The system includes storage components for maintaining a database of previous data packages and their corresponding replies. A processor handles the data processing operations through various modules. A data extraction module extracts data from received packages, while an analysis module examines the extracted data to determine package features, including sender sentiment data. A reply generation module manages the determination of appropriate reply instances, implements adjustments to these instances, and generates the final replies. An input-output device facilitates the reception of data packages and transmission of replies.
[0013] The functionality can be implemented through computer-readable storage media containing executable instructions. When executed by a processor, these instructions enable the reception and processing of data packages, extraction and analysis of data, determination and adjustment of reply instances, and transmission of generated replies.
[0014]
[0015] Examples of the present disclosure will now be described with reference to the accompanying drawings:
[0016] Figure 1 is a flow diagram showing a method of generating a reply to a data package;
[0017] P24-060PCT1Figure 2 is a sequence diagram showing interactions between a sender and a recipient; and
[0018] Figure 3 is a schematic diagram showing a system for generating a reply to a data package.
[0019] Detailed
[0020]
[0021] Method of Generating a Reply to a Data Package
[0022] Figure 1 shows a flow diagram illustrating a method of generating a reply to a data package. The method includes receiving a data package 102 from a sender. The data package 102 may be an email or other digital communication.
[0023] Upon receipt of the data package 102, data is extracted 104 from the data package. The extracted data may include a subject, a body of the data package such as an email body, a previous response part of a chain associated with the data package, a category of the data package such as the type, subtype, or reason for the data package being sent, and a domain associated with the data package 102. It will be appreciated that other data may also be extracted from the data package 102.
[0024] The extracted data 104 is analyzed 106 to determine features of the data package 102. The features may include sentiment data, including the intent of the sender of the data package 102, contextual details such as information relating to an associated case, and the language of the email including grammar. Analyzing the extracted data 104 may comprise aggregating the sentiment data or other features that have been extracted by intent, creating clusters based on the sender's requirements. Within each intent node / cluster, further categorization may be applied such that relevant contextual details provided by the sender and / or extracted from the data package 102 are grouped. This may include applying a large language model or other models, such as an ada-embedding model, to the data package 102 to categorize the features.
[0025] Analysis of the data package 102 may include categorizing specific features such as the query type, a sub-type, the extracted intent, and contextual details including booking numbers, cargo details, payment details, invoice details, and previous replies to similar data packages. The analyzed data may then be indexed and stored in storage, such as storage 304 described below in relation to Figure 3, for future analysis.
[0026] P24-060PCT1Based on the analysis, a reply instance is determined 108 from a database of reply instances. A density-based clustering technique may be applied to identify frequent replies from the database when a similar analysis has occurred. The database comprises optimal reply suggestions for future queries alongside intent and context information relating to the data packages which the reply was being sent in response to. The database may be stored in the same storage, or a different storage as the analysed data.
[0027] At least one adjustment to the reply instance is generated 110. Adjustments may include modifying the content of the determined reply instance to reflect the context and sentiment of the received data package. This may include adjustments to data such as booking numbers, customer details, and other data such as the language and grammar of the originally received data package which are indicative of the context and sentiment of the sender.
[0028] A reply is generated 112 based on the reply instance and the adjustment. This may include replacing text or other data in the determined reply instance with the generated adjustment(s). The generated reply is then sent 114 to the sender of the originally received data package.
[0029] The method assists support teams by providing intelligent recommendations based on detected intent and sentiment to answer customer issues / queries in the customer's language whilst taking into account the emotional state and other unique characteristics of the sender. This includes essential customer information extracted from the data packages 102, enabling tailored replies with minimal edits, reducing response time and improving customer experience.
[0030] The analysis of sentiment data enables the generation of contextually appropriate responses that take into account the emotional state and tone of the sender. The extracted sentiment information allows for nuanced adjustments to standard replies. This helps create more empathetic and personalized communications that resonate with the sender's current state of mind.
[0031] The system and method continuously refines and updates stored reply instances based on ongoing analysis of communications. As new data packages are processed and responses generated, the database of reply instances evolves to incorporate successful
[0032] P24-060PCT1response patterns. This adaptive approach helps maintain relevance and effectiveness of future replies.
[0033] The combination of stored reply instances with dynamic adjustments provides an efficient approach to response generation. Pre-analyzed reply instances serve as templates that can be rapidly customized based on current context and sentiment data. The structured methodology reduces the time needed to craft appropriate responses while maintaining consistency and quality.
[0034] System Components and Operational Environment
[0035] Figure 2 illustrates the sequence of operations described in Figure 1 in more detail. A sender 202 may be a customer of a service provider, and may send data packages using a user device such as a mobile device, desktop device, or wearable device including cellular telephones, smart watches, laptop computers or other suitable devices. The recipient 204 may be a service provider, such as a company providing services to the sender 202, using a system, such as system 302 described below in relation to Figure 3.
[0036] A database 206 may take any suitable form and be stored in storage taking any suitable form. The storage, such as storage 304 described below in relation to Figure 3, may comprise a main memory or primary memory, and may be external or internal to the system. The memory may be off-chip or on-chip, and may comprise magnetic or optical disks, disk drives, solid-state drives (SSDs), synchronous dynamic random-access memory (SDRAM), or double data rate synchronous dynamic random-access memory (DDR-SDRAM).
[0037] The sequence begins when the sender 202 sends a data package 102, such as an email or other digital communication, which forms part of message 208 sent to the recipient 204. At block 210, data is extracted from the data package 102, including elements such as the subject, email body, previous response chains, category information like type and subtype, and associated domain information.
[0038] At block 212, the extracted data is analyzed to determine features including sentiment data, sender intent, contextual details relating to associated cases, and language characteristics including grammar. The analysis aggregates sentiment data and features by intent, creating clusters based on sender requirements. Within each intent node,
[0039] P24-060PCT1categorization groups relevant contextual details using techniques such as large language models or ada-embedding models. The analysis categorizes specific features like query type, sub-type, extracted intent, and contextual details such as booking numbers, cargo details, payment details, and invoice details. This analyzed data is indexed and stored for future reference.
[0040] Message 214 initiates a query to the database to identify frequent replies using density-based clustering techniques. The database contains previous reply suggestions alongside intent and context information from previous data packages. Message 216 returns the matching reply instances from the database.
[0041] At block 218, adjustments to the reply instance are determined based on the context and sentiment of the received data package. These adjustments may include updating booking numbers, customer details, and adapting language and grammar to match the sender's style.
[0042] Block 220 generates the reply by incorporating the determined adjustments into the selected reply instance. Finally, the generated reply is sent back to the sender 202 of the original data package as part of message 222. It will be appreciated that other data may be sent alongside the message 222 or as part of the message 222.
[0043] System for Generating a Reply to a Data Package
[0044] Figure 3 shows a schematic diagram of a system 302 configured to perform the method described in relation to Figure 1. The system 302 generates at least one reply 318 to at least one data package 316 and may form part of a remote server, remote from a sender device used to send the data package from the sender.
[0045] The system 302 includes storage 304 for storing at least a database of previous data packages received from a sender and corresponding replies sent. The storage 304 may comprise a main memory or primary memory, which can be external to the system such as off-chip memory or remote cloud storage. The storage 304 may include magnetic or optical disks, solid-state drives (SSDs), or synchronous dynamic random-access memory (SDRAM).
[0046] A processor 306 processes data packages 316 received by the system 302. The processor 306 includes a data extraction module 308 configured to extract data from
[0047] P24-060PCT1received data packages 316. The extracted data may include a subject, email body, previous response chain, category information, and domain association.
[0048] The processor 306 also includes an analysis module 310 that analyzes the extracted data to determine features of the data package, including sentiment data relating to the sender. Features may encompass the sender's intent, contextual details about associated cases, and language characteristics including grammar. The analysis may involve aggregating sentiment data by intent and creating clusters based on sender requirements. This can include applying large language models or ada-embedding models to categorize features.
[0049] A reply generation module 312 determines reply instances from the database based on the analysis. The reply generation module 312 applies density -based clustering techniques to identify frequent replies from similarly analyzed previous data packages. The database contains optimal reply suggestions alongside intent and context information. The reply generation module 312 determines adjustments to the reply instance based on the current analysis, such as modifying content to reflect context and sentiment, updating booking numbers, customer details, and language style.
[0050] An input-output device 314 facilitates communication between the system and users, either locally or via a wide area network, like the Internet. The input-output device receives data packages 316 from senders and sends generated replies 318. The data packages 316 may be emails or other digital communications, while replies 318 are generated taking into account the context the sentiment in the received data package, the analysis undertaken, the reply instance(s) and the adjustments used by the reply generation module 312 to generate the reply.
[0051] Components of the system may be connected via an interconnection such as a system bus, or other wired or wireless connections enabling communication across networks such as the Internet. The system continuously processes incoming communications, analyzing context and sentiment to generate appropriate responses that maintain consistency while adapting to specific sender needs.
[0052] The system enhances response quality through continuous learning and adaptation. As communications are processed, the analysis results inform adjustments to
[0053] P24-060PCT1stored reply instances. This adaptive approach allows the system to refine and optimize responses based on accumulated insights from ongoing interactions.
[0054] The modular architecture separates key processing functions into dedicated components for extraction, analysis, and generation. Each component performs specialized tasks while maintaining seamless integration through standardized data flows. This structured approach supports efficient processing of communications while allowing independent optimization of each function.
[0055] By incorporating sentiment analysis, the system can generate responses aligned with the emotional context of received communications. The analysis module processes sentiment data to understand the sender's emotional state. This understanding enables generation of replies that appropriately address both content and tone.
[0056] The combination of stored reply instances and dynamic adjustment capabilities provides rapid response generation. Previously successful replies serve as templates that can be modified based on current context. This approach reduces processing overhead while maintaining response relevance and appropriateness.
[0057] The database structure enables systematic organization of historical communications and corresponding replies. This organization facilitates quick retrieval of relevant previous interactions and reply patterns. The structured storage supports efficient analysis of communication history and identification of applicable reply instances.
[0058] The integrated input-output functionality enables continuous processing of communications. Messages are received, analyzed, and responded to without interruption or delay. This streamlined handling maintains communication flow while ensuring appropriate processing of each interaction.
[0059] The implementation using a computer-readable storage medium enables deployment across various computing platforms and environments. The storage medium can be integrated with different processor architectures and operating systems. This flexibility allows the functionality to be utilized on diverse hardware configurations while maintaining consistent operation.
[0060] P24-060PCT1The analysis of sentiment data facilitates generation of responses aligned with the emotional context of received communications. By processing sentiment indicators within the data package, responses can be tailored to acknowledge and appropriately address the sender's emotional state. The generated replies incorporate appropriate tone and language based on the detected sentiment.
[0061] The utilization of stored reply instances combined with dynamic adjustments streamlines the response generation process. Previously successful responses can be rapidly retrieved and modified according to current requirements. This approach reduces processing overhead while ensuring responses remain relevant to each specific situation.
[0062] The system incorporates ongoing analysis of communications to refine stored reply instances overtime. As new data packages are processed, the analysis results inform adjustments to existing responses. This continuous refinement helps maintain and enhance the relevance of stored replies.
[0063] The sequential processing workflow from receipt through analysis to reply ensures methodical handling of incoming communications. Each step builds upon the previous processing results in a structured manner. The organized approach helps maintain consistency in how communications are processed and responded to.
[0064] The combination of stored reply instances with real-time adjustments enables contextually appropriate response generation. Retrieved responses can be modified based on current analysis results. This approach balances efficiency with customization to produce suitable replies for each situation.
[0065] Additional Variations
[0066] In some examples, the storage 304 and database 206 may be implemented using various memory configurations. The storage 304 may comprise either external 'off-chip' memory or internal 'on-chip' memory relative to the system. The storage 304 may be implemented as remote cloud storage or server-based hard drives. The database 206 and storage 304 may utilize magnetic or optical disk drives, solid-state drives (SSDs), synchronous dynamic random-access memory (SDRAM), or double data rate synchronous dynamic random-access memory (DDR-SDRAM).
[0067] P24-060PCT1The data package 316 may be received through various communication channels and formats. While email communications represent one format, the data package 316 may also comprise instant messages, social media posts, text messages, or other forms of digital communication sent from the sender 202 to a recipient.
[0068] The sender 202 may utilize different types of user devices to transmit the data package 316. These devices may include mobile phones, tablets, desktop computers, laptop computers, smart watches, or other wearable devices capable of digital communication. The sender 202 may interact with multiple devices across different communication sessions.
[0069] The supplemental data associated with the sender 202 may be organized and analyzed in various ways. The clustering of supplemental data may be based on different contextual parameters such as time of communication, device type used, location of sender, or topic of communication. Multiple clustering approaches may be applied simultaneously to identify patterns in the sender's communication history.
[0070] The language analysis operations performed on the extracted data may incorporate different analytical techniques. These techniques may include natural language processing, semantic analysis, contextual analysis, or machine learning-based approaches. The sentiment analysis may consider factors such as word choice, punctuation patterns, emoji usage, or communication timing.
[0071] The validation of reply instances may involve multiple verification steps. These steps may include checking for consistency with previous communications, verifying technical accuracy, ensuring appropriate tone, and confirming compliance with any applicable communication policies or guidelines. The validated reply 318 may be stored in various formats within the database 206 for future reference and analysis.
[0072] The analysis of supplementary data from storage allows for deeper contextual understanding of communications. By incorporating historical communication patterns, preferences, and behaviors stored in relation to the sender, the system can develop more comprehensive insights. This expanded context enables generation of responses that are better aligned with the established communication patterns and relationship dynamics between parties.
[0073] P24-060PCT1The analysis of previously sent replies enables the system to refine and optimize future responses. Communication patterns and tendencies from past interactions can inform the generation of new replies. This historical data helps create more contextually appropriate and personalized responses over time.
[0074] The organization of supplemental data through contextual clustering enables streamlined access to historical communication patterns. The grouping of related information based on sender context allows rapid identification of relevant prior interactions and behavioral trends. This structured approach to data organization facilitates more precise and targeted application of historical insights when generating responses.
[0075] The language analysis operations enable a more comprehensive understanding of the communication content. Natural language processing and semantic analysis techniques help interpret nuances, context, and meaning within the extracted data. This deeper linguistic comprehension allows for generating responses that better align with the intended message and communication objectives.
[0076] The sentiment analysis capabilities enable the generation of responses that align with the emotional state of the sender. The analysis considers the sender's tone, mood, and emotional context expressed within their communications. This emotional alignment allows for more natural and appropriate interactions between users.
[0077] The language detection functionality processes the linguistic characteristics present in the communications. Natural language processing techniques analyze grammar patterns, language structures, and contextual elements to determine sentiment indicators. This linguistic analysis contributes to generating contextually appropriate responses that match the sender's communication style.
[0078] The validation of reply characteristics helps maintain high standards of communication quality. Checking specific aspects of replies and their associated previous communications enables verification of accuracy and appropriateness. This validation process supports the generation of consistent and suitable responses across different communication contexts.
[0079] P24-060PCT1The storage of replies in association with their corresponding data packages enables building a comprehensive knowledge base. The linked storage of replies and data packages allows tracking of communication patterns and response histories over time. This organized storage approach facilitates retrieval and analysis of past communications to inform future responses.
[0080] P24-060PCT1
Claims
Claims1. A method of generating a reply to at least one data package, the method comprising:receiving, by a recipient, at least one data package from a sender;extracting data from the at least one data package;analysing the extracted data to determine at least one feature of the at least one data package, the at least one feature comprising sentiment data associated with the sender of the at least one data package;determining, based on the analysis, at least one reply instance associated with previous data packages, the at least one reply instance being stored in a database;determining at least one adjustment to the reply instance based on the analysis; generating the reply based on the reply instance and the at least one adjustment; andsending the reply to the sender.
2. The method of generating a reply to at least one data package according to claim 1, further comprising obtaining supplementary data of the sender, from the storage, and wherein the step of analysing the extracted data comprises analysing the supplementary data of the sender.
3. The method of generating a reply to at least one data package according to claim 2, wherein the supplemental data of the sender comprises data associated with one or more previously sent replies.
4. The method of generating a reply to at least one data package according to claim 2 or claim 3, wherein the supplemental data is aggregated by clustering the supplemental data based on contextual data associated with the sender.
5. The method of generating a reply to at least one data package according to any previous claim, wherein analysing the extracted data comprises performing at least one language analysis operation on the extracted data.P24-060PCT16. The method of generating a reply to at least one data package according to any previous claim, wherein the at least one language analysis operation comprises determining the sentiment data associated with the sender of the data package, based on at least a part of the extracted data.
7. The method of generating a reply to at least one data package according to any previous claim wherein determining the at least one reply instance comprises validating at least one characteristic of the reply instance or the associated previous data packages.
8. The method of generating a reply to at least one data package according to any previous claim further comprising storing the reply in the database in association with the data package.
9. A system for generating at least one reply to at least one data package, the system comprising:storage for storing at least a database of previous data packages received from a sender and corresponding replies sent; anda processor for processing the data package, the processor comprising:a data extraction module configured to extract data from at least one data package sent from a recipient;an analysis module configured to analyse the extracted data and determine at least one feature of the at least one data package, the at least one feature comprising sentiment data relating to the sender of the at least one data package;a reply generation module for:determining a reply instance in the database of previous data packages based on the analysis;determining at least one adjustment to the reply instance based on the analysis; andgenerating the at least one reply based on the reply instance and the at least one adjustment; andan input-output device for:receiving the at least one data package from the sender; and sending the at least one reply to the sender.P24-060PCT110. A computer -readable storage medium, storing instructions that, when executed by a processor, cause the processor to generate a reply to at least one data package, the instructions comprising:receiving, by a recipient, at least one data package from a sender;extracting data from the at least one data package;analysing the extracted data to determine at least one feature of the at least on data package, the at least one feature comprising sentiment data relating to the sender of the at least one data package;determining, based on the analysis, at least one reply instance associated with previous data packages, the at least one reply instance being stored in a database;determining at least one adjustment to the reply instance based on the analysis; generating the reply based on the reply instance and the at least one adjustment; andsending the reply to the sender.P24-060PCT1