Microblog Classification System Using Segmented Image and Text Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems lack an efficient method to classify and determine user actions related to search keywords across mixed text and image information items on microblogs, such as determining user engagement and sentiment analysis in a chronological context.
Innovation Solution
A microblog classification system that accepts search keywords, retrieves relevant information items, classifies image items based on user actions, and determines the sentiment of text items using machine learning models to identify user engagement and sentiment, integrating image and text classification models for comprehensive analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If multiple information items including both text and image are retrieved from microblogs, then the quantity of information is increased, but the complexity of classification and analysis increases
Solution Approach 1:
The patent segments the classification task into distinct modules: an image classification unit that processes image information items separately from a text classification unit that processes text information items. This segmentation allows the system to handle multiple information types independently, reducing the overall complexity of the classification system while maintaining the ability to process large quantities of diverse information items.
2Measurement precision
If image information items are classified to determine user actions, then the precision of user engagement detection is improved, but the time required for processing increases
Solution Approach 1:
The patent applies preliminary action by performing image classification and text classification in parallel independent processes rather than sequentially. The image classification unit and text classification unit operate simultaneously on their respective information items, which reduces the total processing time while maintaining high precision in user action detection through dedicated classification models.
3Measurement precision
If chronological filtering is applied to posted information items, then the relevance of retrieved information is improved, but the complexity of information retrieval increases
Solution Approach 1:
The patent applies local quality by implementing chronological filtering specifically for the image information items retrieved from microblogs, while maintaining separate processing for text information items. This localized application of time-based filtering ensures that image-related information is chronologically relevant without unnecessarily complicating the overall retrieval system architecture.
Data Source
AI summary
There is provided a non-transitory computer-readable medium storing a program causing a computer to execute a process. The process includes: accepting a search keyword; retrieving, from information items posted by users, a posted information item including the accepted search keyword, each of the posted information items including at least either of a text information item and an image information item, and acquiring posted information items which are within a predetermined chronological range with respect to the posted information item including the search keyword; and classifying, as image information items related to the search keyword, some of image information items included in the posted information items that have been acquired, and performing first determination of, for each of the classified image information items, whether or not a user who posted an information item including the classified image information item took an action related to the search keyword.