Bank of America Corporation

Ion Gerald McCusker

Ramakrishna R Yannam

Ravisha Andar

Emad Noorizadeh

Saahithi Chillara

Ashwini Patil

Priyank R Shah

Sushil Golani

Yogesh Raghuvanshi

Donatus Asumu

Prejish Thomas

Pravin Kumar Sankari Bhagavathiappan

Christopher Keith Restorff

Bharathiraja Krishnamoorthy

Devanshu Mukherjee

Systems and methods for increasing accuracy in an online chat interface are provided. Methods may be executed via a machine-learning (ML)-based chat monitoring engine. Methods may include intercepting a request utterance. Methods may include computing, via a trained ML model, an intent of the request utterance; generating a target response to the request utterance; intercepting a response utterance; and calculating a difference between the response utterance and the target response. In response to the difference being less than a threshold difference, methods may include releasing the response utterance to be transmitted as a response message. In response to the difference being more than a threshold difference, methods may include preventing the response utterance from being transmitted as a response message, generating a revised response utterance that is less than a threshold difference apart from the target response, and transmitting the revised response as a response message to a remote computing device.

Machine learning (ML)-based dual layer conversational assist system

A system for horizontal scaling of retraining machine learning models across operational domains is provided. The system may reduce computational overhead associated model retraining. The system may include an artificial intelligence ('AI') engine that determines target machine learning models that need to be retrained in response to changed training data. The AI engine may assign daemons to the target models. The daemons may gather retraining requirements such as source code and training data. The daemons may schedule the target models for retraining on a CPU or a GPU based model training system.

Machine learning model training system

A method for filtering a plurality of agent-customer interactions to determine whether one or more of a plurality of agent-customer interactions should be stored in a library of Artificial Intelligence (AI) files related to an interactive voice response system (IVR) is provided. The method may include receiving an identification of a plurality of IVR flashpoints, monitoring and/or reviewing the plurality of agent-customer interactions, and determining whether one of the plurality of agent-customer interactions meets a threshold number of the IVR flashpoints. For each of the plurality of agent-customer interactions that meets a threshold number of the IVR flashpoints, the method may further direct the IVR to convert the interaction into an IVR workflow and store the IVR workflow in the library of AI related to IVR.

Agent assist design—autoplay

An apparatus includes a memory and a processor. The memory stores first and second machine learning algorithms. The processor receives, from a user, voice signals associated with an information request and converts them into text. The processor uses the first machine learning algorithm to determine, based on the text, to automatically generate a reply to the request, rather than transmitting the request to an agent. The processor uses the second machine learning algorithm to generate, based on the set of text, the reply, which it transmits to the user. The processor receives feedback associated with the reply, indicating that the reply does or does not include the requested information. The processor uses the feedback to update either or both machine learning algorithms.

System and method for voice controlled automatic information access and retrieval

A system for reducing computational load for training machine learning models is provided. The system may provide an end-to-end-solution for automating development, testing and updating of machine learning models in various operational environments. The system may determine which machine learning models included in a computer program product need to be retrained in response to a change in training data. For a computer program product that includes multiple models, the system only retrains target models, resulting in significant savings in computing resources. The system may also reduce the number of machine learning models that need to be generated for testing environments, further reducing consumption of computational resources.

Machine learning model builder

Chatbots may be integrated into a customer service workflow and assist a user before, during and after a user-agent interaction. The chatbot may assist an agent during a user-agent interaction. The chatbot may provide customized responses for a target agent or user. Customized responses may be formulated based on conversation context, account information, sentiment and diagnostic tools. Chabot responses may be customized to meet habits and patterns of a target agent or user. The chatbot may crowdsource questions to other agents or users. The chatbot may employ search engines, entity and slot extraction and heat maps and clustering analysis to generate relevant responses for the agent or user.

Chatbot co-pilot

An apparatus includes a memory and a processor. The memory stores first and second machine learning algorithms. The processor receives, from a user, voice signals associated with an information request and converts them into text. The processor uses the first machine learning algorithm to determine, based on the text, to automatically generate a reply to the request, rather than transmitting the request to an agent. This determination indicates that the text is associated with a probability that the automatically generated reply includes the requested information that is greater than a threshold. The processor uses the second machine learning algorithm to generate, based on the set of text, the reply, which it transmits to the user. The processor receives feedback associated with the reply, indicating that the reply does or does not include the requested information. The processor uses the feedback to update either or both machine learning algorithms.

An apparatus includes a memory and processor. The memory stores previous requests and corresponding previous responses. The processor determines that a user device transmitted a new voice request, converts the voice request into a first set of text, and transmits the text to an agent device. The processor applies the machine learning algorithm to the first set of text to generate suggested responses, by identifying patterns shared by the first set of text and a subset of the previous requests that are associated with the suggested responses. The processor transmits the suggested responses to the agent device. The processor then determines that the agent device transmitted voice signals responding to the new request. The processor converts these voice signals into a second set of text. The processor stores the first set of text as a previous request, and the second set of text as a corresponding previous response.

System and method for dynamically identifying and retrieving information responsive to voice requests

A device that is configured to establish a network connection between a user and an agent. The device is further configured to identify a first issue type for the user and to identify a first resolution type provided by the agent based on a conversation between the user and the agent. The device is further configured to identify a performance score from a resolution mapping based on a combination of the first issue type and the first resolution type. The device is further configured to identify a first knowledge area that is associated with the first issue type and to update a first knowledge score that is associated with the first knowledge area in a performance record for the agent based on the performance score. The device is further configured to send a recommendation to the agent based at least in part on the performance score.

Performance monitoring for communication systems

An apparatus includes a memory and a processor. The memory stores a machine learning algorithm configured to select between forwarding a request to an agent device and transmitting an automatically generated reply to the request. The processor receives feedback for a decision made by the algorithm, indicating whether the automatically generated reply includes the information sought by the request. If the algorithm decided to forward the request to the agent device, a reward is assigned to feedback that indicates that the reply does not include the information, while a punishment is assigned to feedback that indicates that the reply includes the information. If the algorithm decided to transmit the reply, a reward is assigned to feedback that indicates that the reply includes the information, and a punishment is assigned to feedback that indicates that the reply does not include the information. The processor updates the algorithm using the reward/punishment.

Selective use of tools for automatically identifying, accessing, and retrieving information responsive to voice requests

A device that is configured to assign users to an issue cluster based on issue types for the users. The device is further configured to identify available agents and to assign each available agent to one or more knowledge area clusters based on knowledge scores. A knowledge score indicates an expertise level for an agent in a knowledge area. The device is further configured to identify an issue cluster that is associated with an issue type and to identify a user from the issue cluster. The device is further configured to identify a knowledge area cluster that is associated with the issue type and to identify an agent from the knowledge area cluster. The device is further configured to establish a network connection between a user device associated with the user and a user device associated with the agent.

Growing community of inventors

Allen, TX, United States of America

Ion Gerald McCusker