International Business Machines Corporation

Capital One Services, LLC

Sas Institute Inc.

Aysu Ezen Can

Corville O Allen

Roberto Delima

Scott Robert Carrier

Brendan Bull

Joshua S Edwards

Michael Bakr Mossoba

Tyler Maiman

Dwi Sianto Mansjur

Alexander Lin

Mia Rodriguez

Feng Qiu

Guadalupe Bonilla

Meredith Leigh Critzer

Robert Christian Sizemore

Vahid Khanagha

David Contreras

Bob Delima

Chris Symons

Zachary S Brown

Yurdaer Nezihi Doganata

Saratendu Sethi

Xiangqian Hu

Adam R Faulkner

Ethem F Can

Ning Jin

Maury Courtland

Jonathan Galsurkar

David Chen

A system may include processor(s), and memory in communication with the processor(s) and storing instructions configured to cause the system to correct ASR errors. The system may receive a transcription comprising transcribed word(s) and may determine whether the transcribed word(s) exceed associated predefined confidence level(s). Responsive to determining a transcribed word does not exceed a predefined confidence level, the system may generate a predicted word. The system may calculate a distance between numerical representations of the transcribed word and the predicted word and may determine whether the distance exceeds a predefined threshold. Responsive to determining the distance exceeds the predefined threshold, the system may determine whether at least one red flag word of a list of red flag words corresponds to a context of the transcription, and, responsive to making that determination, may classify the transcription as associated with a first category.

Systems and methods for correcting automatic speech recognition errors

In some embodiments, the present disclosure provides an exemplary method that may include steps of monitoring, a conversation script between a call center agent and a customer; utilizing, a speech-to-text deep machine learning model to transcribe the audio call to text; utilizing, a natural language processing deep machine learning model to map intent mappings of the audio call; utilizing, a similarity measurement model to determine a semantic similarity between predefined intent mappings and the intent mappings of the call audio text; determining, an error based on the semantic similarity in the intent mapping call audio text; determining a training session based on the error.

Computer-based systems configured to monitor a communication session and generate a training session and methods of use thereof

In some implementations, a system may capture audio from a call between a calling device and a called device. The system may filter the captured audio to generate a background audio layer. The system may generate an audio footprint that is a representation of sound in the background audio layer. The system may determine that the audio footprint includes a triggering sound footprint based on one or more audio characteristics of the audio footprint. The system may detect synthetic sound based on the audio footprint and after determining that the audio footprint includes the triggering sound footprint, wherein the synthetic sound is indicative of a sound recording. The system may transmit a notification to one or more devices associated with the call based on detecting the synthetic sound.

Detecting synthetic sounds in call audio

Disclosed herein are system, method, and computer program product embodiments for machine learning systems to process incoming call-center calls to provide communication summaries that capture effort levels of statements made during interactive communications. For a given call, the system receives a transcript as the input and generates a textual summary as the output. In order to improve a call summary and customize a summarization task to a call center domain, the technology disclosed herein may employ a classifier that predicts an effort level and attention score for individual utterances within a call transcript, ranks the attention scores and uses selected ones of the ranked utterances in the summary.

Identifying high effort statements for call center summaries

Systems and computer-implemented methods disclosed herein relate to detecting errors in manually entered data. In one embodiment, the system can identify a named entity automatically from a conversation between a customer and service agent with a named entity recognition model that employs natural language processing and machine learning to detect a word or string of words in the conversation that corresponds to a named entity category. In another embodiment, the system can determine whether data entered into a field on a service platform by the service agent includes an error by comparing the data entered with the named entity. In another embodiment, the system can transmit an alert to the service agent through the service platform when there is a mismatch between the named entity and the data entered.

Detection of manual entry error

Disclosed herein are system, method, and computer program product embodiments for machine learning systems to process incoming call-center calls based on caller preferences. Text of historical interactive communications of a set of first callers are used to train one or more machine learning models to extract caller preferences. These trained models extract caller preferences for a second caller to generate a customer specific profile. An automated call center assistance system is configured to selectively route, based on the customer specific profile, a current call from the second caller to a call center agent and communicate one or more of the caller preferences for the second caller to the call center agent for consideration in an interactive communication during the current call.

System and method for providing personalized customer experience in interactive communications

Disclosed herein are system, method, and computer program product embodiments for machine learning systems to process incoming call-center calls based on inferred themes. The machine learning system extracts a topic and keywords associated with the topic from a plurality of interactive communications and ranks the keywords based on a frequency of occurrence within the plurality of interactive communications. The machine learning systems select an N highest ranked keywords from the plurality of interactive communications, compares the N highest ranked keywords to previously extracted N highest ranked keywords to identify new keywords, and determines, based on new keywords, that an emerging topic has been articulated in the plurality of interactive communications.

System and method for identifying themes in interactive communications

A system and method for allowing a single live customer service agent to simultaneously serve multiple customers. According to various embodiments, a virtual agent operates at the front end to receive and attempt to handle customer issues. The virtual agent employs speech recognition and intent mapping in order to generate a proposed response that attempts to identify and resolve customer issues. In some scenarios, the proposed response includes both a response message and a response action to be taken. A chat history and the proposed response is then provided to the live agent. The live agent reviews the information provided, and determines whether the proposed response is appropriate. The live agent then approves the response or revises or replaces the response. The final response is sent back to the virtual agent for processing and providing to the customer.

System and method for serving multiple customers by a live agent

A mechanism is provided in a data processing system comprising least one processor and at least one memory, the at least one memory comprising instructions executed by the at least one processor to cause the at least one processor to implement a clinical decision support system. The mechanism receives a plurality of patient electronic medical records (EMRs) for a patient from a plurality of different sources. For a portion of a patient EMR record of the plurality of patient EMRs, the mechanism detects entities and analyzes a document structure of the portion of the patient EMR to identify a hierarchical structure of the portion of the patient EMR. The mechanism generates a container representation of the portion of the patient EMR based on the hierarchical structure. The mechanism placing each of the one or more sentences within the container representation based on relative position within the hierarchical structure. The mechanism generates a knowledge graph using the detected entities and the container representation.

Container-based knowledge graphs for determining entity relations in non-narrative text

A method, computer system, and a computer program product for active machine learning is provided. The present invention may include annotating a plurality of data entries. The present invention may also include building a first dataset based on the annotated plurality of data entries. The present invention may then include receiving user feedback based on the built first dataset. The present invention may further include assigning a plurality of weights to a plurality of data entry subsets. The present invention may also include generating a second weighted dataset based on the received user feedback.

Active learning for concept disambiguation

Growing community of inventors

Cary, NC, United States of America

Aysu Ezen Can