Attribute Identification Framework for Multi-Label Review Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing number of online product reviews makes it difficult for customers to make informed purchasing decisions and for manufacturers to track customer opinions, as existing text classification methods are inadequate for multi-labeled text classification tasks, which require identifying multiple attributes in reviews.
Innovation Solution
An attribute identification (AI) framework comprising an offline training portion and an online prediction portion, utilizing a support vector machine (SVM) based classification model to enhance attribute identification by considering relationships between attributes, and employing preprocessing, feature selection, and model training to predict attributes and their number in input sentences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If single-labeled text classification is used, then the classification process is simple, but it cannot handle multi-attribute reviews effectively
Solution Approach 1:
The patent segments the text classification task into multiple binary classification problems, one for each attribute. Each binary classifier independently determines whether a review belongs to a specific attribute category, enabling multi-attribute classification while maintaining relatively simple individual classifier designs.
Solution Approach 2:
The patent creates a universal classification framework that can handle multiple attributes simultaneously. The multi-label classification system serves multiple functions by identifying various product attributes (quality, price, service, etc.) within a single classification process, making the system adaptable to different review types and attributes.
2Adaptability or versatility
If multi-label text classification is implemented, then multiple attributes can be identified, but the uncertainty measurement becomes more complex
Solution Approach 1:
The patent introduces an intermediary uncertainty measurement mechanism that bridges the gap between binary classifier outputs and multi-label classification uncertainty. By calculating uncertainty based on the distribution of classifier scores across multiple attributes, the system manages complexity while providing meaningful uncertainty estimates for multi-attribute reviews.
3Quantity of substance
If the number of reviews increases, then more customer opinions are available, but it becomes harder to read and analyze them
Solution Approach 1:
The patent extracts key attribute information from reviews by classifying them into predefined attribute categories. This extraction process identifies and isolates important product attributes (such as quality, price, service) from the full review text, making large volumes of reviews more accessible and analyzable by presenting structured attribute-based summaries.
Data Source
AI summary
Techniques for identifying attributes in a sentence and determining a number of attributes to be associated with the sentence is described. An attribute identification (AI) framework comprises an offline training portion, an online prediction portion, and an AI algorithm module. The offline training portion utilizes the relationships between attributes within sentences input to the offline training portion to improve attribute identification of the AI algorithm module. The online prediction portion predicts, for each sentence input, the attributes of the sentence and the number of attributes the sentence is associated with by employing the AI algorithm module.