Aspect Level Sentiment Analysis: Convolution over Hierarchical Syntactic and Lexical Graphs - Model Architecture and Aggregation

This model utilizes a hierarchical syntactic and lexical graph convolutional approach for aspect-level sentiment analysis. It consists of two primary components:

1. Hierarchical Syntactic and Lexical Graph Modeler: This component converts the input text into hierarchical syntactic and lexical graphs. It comprises two submodules: a hierarchical syntactic parser and a lexical graph builder.

2. Convolutional Neural Network Classifier: This is a sentiment classifier based on convolutional neural networks, designed to categorize the sentiment of each aspect.

The model operates in the following steps:

1. The input text is fed into the hierarchical syntactic and lexical graph modeler.

2. The hierarchical syntactic parser decomposes the text into sentences and words, organizing them into a hierarchical syntactic tree.

3. The lexical graph builder connects relevant words for each aspect into a lexical graph.

4. The lexical graph and syntactic tree are then passed to the convolutional neural network classifier.

5. The convolutional neural network classifier employs convolutional and pooling operations to extract feature vectors for each aspect.

6. These feature vectors are subsequently processed by two aggregators.

7. The two aggregators aggregate the feature vectors, resulting in the final sentiment classification output.

The two aggregators are applied in steps 5 and 7 of the model. The aggregator in step 5 combines feature vectors for each aspect to generate a more representative vector. The aggregator in step 7 combines the vectors from all aspects to produce the final sentiment classification outcome. The aggregators use average pooling operations for the aggregation process.