News Article Categorization Using Dual ML Models and Metadata
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for categorizing news articles by activity and phase face difficulties when dealing with content related to multiple distinct yet related concepts, as they rely solely on semantic equivalence, which is insufficient for accurate classification.
Innovation Solution
A computer-implemented method using a combination of machine learning models and metadata assessment to categorize news articles by activity and phase, employing a first machine learning model to assign activities and a second model to identify activity phases, with the confirmation attribute based on Term Frequency—Inverse Document Frequency (TFIDF) values and time-related metadata to determine the phase of focus.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If semantic equivalence alone is used to group news articles, then the categorization process is simple, but the accuracy of phase identification deteriorates when articles contain multiple related concepts
Solution Approach 1:
The patent segments the categorization process into three distinct stages: (1) activity identification using a first machine learning model, (2) activity phase candidate identification using a second machine learning model, and (3) confirmation of the correct phase using metadata attributes. This segmentation allows each stage to focus on specific aspects of categorization, improving overall accuracy while maintaining operational clarity.
Solution Approach 2:
The patent introduces metadata attributes (such as TF-IDF values and time-related metadata) as intermediaries to bridge the gap between semantic analysis and accurate phase identification. These attributes serve as additional evidence to confirm which activity phase candidate is correct, especially when semantic equivalence alone is insufficient to distinguish between multiple related concepts.
2Adaptability or versatility
If multiple activity phase candidates are identified based on semantic analysis, then more potential phases are considered, but the difficulty of determining the correct phase increases
Solution Approach 1:
The patent employs feedback mechanisms where metadata attributes (TF-IDF values, time-related metadata) are used to evaluate and rank activity phase candidates. The system receives feedback from these attributes to confirm which candidate is the correct phase, reducing the difficulty of determination while maintaining comprehensive candidate coverage.
Solution Approach 2:
The patent changes the parameters used for phase identification by incorporating metadata attributes beyond semantic analysis. By using TF-IDF values and time-related metadata as additional parameters, the system can objectively rank and select the correct phase from multiple candidates, reducing determination difficulty.
3Productivity
If only semantic analysis is used for categorization, then the system is faster to implement, but it cannot eliminate candidates that are semantically similar but temporally impossible
Solution Approach 1:
The patent performs preliminary action by extracting and analyzing metadata attributes (such as time-related metadata) before final phase identification. This allows the system to pre-filter or pre-rank activity phase candidates based on temporal feasibility, eliminating impossible candidates early in the process while maintaining categorization efficiency.
Data Source
AI summary
A computer categorizes a news article by an activity and an associated activity phase, includes receiving, by a computer, at least one news article from an article source. The computer assigns for each news article, an activity of focus selected from a list of target activities, using a first machine learning model. The computer identifies, for each news article, at least one activity phase candidate selected from a list of activity phases associated with the activity of focus, using a second machine learning model. The computer determines, for each of new article, an activity phase of focus from among the activity phase candidates. The determination is based, at least in part on a confirmation attribute associated with the article. The computer categorizes each of the articles by said activity of focus and said activity phase of focus.