Multi-level Graph Pooling for Text Sentiment Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text sentiment analysis techniques based on machine learning often ignore global word collocations and long-distance semantics, leading to insufficient mining of text features and key sentiment information, which affects the accuracy of sentiment analysis.
Innovation Solution
A text sentiment analysis method using multi-level graph pooling, which builds a graph from text data by calculating collocate point mutual information between word nodes and employs a gated graph neural network layer, graph self-attention pooling layers, and graph convolutional layers to extract low-level and high-level feature representations, followed by feature fusion and sentiment category output.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sequence models (recurrent neural networks or one-dimensional convolutional neural networks) are used as base modules, then local continuous word sequence semantic information can be captured, but global word collocate with non-continuity and long-distance semantics are ignored
Solution Approach 1:
The patent segments the text processing into multiple levels: word level (basic tokens), phrase level (n-gram collocations), and sentence level (complete semantic units). This segmentation allows the model to capture both local continuous semantics at the word level and global non-continuous collocations at the phrase and sentence levels, resolving the contradiction between capturing local semantics and preserving global information.
Solution Approach 2:
The patent introduces a hierarchical dimension to the traditional sequence model by adding phrase-level and sentence-level processing layers. This dimensional expansion allows the model to simultaneously process local word sequences and global collocations, capturing both continuous and non-continuous semantic relationships that single-dimensional sequence models miss.
2Productivity
If traditional sequence models are used for text sentiment analysis, then processing speed is maintained, but key sentiment information and structural features of text are not adequately mined
Solution Approach 1:
The patent performs preliminary actions by pre-processing text into multiple hierarchical levels (word segmentation, phrase extraction, sentence structure analysis) before the main sentiment analysis. This preliminary structuring enables the model to efficiently process pre-organized information, maintaining processing speed while significantly improving sentiment information extraction through pre-computed structural features and collocation patterns.
3Use of energy by moving object
If feature mining is insufficient in text sentiment analysis, then computational resources are saved, but the final sentiment analysis effect is affected
Solution Approach 1:
The patent applies local quality by differentiating the depth of feature mining across different text levels. At the word level, basic semantic features are extracted; at the phrase level, collocation patterns and semantic relationships are mined with greater depth; at the sentence level, structural features and sentiment patterns are extensively analyzed. This differentiated approach ensures computational resources are allocated efficiently to where they yield the most sentiment analysis value.
Data Source
AI summary
A text sentiment analysis method based on multi-level graph pooling includes steps of: preprocessing a target text; taking collocate point mutual information between word nodes as an edge weight between the word nodes, and building a graph for each text independently; constructing a multi-level graph pooling model, of which a gated graph neural network layer transfers low-level information, a first graph self-attention pooling layer performs an initial graph pooling operation and uses a Readout function to extract low-level features, a second graph self-attention pooling layer performs a graph pooling operation again, performs a pruning update on the graph structure by calculating attention scores of nodes in the graph and uses a Readout function to extract high-level features; obtaining a multi-level final vector representation through a feature fusion function; and selecting a sentiment category corresponding to a maximum probability value as a final sentiment category output of the text.