Parallel computing categorization process

Abstract

To provide a computer-implemented method for parallel computing categorization, a program, and an information processing apparatus.SOLUTION: A method comprises: obtaining real-time text data relating to a matter, the text data comprising a plurality of portions of information; performing a categorization process, the categorization process being configured to run a plurality of threads in parallel, each thread of the plurality of threads acting on one portion of information at a time; obtaining a sentiment score on the basis of the portion of information using a Sentiment Analysis machine learning (ML) model for each thread; assigning a category to the matter on the basis of the sentiment score using a classification ML model trained on historical data; updating a live category on the basis of the category assigned to the matter; and outputting the live category to a user in real-time.SELECTED DRAWING: Figure 8

Description

[Technical Field]

[0001] FIELD OF THE INVENTION The embodiments of the invention described herein relate to parallel computing categorization processes, and in particular to computer-implemented methods, computer programs, and information processing apparatus. [Background technology]

[0002] Categorizing matters based on calls and / or messages is crucial for efficient communication management across various domains. For example, in customer service, categorization helps improve inquiry flow and ensures that inquiries reach the appropriate department or personnel promptly. By labeling calls based on the nature of the inquiry, companies can effectively prioritize and allocate resources, resulting in improved customer satisfaction and resolution times. In another example, emergency calls are often categorized by severity to prioritize responses and allocate resources effectively. While the exact categorization system can vary depending on the emergency response organization and country, severity levels are generally assigned based on the perceived urgency and potential threat to life or property. Emergency dispatchers or call takers often use protocols and guidelines provided by their organizations to assess the seriousness of each call and assign the appropriate response level. They may ask specific questions to gather more information and make a quick assessment of the situation. Overall, effective categorization of matters based on calls and / or messages is essential to increase efficiency, responsiveness, and overall service quality across a variety of industries and sectors. Summary of the Invention [Problem to be solved by the invention]

[0003] Improved methods for categorization are desirable. [Means for solving the problem]

[0004] According to one embodiment, a computer-implemented method is disclosed herein that includes obtaining real-time text data related to an item, the text data including a plurality of information portions; and performing a categorization process, the categorization process configured to execute multiple threads in parallel, each thread operating on one information portion at a time, each thread performing the steps of: (i) obtaining a sentiment score based on the information portion using a sentiment analysis machine learning (ML) model; (ii) assigning a category to the item based on the sentiment score using a classification ML model trained on historical data; and (iii) updating a live category based on the category assigned to the item; and outputting the live category to a user in real time. [Brief explanation of the drawings]

[0005] Embodiments of the present invention will now be further described, by way of example only, with reference to the accompanying drawings, in which: [Figure 1] 1 illustrates a prior art process for categorizing emergency calls. [Figure 2] 1 illustrates an example of a real-time automated dynamic event categorization framework using parallel language processing, according to some embodiments of the present invention. [Figure 3] 1 illustrates an exemplary conceptual interface of an operator-viewed visualization dashboard (unfilled), according to some embodiments of the present invention. [Figure 4] 1 illustrates an exemplary conceptual interface of a visualization dashboard (filled) viewed by an operator, according to some embodiments of the present invention. [Figure 5]1 illustrates an example of a Named Entity Recognition (NER) component within a Parallel Event Categorization (PEC) module, according to some embodiments of the present invention. [Figure 6] 1 illustrates an example of a sentiment analysis component within a PEC module, according to some embodiments of the present invention. [Figure 7] 10 illustrates an example of a classification component within a PEC module, according to some embodiments of the present invention. [Figure 8] 1 is a flowchart according to some embodiments of the present invention. [Figure 9] 1 illustrates an apparatus according to some embodiments of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0006] Overview Embodiments of the present invention provide a Real-Time Automated Dynamic Event Categorization (rt-ADEC) framework using Parallel Language Processing (PLP), which handles event categorization using a parallelized framework with several layers of supervised machine learning processes to assign categories to items. Embodiments of the present invention use a parallel computing pipeline to categorize items (e.g., events, incidents, problems, queries, and inquiries) using historical data and classification-based algorithms on text-based data. Categories can be updated and changed as more information is acquired in real time. One application of some embodiments of the present invention is to support the management of emergency services resources by providing tools to support emergency dispatchers for better needs categorization. The present invention is described below primarily with reference to this application. However, embodiments of the present invention focus on determining the importance and urgency of tasks based on historical data and real-time text data (which may be speech-to-text data), which can have applications in a variety of domains beyond emergency services. Embodiments of the present invention may be used equally for any industry or sector that would benefit from categorizing items based on incoming text data in real time, since the text data can relate to any item, the categorization can relate to any set of categories, and the historical data can relate to any context.

[0007] Categorizing calls and / or messages is important across a variety of industries and sectors, including: 1. Customer ServiceIn businesses, categorization is key to efficiently managing customer inquiries. By sorting calls and messages into categories based on topic and / or urgency, companies can ensure that each query is directed to the appropriate department or representative, resulting in faster resolution times and improved customer satisfaction. 2. Emergency Services In emergency response systems, call categorization plays a vital role in effectively prioritizing and dispatching resources. Calls can be categorized based on the severity or urgency or type of emergency. This allows dispatchers to quickly allocate the appropriate personnel and equipment to address the situation. 3. Healthcare In healthcare settings, categorizing patient calls and messages helps healthcare staff prioritize and address inquiries and concerns. Calls may be categorized based on urgency, medical condition, or department, allowing healthcare providers to provide timely and appropriate care. 4. Technical Support In the technology industry, categorizing support calls and messages based on urgency, topic, or issue allows support teams to troubleshoot and resolve issues efficiently. This categorization ensures specialized support personnel are assigned to handle specific types of technical issues, resulting in faster resolution times and improved customer satisfaction.

[0008] Overall, categorizing matters based on calls and messages is important to optimize communication management, improve response times and ensure efficient resource allocation across various sectors.

[0009] One example of an immediate technical challenge relates to traditional methods used by emergency services for categorizing the severity of incoming calls. This process typically involves manual steps in which a call handler engages with the caller, gathers information using verbal prompts, and then subjectively categorizes the emergency call into various severity levels based on their own experience and training. This method is deficient in several ways: it relies on limited and potentially biased information, introduces subjectivity into the categorization process, and often results in inefficient resource allocation that costs emergency services significant amounts of money. As a result, incidents that may sound similar can be very different in nature, and relying on subjective judgment can result in over- or under-allocation of resources. This challenge highlights the need for more sophisticated, data-driven solutions for optimizing emergency call categorization and resource allocation.

[0010] Embodiments of the present invention address these challenges by incorporating automation and advanced techniques, such as language processing algorithms, to improve the accuracy of call categorization. By moving away from subjective assessment and incorporating data-driven insights, the system aims to ensure that the most appropriate resources are allocated to each emergency call, improving the overall efficiency and effectiveness of emergency services and ultimately enhancing public safety. When applied to categorizing the severity of emergency events, embodiments of the present invention provide more efficient categorization, optimizing emergency services' call handling and reducing the mental pressure on call operators who must subjectively identify severity from verbal cues.

[0011] There are four areas where technical challenges arise, which are described in more detail below.

[0012] 1. Outdated technology and cost constraintsThe current technology infrastructure used by emergency services often relies on outdated software and hardware systems. Upgrading to new technology would be advantageous for improving efficiency and accuracy in categorizing emergency calls. However, implementing these changes poses several challenges. First, it requires significant financial investments, both to acquire modern hardware and software solutions and to train personnel to use them effectively. Emergency services typically operate under budget constraints, making it difficult to justify the significant upfront costs of such technological upgrades. Furthermore, integrating new technology while ensuring minimal disruption to ongoing emergency operations is a complex and delicate process.

[0013] 2. Scalability for real-time call processing Handling large volumes of incoming emergency calls in real time requires a highly scalable infrastructure. Emergency services must be prepared to handle surges in call volume during crises or disasters. Achieving this scalability while maintaining low response times and efficient resource allocation is a significant technical challenge. This involves designing and implementing algorithms and systems that can dynamically allocate resources, prioritize calls, and ensure that response times remain within acceptable limits, even during periods of peak demand.

[0014] 3. Advanced algorithms for call categorizationDeveloping accurate algorithms for categorizing emergency calls based on urgency and seriousness requires advanced language processing and pattern recognition techniques. This challenge involves creating algorithms that can analyze the content of emergency calls, identify keywords or patterns that indicate seriousness, and assign appropriate priority levels. These algorithms must adapt to a variety of languages, accents, and dialects, and they must effectively handle both structured and unstructured data. Achieving high accuracy in call categorization is critical to ensuring that the most urgent cases receive immediate attention.

[0015] 4. Seamless Integration, Data Privacy and Security Implementing a new emergency categorization system must seamlessly integrate with existing legacy systems used by emergency services. This integration is critical to ensure a smooth transition and avoid disruptions to critical services. Furthermore, maintaining robust data privacy and security practices is paramount when handling sensitive information during emergency calls. Protecting the confidentiality and integrity of caller data while facilitating effective categorization poses complex challenges. In addition, establishing collaborative partnerships among multiple stakeholders, including government agencies, law enforcement, and healthcare providers, is critical to ensuring a coordinated and efficient emergency response system, but it involves navigating a complex regulatory and organizational landscape.

[0016] Embodiments of the present invention address the above challenges when applied to emergency call categorization by providing:

[0017] · Scalability and FlexibilityModern advanced computing interfaces provide emergency services with the ability to dynamically scale resources, ensuring efficient handling of large volumes of calls during peak hours. This scalability is made possible through virtualization technologies that abstract the physical infrastructure and allow for rapid provisioning of virtual instances. Advanced computing databases further support growing volumes of call data and ensure that the data is accessible for training machine learning models. Furthermore, the global availability of data centers ensures redundancy and fault tolerance, increasing the overall resilience of the system. These technological elements come together to enable emergency services to efficiently adapt to fluctuating call traffic and allocate resources effectively.

[0018] · Automation and Efficiency Automation in emergency call handling relies on advanced technologies such as language processing and machine learning. Language processing algorithms use machine learning models, including neural networks, to process and understand emergency call content in both audio and text format. Machine learning models are trained on historical call data, enabling the development of categorization algorithms. These algorithms are often based on supervised learning and classify calls into predefined categories based on urgency and seriousness. Furthermore, by incorporating parallel computing methods and big data analytics, real-time event categorization is developed, which subsequently optimizes resource usage and reduces response times throughout the emergency resource dispatch process.

[0019] · Real-time processing and decision making--Real-time computing plays a key role in improving the responsiveness of emergency services. Real-time data processing methods can facilitate rapid information sharing and lead to location-based decisions in a timely manner. Real-time data streams allow for immediate updates and coordination between emergency services and call centers. Large-scale data stream processing frameworks process large amounts of data in real time and perform data enrichment, categorization, and decision-making. Real-time data processing frameworks ensure that emergency services can process incoming calls promptly, make more informed decisions faster, and overcome high-volume call bottlenecks.

[0020] · Improved Data Security and Privacy Maintaining the security and privacy of caller information is a paramount concern, especially with regard to legal compliance. Technical measures typically include robust data encryption methods, such as SSL / TLS for data in transit and AES, DES, or other forms of encryption for data at rest, to ensure the confidentiality and integrity of call data. Access control mechanisms, such as role-based access control (RBAC) and access control lists (ACLs), limit access to authorized personnel and prevent unauthorized access or modification of call data. Secure data storage solutions implement techniques such as data marking, tokenization, and anonymization to protect sensitive information stored in databases or logs. These security measures align with compliance requirements for data privacy regulations such as GDPR or HIPAA, requiring rigorous auditing, monitoring, and data anonymization practices. Overall, these technical elements combine to ensure the secure and compliant handling of caller information and instill trust in emergency services.

[0021] The following non-exhaustive, illustrative list explains some technical terms:

[0022] (Near) real-time: Refers to the ability of a computer system, process, or application to respond instantly or with minimal delay to events or inputs as they occur in the physical world.

[0023] · Emergency Services: Refers to organizations and agencies responsible for responding to emergency situations, including situations requiring medical, fire, or police assistance.

[0024] Public Safety Answering Point (PSAP): An emergency call center responsible for receiving and processing emergency calls, determining the appropriate response, and dispatching necessary resources.

[0025] Call Handler or Dispatcher: A trained professional at a PSAP who receives emergency calls, assesses the nature of the emergency, and gathers key information from the caller.

[0026] · Emergency call categorization: The process of classifying emergency calls based on their perceived seriousness, ranging from life-threatening situations to non-emergency incidents.

[0027] Automated Speech Recognition (ASR): A technology that converts spoken language into written text. Often used to transcribe emergency calls.

[0028] Natural Language Processing (NLP): A branch of artificial intelligence that focuses on enabling computers to understand, interpret, and generate human language.

[0029] · Parallel computing: A type of computing in which many calculations or processes are performed simultaneously, improving the efficiency and speed of data processing.

[0030] · Caller Type: Classification of the person making the emergency call, for example, caregiver or elderly.

[0031] Risk Level: An assessment of the potential danger or seriousness associated with an emergency situation.

[0032] Conversation Metrics: Measures related to the content and structure of the conversation between the caller and the call handler, including word count, turn length, and / or words per minute.

[0033] Incident Resolution: Actions taken by emergency responders to address and mitigate an emergency, which may include providing medical care, ensuring public safety, or other interventions.

[0034] · Quality Assurance and Review: The process of evaluating and assessing the handling of each emergency call to ensure adherence to protocols and identify areas for improvement.

[0035] · Patient Identifiers: Information used to distinguish individuals who frequently or urgently require emergency services, often used in healthcare emergency situations.

[0036] · Grading System: A method used by emergency services to assess the priority and seriousness of an incident, often based on factors such as time, location, and caller information.

[0037] · Central Graded Response Manual: A set of guidelines and protocols used by emergency services to categorize and respond to incidents based on law and predetermined criteria.

[0038] Emergency Medical Dispatcher: A specialized call handler responsible for assessing medical emergencies and providing instructions or dispatching medical resources.

[0039] · Out-of-hospital cardiac arrest: A medical emergency in which a person's heart stops beating while outside a medical facility and requires immediate medical attention.

[0040] Discourse asymmetry: refers to the uneven distribution of conversational control and status between the call-taker (dispatcher) and the caller during an emergency call.

[0041] · Colloquial communication: informal, everyday language used in everyday conversation, as opposed to formal or institutional discourse.

[0042] Institutional Discourse: The formal, structured language used by emergency call-takers and institutions when interacting with callers during an emergency situation.

[0043] The following is a typical process emergency services follow to handle calls and dispatch resources, with the area of ​​focus being classification of the severity of the incident. While this process will vary depending on the country, jurisdiction, and law, a typical process is described as follows:

[0044] 1. Receiving an emergency call: a. The process begins when an individual places an emergency call to a designated emergency number such as 999 (UK), 911 (USA), etc. b. The call is routed to a Public Safety Answering Point (PSAP), also known as an emergency call center, based on the caller's location or the type of emergency service needed.

[0045] 2. Call triage and call handler assignment: a. At a public safety response center, the call is received by a call handler or dispatcher who assesses the nature of the emergency. b. Information gathering begins with the call handler asking a series of standardized questions to prompt for key details about the situation, which may include the caller's location, the type of incident, and any injuries or dangers involved. c. Call handlers categorize calls based on their perceived seriousness, which may range from life-threatening emergencies to non-emergency situations.

[0046] 3.Location Identification: a. In parallel with evaluating the call, call handlers use a variety of tools and techniques to pinpoint the caller's location, which may involve GPS data from the caller's mobile phone, location-based services, or information provided by the caller.

[0047] 4. Resource dispatch: a. Once the call is categorized, the call handler determines the appropriate response and resource allocation. b. Emergency services, including police, fire, or medical personnel, will be dispatched to the location of the incident based on the information gathered and the seriousness of the situation. c. For complex incidents, multiple units and specialized teams may be dispatched to handle different aspects of the emergency.

[0048] 5. Continuous communication: a. Throughout the process, the call handler maintains communication with the caller to provide instructions, reassurance, and updates on the status of the response. b. Dispatchers coordinate and relay information between emergency responders, ensuring that they have the details they need to effectively deal with the situation.

[0049] 6. Real-time updates and monitoring: a. Emergency services use real-time tracking and mapping systems to monitor the location and status of response units, which helps optimize response times and resource allocation. b. Dispatchers may continually assess the situation and adjust resource deployment as needed based on changing conditions.

[0050] 7. Incident Resolution: a. Emergency responders arrive at the location of the incident and, according to their training and protocols, take appropriate action to mitigate the emergency, provide medical care, ensure public safety, or address the crisis. b. The call handler or dispatcher will remain in contact with the responder and provide updates and guidance as needed until the incident is resolved.

[0051] 8. Post-incident documentation: a. After the incident is resolved, detailed documentation is prepared, including call logs, incident reports, and medical or legal documentation, if any, as required. b. This documentation is important for legal and administrative purposes and can be used for debriefing and continuous improvement.

[0052] 9. Quality Assurance and Review: Many emergency services conduct quality assurance reviews to evaluate the handling of each emergency call. This process ensures adherence to protocols and identifies opportunities for training and improvement.

[0053] An example 100 of this current process of emergency call categorization is shown in FIG. 1. In FIG. 1, a first call handler 103 first asks the caller what type of emergency service they need. The caller then responds with the required service (e.g., an ambulance). The call is then redirected to a second call handler 106, who asks the caller a series of prompts 108 to obtain information about the emergency (e.g., "What is your emergency?", "Are you still breathing?"). The second call handler subjectively categorizes the emergency based on the caller's responses and their own past experience. If appropriate, emergency services 110 are dispatched, and the second call handler notifies the caller that emergency services have been dispatched and remains online until emergency services arrive on scene.

[0054] Embodiments of the present invention are directed to, among other things, addressing the problem of categorizing the severity of emergency calls (although, as noted above, embodiments of the present invention may be applied to any industry or sector). Emergency service dispatch centers handle a significant influx of calls, especially during large-scale events—situations that can cause bottlenecks in processing and resulting delays. Furthermore, the decision to dispatch resources or not is heavily dependent on the individual experience and inherently subjective judgment of the dispatcher. This reliance on dispatcher subjectivity introduces an element of ambiguity and the potential for miscommunication, which can significantly increase the likelihood of incorrect information being shared.

[0055] As a direct result of this inherent subjectivity, incidents are often misclassified as to their severity. This misclassification, in turn, leads to two distinct and problematic scenarios: over-resourced and under-resourced. In the former, excessive resources are allocated to a particular incident, potentially resulting in a depletion of available resources for other emergencies. Conversely, under-resourced occurs when ill-equipped or inappropriate resources are dispatched to handle a given situation, resulting in an inadequate emergency response, which can have direct impacts on people's health and well-being.

[0056] Added to these challenges is the emotional state of callers during a crisis. Often, callers experience shock and panic, and their initial assessment of the severity of the incident is likely to fluctuate over the course of the call. It is against this background that the implementation of a robust, adaptive system capable of real-time severity categorization is paramount. Such a system has the potential to proactively mitigate the aforementioned issues by providing a dynamic assessment that adapts to the evolving situation, thus improving the accuracy and efficiency of emergency resource allocation and response.

[0057] Various aspects and details of these key concepts are described below, by way of example only, with reference to Figures 2-9.

[0058] 8 is a flowchart illustrating method steps in a computer-implemented method 800. The method includes steps 802-812.

[0059] In step 802, real-time text data regarding a matter (e.g., an event, incident, problem, query, inquiry, emergency incident) is obtained. The text data includes multiple information portions. In some examples, the real-time text data may be converted from real-time voice data, for example, if the source of the text data is a telephone conversation (e.g., an emergency call), the voice data may be converted to text data. In other examples, the source of the text data may be a text-based conversation from an online chat, text message, email message, etc. The multiple information portions may refer to individual messages within the conversation, for example, responses to an operator.

[0060] In step 804, a categorization process is performed. The categorization process may be performed by a Parallel Event Categorization (PEC) module described herein. Each conversation may have its own instance of the PEC module. The categorization process is configured to run multiple threads (Thread 1, 2, 3, ... N) in parallel. Each thread operates on one piece of information (e.g., a message) at a time. In other words, multiple threads each take one message at a time, so multiple messages can be analyzed at once. This means that longer messages can be analyzed by one thread while other threads process shorter messages. Longer messages take longer to analyze than shorter messages. Parallelization speeds up the analysis so that the conversation can be analyzed in real time. This is especially important for live conversations such as phone calls or online chats.

[0061] In step 806, each thread uses a sentiment analysis ML model to obtain a sentiment score (or equivalently, a sentiment label) based on the information portion (e.g., message). The sentiment analysis model may be an off-the-shelf sentiment analysis model. Like NER models, sentiment analysis models may be location-specific, e.g., UK-based, US-based, global, etc.

[0062] In step 808, each thread assigns a category (e.g., a severity category, an urgency category, a subject matter category) to the item based on the sentiment score using a classification model trained on the historical data. The classification model may be trained on the historical data by obtaining a historical database of the historical data that includes multiple pieces of text data, each labeled with a category, and training the classification model based on the historical database.

[0063] Classification and / or sentiment analysis models can be trained based on domain knowledge, i.e., a set of categories appropriate to the context. For example, in an emergency services application, a classification model may be trained using both historical text data and domain knowledge of incident categorization standards from emergency services. The standard, for example, ISER for police, where I is immediate (high urgency), S is significant (low urgency), E is extended (non-urgent), and R is referred (no attention required). In a customer service application, a classification model may be trained using both historical text data and domain knowledge categorization standards within the enterprise.

[0064] In step 810, each thread updates the live category based on the category assigned to the item. In other words, there is a live category variable that is updated based on the category assigned by the thread. As each thread finishes analyzing a message, the live category is updated with the latest analysis. Thus, it is always the latest category that is output to the user in step 812. The output may be displayed to the user via a GUI.

[0065] Optionally, before step 806, each thread may further perform the step of using a Name Entity Recognition (NER) ML model to retrieve one or more keywords from the information portion. The NER ML model may be a pre-built NER model. The NER model may be location-specific, e.g., UK-based, US-based, global, etc. The NER model may be trained on in-house data; for example, if used in a police context, the NER model may be trained on police data. These keywords are then used as input to a sentiment analysis ML model and / or a classification ML model. Similar to the live categorical variables described above, the keywords may be used to update a live keyword output, and the method may further include outputting the live keyword output to a user in real time via a GUI. For example, the keywords can be used to highlight important information in a transcript window, as described below.

[0066] It should be noted that embodiments of the present invention are described below in the context of emergency service calls. However, the present invention is not limited to this application, and those skilled in the art will appreciate that embodiments of the present invention may be equally used for any industry or sector that benefits from categorizing items based on incoming text data in real time.

[0067] For example, with reference to an embodiment of the present invention, real-time text data may include data from calls / messages from customers needing urgent help. The matter in this instance may be a problem. The pieces of information may be each message from the customer. The categorization may relate to the level of urgency and / or topic of the problem. The historical data used to train the classification model may relate to historical customer queries labeled with categorizations.

[0068] In another example, the real-time text data may include data from an emergency call / message from a caller needing urgent help from emergency services (e.g., ambulance, fire, police, coast guard, military, etc.). The matter in this example may be the emergency event / situation. The pieces of information may be each response from the caller to the operator's prompts. The categorization may be related to the seriousness of the situation. The categorization may follow a categorization standard, for example, the police standard is ISER, where I is immediate (high urgency), S is significant (low urgency), E is extended (non-urgent), and R is referred (no attention required). The historical data used to train the classification model may relate to historical text data about emergency service calls labeled with categorizations.

[0069] In another example, the real-time text data may include data from calls / messages from callers seeking healthcare assistance. The matter in this instance may be an inquiry. The pieces of information may be respective responses from the caller to the operator's prompts. The categorization may relate to, for example, urgency and / or medical condition. The categorization may be in accordance with categorization standards for the healthcare facility. The historical data used to train the classification model may relate to historical text data regarding healthcare inquiries labeled with categorizations.

[0070] In another example, the real-time text data may include data from calls / messages from a person needing technical support. The matter in this example is a technical issue. Multiple pieces of information may be each message from the person to a help desk system. The categorization may be related to urgency and / or topic, for example. The categorization may follow categorization standards for the help desk system. The historical data used to train the classification model may involve historical text data related to technical issues and labeled with categorizations.

[0071] An optional NER ML model can take (a portion of) real-time text data as its input and output keywords from the text data.

[0072] A sentiment analysis ML model can take as its input (a portion of) real-time text data and, optionally, keywords from an NER model, and output a sentiment score (or sentiment label).

[0073] The classification ML model may take as its input sentiment scores (from the sentiment analysis model) and, optionally, keywords (from the NER model), and outputs categories.

[0074] The sentiment analysis model and classification model (and optionally the NER ML model) together form a thread. Multiple threads form a Parallel Event Categorization (PEC) module that performs the categorization process. Parallel threads allow multiple parts of a conversation to be analyzed at once, resulting in faster categorization that can be updated in real time as the conversation continues.

[0075] In some examples, the real-time text data is derived from real-time voice data (e.g., from a call, e.g., an emergency call), which is converted into real-time text data using a speech-to-text converter.

[0076] In some examples, the categories of items are output to the user in real time via a GUI, such as the GUI shown in Figures 3 and 4 and described below.

[0077] Figure 2 illustrates concepts according to an embodiment of the present invention in the context of an exemplary application of an emergency service call. Figure 2 provides a detailed workflow of how an emergency call is processed using an embodiment of the present invention. Please refer to Figure 2.

[0078] The process begins with an incoming emergency call received at the appropriate emergency services call center. The call center operator begins the conversation 202 by using general prompts to gather key information. These prompts are intended to understand the nature of the emergency, the caller's location, and the type of incident being reported.

[0079] Next, each response provided by the caller is channeled through a robust data security protocol 204. This protocol serves the important purpose of tagging the data for identification and security purposes. It acts as a safeguard to ensure the integrity and privacy of the information shared during the conversation 202.

[0080] After passing through data security protocols 204, the tagged data is sent to a cloud infrastructure 206. This cloud-based system 206 is designed to provide the necessary computing power, storage capacity, and flexibility required for real-time data processing and analysis for many calls at once (tagged messages are appropriately routed to the same parallel event categorization instances).

[0081] Within the cloud infrastructure 206, each message, e.g., MSG:1, is directed to a parallel event categorization (PEC) module instance 210. This module, a component of the system 200, facilitates the concurrent processing of messages. A parallel algorithm is employed, which simultaneously launches multiple threads, N, to concurrently classify each message. There is one PEC module instance per incoming call. The PEC module handles the parallelization of the stream of messages coming from the incoming call. This parallelization is important for efficiently analyzing messages from the incoming call. Parsing messages in parallel is advantageous because messages (i.e., text data) vary in length, and short messages take less time to parse than long messages. By having parallelization, the process is not bogged down by the analysis of long messages. While one thread of the PEC module analyzes a longer message, other threads of the PEC module can process multiple, shorter messages.

[0082] Once messages 208 are classified, the system 200 dynamically assigns a category 212 to each message in real time. These categories 212 are dynamic and change as new information comes in from the caller's responses and the ongoing conversation 202. The evolving categories 212 are displayed 214 to the operator in real time, ensuring that the categories are continually updated with the severity of the incident as it unfolds. Here, m(t) represents the severity category at time t, and S represents the superset of all possible severity categories.

[0083] Once the operator is confident in the identified severity, a decision is made regarding the dispatch of relevant resources. This decision may be informed not only by real-time categorization, but also by historical incident data and the current conversation with the caller. The timely dispatch of appropriate resources is a key outcome of this system, optimizing emergency response.

[0084] Following resolution of the incident, the conversation data undergoes another round of processing through data security protocols 204. This processed data is then securely labeled and stored in a historical text database 216. This archiving of data serves multiple purposes, including compliance with data retention regulations, potential future reference, training for ML models (e.g., classification ML models), and analysis for continuous improvement of emergency response procedures.

[0085] An embodiment of the present invention uses parallel language processing techniques that can be used to significantly improve the management of high volumes of emergency telephone calls to emergency lines in real time. Furthermore, it can reduce the pressure on call handlers who must assess the seriousness of a situation while conversing with distressed callers. Below are key points about how this technology can solve the problems already mentioned above.

[0086] Language processing algorithms enable the extraction of key information from emergency calls by thoroughly analyzing various components. This includes decoding the caller's verbal description, speech recognition to translate speech to text, and semantic understanding to grasp context. Through this sophisticated analysis and training on a large archive of previous conversation data, the system can accurately categorize incoming calls based on their urgency and seriousness, in real time. This not only expedites the response process but also ensures that each call receives the appropriate level of attention.

[0087] By leveraging cloud computing technology, the system gains substantial scalability and flexibility. Emergency services often deal with sudden surges in call volume, especially during crises. A cloud-based infrastructure provides the necessary computing power and storage capacity to process and analyze these incoming calls in real time. This means the system can seamlessly adapt to fluctuations in demand and ensure a consistently efficient response, even during high-call-volume scenarios.

[0088] Recognizing that the pace at which callers communicate their issues can vary significantly, incorporating parallel computing becomes important. By parallelizing the event categorization module, the system can process multiple messages for the same call simultaneously, rather than processing one message at a time and updating classification metrics. This parallelization optimizes resource allocation and allows for faster processing of calls in an automated manner. The result is reduced strain on operators, faster response times, and faster delivery of emergency services, especially important in situations where every second counts.

[0089] The integrity and security of caller information are of paramount importance. Robust security measures, including the latest encryption technologies, strict access controls, and secure data storage protocols, are key components of handling sensitive data. These measures not only protect sensitive caller information from unauthorized access, but also ensure strict compliance with stringent data privacy regulations. Caller trust and confidentiality are maintained throughout the emergency call handling process, instilling confidence in both callers and emergency service providers.

[0090] Below is a detailed description of each stage of FIG. 2 illustrating an embodiment of the present invention in the exemplary context of an emergency call.

[0091] Step 1. The emergency operator transfers the incoming call to the requested service operator (police, ambulance, or fire).

[0092] Step 2. The service operator will start the prompt "What is your emergency" and the caller will begin to explain the problem.

[0093] Step 3. The incoming response is converted from speech to text.

[0094] Step 4. The converted response is sent to the DSP (Data Security Protocol) module, which performs the following steps: a. The transformed response is tagged with a call_ID which is a unique 12 character identifier, e.g. 172937194827, and a message_number which is the order in which the response comes into the system (the first one is 1). b. The tagged response, response(172937194827,1,RESP_MSG), goes into an entity recognition module where personal information such as names is anonymized. Note that RESP_MSG is the raw text response from the caller. c. Now there are two response messages: response(172937194827,1,RESP_MSG) and response_anonymised(172937194827,1,RESP_MSG_ANON). Response_anonymised is sent to a historical text database for recording, and response is sent to a cloud computing infrastructure. Here, RESP_MSG_ANON is the anonymised version of the caller response.

[0095] Step 5. The response_anonymized text data is used as a record in the historical text database (unlabeled). The tagged message response is sent to the cloud infrastructure, which performs the following steps: a. response(172937194827,1,RESP_MSG) goes to the real-time message intake module, which stores the message waiting for further processing.

[0096] b. The cloud infrastructure can handle multiple incoming response(call_ID, message_number, message) variables and submit these responses in parallel to the event categorization instance.

[0097] Step 6. Parallel Event Categorization The framework parallelizes the incoming response and performs the following process: a. RESP_MSG is extracted from the response variable and used as input to a trained Name Entity Recognition (001) process that highlights keywords such as location, name, and type of emergency for each parallel thread. RESP_MSG is used to identify the message "WE HAD A CAR ACCIDENT, MY HUSBAND MARC IS NOT BREATHING, WE ARE IN WINDSOR " [ car My husband was in an accident. mark but I'm not breathing , Windsor These keywords are then compiled into a data structure x = {[172937194827,1]:[CAR,MARC,NOT BREATHING,WINDSOR], …,[172937194827,N]:[, …,]}, and new incoming messages are appended to x, with messages appended in parallel. b. Each RESP_MSG, in this case = "WE HAD A CAR ACCIDENT, MY HUSBAND MARC IS NOT BREATHING, WE ARE IN WINDSOR", is used as input to a Sentiment Analysis Instance (002) that tags the message with a sentiment, in this case y=Negative (because the content of the message has a negative sentiment - husband not breathing after car accident), the algorithm may have been pre-trained using both domain knowledge from emergency services and labeled data from past history. c. The final step involves a classification process (003), which may be trained using both historical text data and domain knowledge of incident categorization standards from emergency services. The categorization standard, for example in police, is I,S,E,R, where I is immediate (high urgency), S is significant (low urgency), E is extended (non-urgent), and R is referral (no attention required). The inputs for the classification module are x, y, and response->X = (x, y, response) for each thread. The output from the module is (timestamp, category, confidence, call_ID), which is fed into a visualization dashboard (004), allowing dispatchers to see in real time how the Automated Dynamic Event Categorization (ADEC) framework categorizes the call; this categorization process can change as the conversation continues. Given that people can subjectively perceive incidents differently, sometimes a call may sound like a Category 1 high urgency event, but in reality, as the conversation continues, this may not be the case.

[0098] Step 7. Once the call is processed, the compiled data is labeled with severity and sent through data security protocols to a historical database for training tasks.

[0099] The dashboard 300 shown in FIG. 3 illustrates an exemplary GUI and describes an exemplary window an operator might observe while handling an emergency call. It has four sections: a live window 302, a transcript window 304, a raw analysis results window 306, and a live categorization window 308. The live window 302 highlights a 12-character identifier 310 and the call's originating location 312. A red circle 314 indicates an operator is on the call. The transcript window 304 displays a live speech-to-text transcription of the ongoing conversation for the operator and caller. The raw analysis results window 306 displays aggregated results for urgency based on confidence (a reliability metric based on the conversation and a trained model), sentiment, and severity score. The live categorization window 308 displays a pie chart 316 or any other easy-to-read visualization that displays the severity score for the ongoing call in real time. On the right side is a quantitative description 318 of this severity score. The operator can also terminate the call using the "End Call" button 320 or redirect the call to another operator using the "Redirect Call" button 322, where the current accumulated information is delivered to the other operator. In Figure 4, the visualization dashboard is filled with mock data (similar reference numbers are used).

[0100] The parallel event categorization (PEC) module 500 is shown in FIGS. 5-7 and performs step 6 described above. The PEC module includes N parallel threads, e.g., 502a, 502b, and 502c. For each parallel thread, N=X, where N is the number of threads and X is the number of messages currently in the module. Each parallel thread, e.g., 502a, 502b, and 502c, includes a sentiment analysis component 506 and a classification component 508 (and optionally a named entity recognition (NER) component 504), as described below.

[0101] The optional NER component 504 goes through several stages of data wrangling, analysis, and result extraction. Each stage is described on the right side of FIG. 5. Some of these components are interchangeable and may be removed or replaced with other methods depending on the domain in which they are used. The optional use of the NER component 504 is described in connection with step 6a above. More specifically, referring to FIG. 5, the first stage of the entity recognition component is to receive / acquire text input 510 (e.g., the RESP_MSG described above). Next, in the tokenization stage 512, the text input is broken down into smaller tokens. Next, in the feature extraction stage 514, relevant features are extracted from each token. Next, in the ML model stage 516, an ML model trained on annotated data, such as people's names and locations, is used on the extracted features. The output of the ML model is then the “NER output” 518, where each corresponding token is labeled, e.g., B-PERSON, I-PERSON (e.g., the beginning and end of a person's name), to track who said what in the conversation and in what order those tokens were said. For example, the first token may be labeled as said by the operator, and the next token may be labeled as said by the caller. Next is the post-processing stage 520, where consecutive tokens with the same entity label (e.g., location) are merged to form a complete named entity. For example, if “New” and “York” are tagged as [B-LOC, I-LOC], they are merged into a single entity, “New York.” The final output of the NER process (“NER result”) 522 is a list of identified named entities (keywords) and their respective categories (e.g., [CAR, MARC, NOT BREATHING, WINDSOR], as in the example of step 6a above).Within each thread, the output (keyword(s)) of the NER process 504 may be used to update a live keyword output, which is then output to a user in real time via a GUI (e.g., the GUI shown in Figures 3 and 4). For example, the live keyword output may be used to highlight keywords in the transcript window 304.

[0102] Once the NER process 504 is complete, the results may then optionally be forwarded to the sentiment processing phase 506 (FIG. 6). The sentiment processing phase 506 does not necessarily require the NER results and can take text data as its only input. However, by inputting the NER results into the sentiment analysis component 506, more accurate results can be achieved. The stages of the sentiment processing phase are described on the right side of FIG. 6. The use of the sentiment analysis component 506 is described in connection with step 6b above. More specifically, with reference to FIG. 6, the first stage of the sentiment analysis component 506 is to receive / acquire text input 602, which includes the original message text, e.g., RESP_MSG, discussed above, and optionally includes the output from the NER component 504. Next, the text input is preprocessed (604). The preprocessing phase may include a tokenization process, converting the text to lowercase, and / or removing punctuation. The sentiment analysis model 606 then operates on the preprocessed text. The sentiment analysis model 606 is a machine learning algorithm trained on labeled data (a corpus with emotion labels) from historical phone conversations. Suitable machine learning algorithms include support vector machines (SVMs), naive Bayes, recurrent neural networks (RNNs), and transformers (e.g., BERT). The model learns to recognize emotion patterns and make predictions, thereby learning to identify the correct emotion given new input from real-time conversations. The emotion is then classified by an emotion classification model, which outputs a classification determining whether the emotion is positive, negative, or neutral. An emotion score is then assigned indicating the strength of the expressed emotion, which in turn influences the severity output. The emotion score may be a normalized value between −1 and +1, with −1 representing the most negative emotion (e.g., very upset, very panicked), 0 representing a neutral emotion, and +1 representing the most positive emotion (e.g., harmonious conversation). Similarly, the sentiment score may be a normalized value between 0 and 1, with 0 being most negative and 1 being most positive.The output of the sentiment analysis component is an output that may include a sentiment label (positive, negative, or neutral) and / or a sentiment score. Sentiment scores and sentiment labels can be easily converted between each other—sentiment scores are a numerical form of sentiment labels. Sentiment analysis can help inform the model how quickly to take action. For example, if the caller is very panicked and therefore has a very negative sentiment score, the model may suggest to the user to take action (e.g., dispatch emergency services) more quickly than if the caller were calm and had a more positive sentiment score (thus basing the classification on less information because the conversation would be shorter). The main purpose of sentiment analysis is to distinguish very negative scenarios compared to positive scenarios (although in the system's application to emergency service calls, there are unlikely to be any positive calls). If the majority of messages from an incoming call are negative in sentiment, this will affect the classification score; i.e., given that the majority of messages were labeled with negative sentiment, the call is likely to be classified as very urgent).

[0103] The classification component 508 is shown in FIG. 7. Training of the classification component 508 will first be described with reference to FIG. 7. The first step in training the classification component is receiving / acquiring historical data 702. The classification component is initially trained using the latest version of the historical database 702 of conversations labeled with severity metrics (step 7 above). The historical data 702 from emergency calls is parsed. The text from the historical data 702 is then preprocessed (704) to clean and prepare the text data and make it suitable for analysis. The text data then undergoes feature extraction 706 to convert the data into numerical features so that a machine learning algorithm can manipulate the data. The dataset is then split (708) into a training set, a validation set, and a test set. A machine learning model is then selected (710) (e.g., a Naive Bayes model, a logistic regression model, etc.) and trained (712). The performance of the model is then assessed / evaluated on the test set using an evaluation metric (e.g., root mean square error, Pearson's correlation coefficient, T-test, etc.) (714). The trained model is then deployed (716).

[0104] The use of the classification component 508 is described in connection with step 6c above. As described in step 6c above, the classification module 508 receives as its input the output of the sentiment analysis component (and optionally the output of the NER component), referred to as "Message" 718 in Figure 7, and outputs a severity category for the currently received emergency call data. Note that some of these internal components can be replaced with other processes or algorithms, and in some cases, not all components are required, while in other cases, all processes are required. As new emergency call messages are received, they are fed into the trained model and instantly classified into severity levels in real time. These results are then used to update the visualization dashboard, an example of which is shown in Figure 4.

[0105] Worked Example: The following scenario could be played out in the UK and enforced by the Metropolitan Police.

[0106] The following steps correspond to those already described above in connection with FIG.

[0107] Step 1: Emergency Operator Transfer An emergency operator receives the call and identifies it as a request for police service. The call will be transferred to a police service operator.

[0108] Step 2: Prompt for emergency details · Police service · The operator will begin the prompt "What is your emergency?" The caller begins explaining the problem: "I witnessed a robbery in progress near Oxford Street."

[0109] Step 3: Speech-to-text The caller's spoken response, "I witnessed a robbery in progress near Oxford Street," is converted from speech to text.

[0110] Step 4: Data Security Protocol Module The transformed response is tagged with a unique call_ID (e.g., 172937194827) and message_number (1). The tagged responses go through an entity recognition module for anonymization. Two responses are generated: ○「response(172937194827,1,"I witnessed a robbery in progress near Oxford Street") response_anonymized(172937194827,1,"I witnessed a robbery in progress near [LOCATINO]"), where [LOCATION] has been anonymized.

[0111] Step 5: Data processing "response_anonymised" is sent to a historical text database for record-keeping purposes. The response is sent to the cloud computing infrastructure.

[0112] Step 6: Parallel Event Categorization Framework The cloud infrastructure processes the incoming responses and parallelizes the categorization process. For each parallel thread ○RESP_MSG in this case is "I witnessed a robbery in progress near Oxford Street". Name Entity Recognition (001) highlights the keywords "[LOCATION: Oxford Street],[INCIDENT TYPE: robbery],[WITNESS]" ○Sentiment Analysis (002) tags the sentiment as "negative." Classification (003) categorizes incidents as "I (Immediate)" based on domain knowledge and historical data.

[0113] Step 7: Visualization and final data processing The categorization output for a thread is (timestamp,category,confidence,call_ID) where "category" is "I", "confidence" is a number indicating the certainty of the output given the input data, and "call_ID" is 172937194827. This data is fed into a visualization dashboard (004) for dispatchers to monitor in real time. · Categorizations may change as the conversation continues. Once the call is processed, the compiled data is labeled with a severity (in this case, I) and sent through data security protocols to a historical database for training tasks.

[0114] This process ensures that emergency calls are efficiently categorized and that dispatcher decisions are based on real-time analysis of the caller's information. The system can adapt as the conversation progresses, providing more accurate classification and response to the emergency.

[0115] Hardware 9 is a block diagram of an information processing apparatus 10 or computing device 10, such as a data storage server, that may be used to embody the present invention, to implement some or all of the operations of the methods embodying the present invention, and to perform some or all of the tasks of the apparatus of the embodiments. The computing device 10 may be used to perform any of the method steps described above, e.g., S.802-812.

[0116] The computing device 10 has a processor 993 and memory 994. Optionally, the computing device also includes a network interface 997 for communicating with other such computing devices, such as other computing devices of invention embodiments. Optionally, the computing device also includes one or more input mechanisms, such as a keyboard and mouse 996, and one or more display units, such as a monitor or monitors 995. These elements can facilitate user interaction. These components can be connected to each other via a bus 992.

[0117] The memory 994 may include a computer-readable medium, which term may refer to a single medium or multiple media (e.g., centralized or distributed databases and / or associated caches and servers) configured to carry computer-executable instructions. Computer-executable instructions may include, for example, instructions and data accessible by a general-purpose computer, a special-purpose computer, or a special-purpose processing device (e.g., one or more processors) to cause it to perform one or more functions or operations. For example, computer-executable instructions may include instructions for implementing a method disclosed herein or any method step(s) disclosed herein, e.g., the method or any method step(s) shown in steps S.801-S.812. Thus, the term "computer-readable storage medium" may also include any medium capable of storing, encoding, or carrying a set of instructions for execution by a machine, causing the machine to perform any one or more of the method steps of the present disclosure. Thus, the term "computer-readable storage medium" may be interpreted to include, but is not limited to, solid-state memory, optical media, and magnetic media. By way of example, and not limitation, such computer-readable media may include non-transitory computer-readable storage media including random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disk storage, magnetic disk storage or other magnetic storage devices, flash memory devices (e.g., solid-state memory devices).

[0118] The processor 993 is configured to control the computing device and perform processing operations, for example, by executing computer program code stored in the memory 994 to perform any of the method steps described herein. The memory 994 stores data read and written by the processor 993, and may store the above-mentioned historical data, real-time text data, keyword data, sentiment score data, and / or category data, and / or programs for executing any of the method steps S.801-812. As referred to herein, a processor may include one or more general-purpose processing devices, such as a microprocessor, a central processing unit, or the like. The processor may include a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, or a processor implementing other instruction sets or a combination of instruction sets. The processor may also include one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), a network processor, or the like. In one or more embodiments, a processor is configured to execute instructions to perform the operations described above and elsewhere herein. The processor 993 may be considered to include any of the modules described above. Any operation described as being implemented by a module may be implemented as a method by a computer, such as by the processor 993.

[0119] The memory 994 and the processor 993 may collectively be configured to provide a parallel event categorization (PEC) module 500 configured to perform a categorization process, the PEC module 500 including multiple threads, each thread including a sentiment analysis ML model 506, a classification ML model 508, and optionally an NER ML model 504.

[0120] The display unit 995 may display representations of data stored by the computing device, such as the live categorization or recommendation and / or GUI windows described above in connection with Figures 3 and 4, and / or interactive representations that allow a user to interact with apparatus 10 as described above with reference to any of steps S.802-812, for example, by drag-and-drop or selection interactions, and / or any other output described above, and may also display cursors and dialog boxes and screens that allow interaction between a user and programs and data stored on the computing device. The input mechanism 996 may allow a user to input data and instructions into the computing device, such as allowing a user to enter any of the user inputs described above.

[0121] The network interface (network I / F) 997 may be connected to a network such as the Internet and may be connected to other such computing devices via the network. The network I / F 997 may control data input and output to and from other devices via the network. Other peripheral devices such as a microphone, speakers, printer, power supply unit, fan, case, scanner, trackball, etc. may be included in the computing device.

[0122] A method embodying the present invention may be performed on a computing device / apparatus 10 such as that shown in FIG. 9. Such a computing device need not have all of the components shown in FIG. 9, but may instead consist of a subset of those components. For example, apparatus 10 may include a processor 993 and a memory 994 connected to the processor 993. Alternatively, apparatus 10 may include a processor 993, a memory 994 connected to the processor 993, and a display 995. A method embodying the present invention may be performed by a single computing device that communicates with one or more data storage servers over a network. The computing device may also be a data storage device itself that stores at least a portion of the data.

[0123] Methods embodying the present invention may be performed by a plurality of computing devices operating in cooperation with one another, one or more of which may be data storage servers that store at least a portion of the data.

[0124] The invention may be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or in combinations of them. The invention can also be implemented as a computer program or computer program product, i.e., a computer program tangibly embodied in a non-transitory information carrier, for example a machine-readable storage device or a propagated signal, for execution by, or to control the operation of, one or more hardware modules.

[0125] A computer program may be in the form of a stand-alone program, a computer program portion, or two or more computer programs, may be written in any programming language, including compiled or interpreted languages, and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a data processing environment. A computer program may be deployed to be executed on one module or on multiple modules at one site, or may be distributed across multiple sites and interconnected by a communications network.

[0126] The method steps of the present invention may be performed by one or more programmable processors executing a computer program that performs the functions of the present invention by operating on input data and generating output. The apparatus of the present invention may be implemented as programmed hardware or as special purpose logic circuitry including, for example, an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit).

[0127] Processors suitable for the execution of a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Generally, a processor receives instructions and data from a read-only memory or a random-access memory or both. The essential elements of a computer are a processor for executing instructions coupled to one or more memory devices for storing instructions and data.

[0128] The above-described embodiments of the invention may be used to advantage independently of other embodiments or in any feasible combination with one or more other embodiments.

[0129] Various modifications, either by adding, deleting, or substituting features, may be made to the above-described embodiments to provide further embodiments, any and all of which are intended to be encompassed by the appended claims.

[0130] A list of numbered statements. [Statement 1] 1. A computer-implemented method comprising: acquiring real-time text data relating to the matter, the text data including a plurality of information portions; executing a categorization process configured to execute a plurality of threads in parallel, each thread of the plurality of threads operating on one piece of information at a time, each thread performing the following steps: obtaining a sentiment score based on the information portion using a sentiment analysis machine learning (ML) model; assigning a category to the item based on the sentiment score using a classification ML model trained on historical data; and performing the step of updating a live category based on the category assigned to the item; stages and; and outputting the live categories to a user in real time. method. [Statement 2] The method of claim 1, further comprising receiving real-time voice data relating to the item and converting the real-time voice data into real-time text data. [Statement 3] 3. The method of claim 1 or 2, wherein the item includes an event, an incident, a problem, a query, or an inquiry. [Statement 4] 4. The method of claim 3, wherein the matter includes an emergency incident. [Statement 5] 5. The method of claim 4, wherein the real-time text data is derived from an emergency call. [Statement 6] 6. The method of any one of statements 1 to 5, wherein the assigned category is a severity category indicating how serious and / or urgent the matter is. [Statement 7] 7. The method of any one of claims 1 to 6, wherein outputting the live categories to a user in real time includes displaying the live categories to a user via a graphical user interface (GUI). [Statement 8] 8. The method of any one of statements 1 to 7, wherein each thread further performs the step of obtaining one or more keywords from the information portion using a Name Entity Recognition (NER) ML model. [Statement 9] 9. The method of claim 8, wherein the NER ML model receives the information portion as its input and outputs the one or more keywords. [Statement 10] 10. The method of claim 8 or 9, wherein the sentiment analysis ML model further bases the sentiment score on the one or more keywords. [Statement 11] 11. The method of any one of statements 8 to 10, wherein the classification ML model further bases the categories on the one or more keywords. [Statement 12] 12. The method of any one of statements 8 to 11, wherein each thread further performs a step of updating a live keyword output based on the one or more keywords, and the method further includes outputting the live keyword output to a user in real time via a GUI. [Statement 13] 13. The method of any one of statements 1 to 12, wherein the classification ML model trained on historical data is trained by obtaining a historical database of historical data including multiple portions of text data, each portion labeled with a category; and training the classification ML model based on the historical database. [Statement 14] 14. The method of claim 13, further comprising labeling the text data with its assigned category and adding it to the historical database for training purposes. [Statement 15] 15. The method of any one of statements 1 to 14, wherein the sentiment analysis ML model takes the information portion (and optionally the one or more keywords) as its input and outputs the sentiment score. [Statement 16] 16. The method of any one of statements 1 to 15, wherein the classification ML model takes the sentiment score (and optionally one or more keywords) as its input and outputs the category. [Statement 17] 17. A computer program product which, when run on a computer, causes the computer to carry out a method according to any one of statements 1 to 16. [Statement 18] 17. An information processing device having a memory and a processor connected to the memory, the processor configured to perform a method according to any one of statements 1 to 16. [Statement 19] 19. The information processing apparatus of claim 18, wherein the memory and the processor are collectively configured to provide a parallel event categorization (PEC) module configured to perform the categorization process, the PEC module including the plurality of threads, each thread including the sentiment analysis ML model and the classification ML model (and optionally the NER ML model).

[0131] The following additional notes are provided regarding the embodiments including the above examples. (Appendix 1) 1. A computer-implemented method comprising: acquiring real-time text data relating to the matter, the text data comprising a plurality of information portions; executing a categorization process configured to execute a plurality of threads in parallel, each thread of the plurality of threads operating on one piece of information at a time, each thread performing the following steps: obtaining a sentiment score based on the information portion using a sentiment analysis machine learning (ML) model; assigning a category to the item based on the sentiment score using a classification ML model trained on historical data; and performing the step of updating a live category based on the category assigned to the item; stages and; and outputting the live categories to a user in real time. A computer-implemented method. (Appendix 2) 2. The computer-implemented method of claim 1, further comprising receiving real-time audio data relating to the item and converting the real-time audio data into the real-time text data. (Appendix 3) The matter may include an event, incident, problem, query, or inquiry; Preferably, the matter includes an emergency incident; Further preferably, the real-time text data is derived from an emergency call. (Appendix 4) 2. The computer-implemented method of claim 1, wherein the assigned category is a severity category indicating how serious and / or urgent the matter is. (Appendix 5) 2. The computer-implemented method of claim 1, wherein outputting the live categories to the user in real time includes displaying the live categories to the user via a graphical user interface (GUI). (Appendix 6) 2. The computer-implemented method of claim 1, wherein each thread further performs the step of using a Name Entity Recognition (NER) ML model to obtain one or more keywords from the information portion; preferably, the sentiment analysis ML model further bases the sentiment score on the one or more keywords; and more preferably, the classification ML model further bases the category on the one or more keywords. (Appendix 7) 7. The computer-implemented method of claim 6, wherein each thread further performs a step of updating a live keyword output based on the one or more keywords, and the method further includes outputting the live keyword output to a user in real time via a GUI. (Appendix 8) The classification ML model trained on historical data: obtaining a historical database of historical data including a plurality of portions of text data, each portion labeled with a category; and training the classification ML model based on the historical database; Preferably, the method further comprises labeling said text data with its assigned category and adding it to said historical database for training purposes. 8. A computer-implemented method according to any one of claims 1 to 7. (Appendix 9) A computer program product which, when run on a computer, causes the computer to perform the method described in Appendix 1. (Appendix 10) 10. An information processing device having a memory and a processor connected to the memory, the processor configured to perform the method described in claim 1. [Explanation of symbols]

[0132] 102 Classical Computing Devices 104 datasets 106 Subset Selection 108 Data Subsets 110 Quantum Computing Devices 112 Parameters 114 training parameters 116 solution 118 Threshold Solution 120 Compare Solutions 122 Adjusting quantum circuit depth Select a data subset from a 302 dataset 304 The parameters of the quantum computing system model are trained using a quantum computer over multiple iterations using a data subset and quantum circuit depth. 306 Generate solutions for those data subsets using a quantum computing system model and a quantum computer. 308 The solution is compared with the threshold solution 310 Adjusting quantum circuit depth in response to the solution of a quantum computing system model not satisfying a threshold solution 312 Retraining the parameters of the quantum computing system model using a quantum computer with the data subset and adjusted quantum circuit depth Selection of data subsets, quantum circuit depth, and threshold quality from the 402 dataset Using a quantum computer over 404 iterations, the data subset and quantum circuit depth are used to train the parameters of the quantum computing system model. 406. Using a quantum computing system model and a quantum computer to generate solutions for said subsets of data. 408 Does the solution of the quantum computing system model satisfy the threshold solution? 410 Adjusting quantum circuit depth 412 Is the size of the data subset smaller than the size of the data set and / or are available qubits not being utilized? 414 Adjust the size of the data subset Generate 416 quantum computing system models 500 Systems 502 processor 504 memory 506 Data Storage 508 Communication Unit

Claims

1. 1. A computer-implemented method comprising: obtaining real-time text data relating to an item, the text data comprising a plurality of information portions; executing a categorization process configured to execute a plurality of threads in parallel, each thread operating on one piece of information at a time, each thread performing the following steps: obtaining a sentiment score based on the information portion using a sentiment analysis machine learning (ML) model; assigning a category to the item based on the sentiment score using a classification ML model trained on historical data; and performing a step of updating a live category based on the category assigned to the item; Stages and; outputting the live categories to a user in real time. A computer-implemented method.

2. 2. The computer-implemented method of claim 1, further comprising receiving real-time audio data relating to the matter and converting the real-time audio data into the real-time text data.

3. The matter may include an event, incident, problem, query, or inquiry; Preferably, the matter includes an emergency incident; More preferably, the real-time text data is derived from an emergency call.

4. The computer-implemented method of claim 1 , wherein the assigned category is a severity category indicating how serious and / or urgent the matter is.

5. 10. The computer-implemented method of claim 1, wherein outputting the live categories to a user in real time comprises displaying the live categories to a user via a graphical user interface (GUI).

6. 2. The computer-implemented method of claim 1, wherein each thread further performs the step of using a Named Entity Recognition (NER) ML model to obtain one or more keywords from the information portion; preferably, the sentiment analysis ML model further bases the sentiment score on the one or more keywords; and more preferably, the classification ML model further bases the category on the one or more keywords.

7. 7. The computer-implemented method of claim 6, wherein each thread further performs the step of updating a live keyword output based on the one or more keywords, the method further comprising outputting the live keyword output to a user in real time via a GUI.

8. The classification ML model trained on historical data: obtaining a historical database of historical data including a plurality of portions of text data, each portion labeled with a category; and training the classification ML model based on the historical database; Preferably, the method further comprises labelling said text data with its assigned category and adding it to said historical database for training purposes. A computer-implemented method according to any one of claims 1 to 7.

9. A computer program product which, when executed on a computer, causes the computer to carry out the method of claim 1.

10. An information processing device having a memory and a processor coupled to said memory, said processor configured to perform the method of claim 1.

Images