Anaphora Resolution via ML Feature Vector Matching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing computer systems are inefficient in processing messages that contain implicit references, leading to misinterpretation of user requests and inefficient resource utilization, which reduces their ability to process data and maintain throughput.

Innovation Solution

A natural language processing system that identifies explicit references and replaces implicit references with their corresponding explicit forms using a machine learning model, allowing for more accurate message interpretation and resource management.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the system processes messages with implicit references using traditional methods, then the processing complexity increases and resource consumption increases, but the system lacks the ability to correctly interpret user requests

Engineering Contradiction:

Improvemessage interpretation accuracyVSAvoidprocessing complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent introduces an anaphora resolution component as an intermediary between message reception and processing. This component resolves implicit references by identifying their explicit antecedents in the conversation history, transforming messages with implicit references into messages with explicit references that can be processed by existing systems without increasing overall complexity

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system performs preliminary anaphora resolution before main message processing. By pre-resolving implicit references and replacing them with explicit references in advance, the system prepares messages for efficient processing by downstream components, avoiding the need for complex on-the-fly interpretation during main processing

Inventive Principle:

Principle #10Preliminary action

2Reliability

If the system uses detailed analysis to correctly interpret implicit references, then interpretation accuracy improves, but processing time increases and throughput decreases

Engineering Contradiction:

Improverequest interpretation accuracyVSAvoiddata processing throughput

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The message processing system is segmented into distinct components: message reception, anaphora resolution, and main processing. The anaphora resolution component handles only the specific task of resolving implicit references, while the main processing component handles general message processing. This segmentation allows each component to be optimized for its specific function, maintaining accuracy while improving overall throughput

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The anaphora resolution component serves itself by maintaining its own context window and resolution logic, independently resolving implicit references before passing processed messages to the main system. This self-contained approach avoids burdening the main processing system with complex reference resolution tasks, thereby maintaining high throughput while ensuring accurate interpretation

Inventive Principle:

Principle #25Self-service

3Reliability

If the system processes all messages with full analysis, then interpretation accuracy improves, but resource consumption increases

Engineering Contradiction:

Improvemessage processing accuracyVSAvoidprocessing resource consumption

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system applies detailed anaphora resolution only where needed - specifically when implicit references are detected in messages. Messages without implicit references bypass this detailed analysis and proceed directly to main processing with minimal resource consumption. This localized application of complex processing ensures high accuracy for challenging cases while maintaining efficient resource utilization overall

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS11977852B2Anaphoric reference resolution using natural language processing and machine learning

Publication Date: 2024.05.07 BANK OF AMERICA CORP

AI summary

A device configured to receive a sentence that includes a plurality of words. The device is further configured to input the words into a machine learning model that is configured to output a first feature vector based on the words. The device is further configured to identify a keyword within the sentence and to determine that the keyword is an implicit reference to an item. The device is further configured to identify a second feature vector in a reference list that closest matches a numeric value of the first feature vector and to identify an explicit reference in the reference list that is associated with the second feature vector. The device is further configured to replace the keyword with the explicit reference in the sentence and to output the sentence that includes the first explicit reference.