Aspect-based sentiment analysis
AssigneeAdobe Inc.
InventorsAmir Pouran Ben Veyseh, Franck Dernoncourt
Life of this patent
Estimating the term…
Abstract
Systems and methods for natural language processing are described. One or more embodiments of the present disclosure generate a word embedding for each word of an input phrase, wherein the input phrase indicates a sentiment toward an aspect term, compute a gate vector based on the aspect term, identify a dependency tree representing relations between words of the input phrase, generate a representation vector based on the dependency tree and the word embedding using a graph convolution network, wherein the gate vector is applied to a layer of the graph convolution network, and generate a probability distribution over a plurality of sentiments based on the representation vector.
Claims20
A method for natural language processing, comprising:
- generating a word embedding for each word of an input phrase, wherein the input phrase indicates a sentiment toward an aspect term;
- computing a gate vector based on the aspect term;
- identifying a dependency tree representing relations between words of the input phrase;
- generating a representation vector based on the dependency tree and the word embedding using a graph convolution network, wherein the gate vector is applied to a layer of the graph convolution network;
- and
- generating a probability distribution over a plurality of sentiments based on the representation vector.
An apparatus for natural language processing, comprising:
- a dependency component configured to generate a dependency tree for an input phrase;
- a graph convolution network configured to generate a representation vector for the input phrase based on the dependency tree, wherein the representation vector is generated by applying a gate vector representing an aspect term of the input phrase to a layer of the graph convolution network;
- and
- a decoder network configured to generate a probability distribution over a plurality of sentiments toward the aspect term based on the representation vector.
A method for training a neural network, comprising:
- identifying a training set comprising an input phrase, an aspect term, a dependency tree, and a ground truth sentiment of the input phrase toward the aspect term;
- generating a gate vector based on the aspect term;
- generating a representation vector for the input phrase based on the dependency tree by applying the gate vector to a layer of a graph convolution network;
- generating a probability distribution over a plurality of sentiments based on the representation vector;
- computing a multi-task loss function including a prediction loss based on comparing the ground truth sentiment with the probability distribution over the plurality of sentiments;
- and
- training the graph convolution network based on the multi-task loss function.
Dependencies are inferred from the claim text.