NLG Text Generation Scoring and Ranking Method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current natural language generation (NLG) systems struggle to generate multiple versions of stories from identical data and lack the ability to automatically, consistently, and repeatably evaluate different outputs, leading to inconsistent and uninteresting content.
Innovation Solution
A computerized method that generates multiple text stories using a corpus and scoring system, including geographic, distance, information content, replacement, and extra aspect scores, to rank and select the most relevant story instances based on user input, ensuring objective evaluation and variation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If NLG systems use structured templates to generate text stories, then the generation process is efficient and repeatable, but the output becomes hard to differentiate from every other story generated by the same system
Solution Approach 1:
The patent applies local quality by allowing different portions of the text generation process to have different characteristics. The template provides structured efficiency, while randomization elements (such as varying sentence structures, adjectives, and narrative flows) introduce local variations that make each story unique while maintaining overall consistency in format and style.
Solution Approach 2:
The system changes parameters during text generation by incorporating randomization factors that modify narrative elements. These parameter changes include varying the order of information presentation, selecting different descriptive terms, and adjusting sentence complexity while maintaining the underlying template structure, thereby differentiating outputs without sacrificing generation efficiency.
2Reliability
If NLG systems generate the same story every time from identical data, then the output is consistent and reliable, but the content becomes uninteresting and lacks variety
Solution Approach 1:
The patent introduces dynamics into the otherwise static template-based generation process. By incorporating randomization elements that can vary with each generation, the system maintains reliability through consistent template structure while enabling content variety through dynamic selection of narrative elements, ensuring that identical data produces different but equally valid stories each time.
Solution Approach 2:
The system performs preliminary actions by pre-defining template structures and categories, then applies randomization within those pre-established frameworks. This preliminary structuring ensures reliability and consistency in format, while the subsequent randomization within the template allows for content variety and prevents monotony in the generated stories.
3Adaptability or versatility
If systems use after-the-fact performance metrics to rewrite titles and descriptions, then the adaptivity responds to actual user behavior, but the presentation becomes inconsistent and confusing for users
Solution Approach 1:
The patent applies preliminary action by generating multiple story variants in advance using randomization elements before viewing user behavior data. This allows the system to prepare diverse content options proactively rather than reactively, maintaining presentation consistency while still adapting to user preferences through the pre-generated variety, thereby eliminating the confusion caused by changing descriptions based on after-the-fact metrics.
4Manufacturing precision
If human intervention is introduced to review and edit NLG output, then the quality can be improved through human judgment, but the process becomes subjective and inconsistent depending on the individual reviewer
Solution Approach 1:
The patent implements self-service by enabling the system to automatically evaluate and select from multiple generated story variants using objective criteria and randomization mechanisms. This self-evaluation process eliminates the need for human reviewers, providing consistent and repeatable quality control that is not influenced by individual reviewer subjectivity, time of day, or mood, while still producing high-quality diverse content.
Data Source
AI summary
A computerized method for generating and evaluating natural language-generated text involves receiving, in a computer, data input by a user, generating, using a natural language generation technique, multiple instances of text stories based upon both contents of a corpus and the received data; analyzing the multiple instances of text stories as a weighted combination of computed geographic scores, distance scores, information content scores, replacement scores and extra aspect scores, providing a ranked set of the generated text stories to a user, receiving a selection of one of the text stories in the ranked set, and storing the selected story.