Aspect category detection system and method based on evaluation object reinforcement and constrained label embedding

Abstract

The invention discloses an aspect category detection system and method based on evaluation object enhancement and constrained tag embedding, which introduces evaluation object information enhanced aspect category detection features, a constrained tag embedding mechanism and multiple attention mechanisms in a multi-task manner In combination, the aspect category detection of commodity reviews is realized. The present invention uses entity words in sentences as external information, and integrates them into the model through evaluation object extraction tasks. And a corresponding gating mechanism is designed to allow the entity word information to assist in the category detection task. Second, the present invention improves the discrimination between individual topic tags by adding constraints to tag embedding. At the same time, the present invention extracts different types of semantic features through different types of attention mechanisms, thereby solving the problem of insufficient feature extraction under the traditional single attention mechanism.

Description

【Technical field】 [0001] The invention belongs to the technical field of electronic information, and relates to an aspect category detection system and method based on evaluation object reinforcement and constrained label embedding. 【Background technique】 [0002] With the rapid development of the Internet and e-commerce, people are more and more online shopping, ordering food, booking hotels, etc. through the Internet. People often post their opinions and suggestions on the product in the comment area of ​​the product after shopping. These product reviews are of great significance to both consumers and businesses. For consumers, these commodity reviews are an important channel for consumers to understand product performance and will have a direct impact on consumers' purchase decisions. For companies that produce products, product reviews are of great significance in terms of product improvement, new function design, and competitive product analysis. However, the number o...

AI Technical Summary

Problems solved by technology

However, the existing model lacks pertinence in the process of introducing external knowledge. It only integrates external knowledge as auxiliary information, and fails to filter external knowledge so as to incorporate information that is really beneficial to aspect category detection.

In addition, label embedding has been proved to have a representative role in many natural language processing tasks. However, as far as the current research is concerned, label embedding has not been applied to aspect category detection. At the same time, previous label embeddings often did not take into account Distinction between labels

Finally, attention mechanisms are widely used in aspect category detection tasks, but existing attention mechanisms are often only one-sided attention mechanisms

Such an attention mechanism often screens all types of features through a unified attention mechanism, and cannot take into account different types of features, so the probability of noise in the attention mechanism is increased.

[0005] In summary, the existing aspect category detection models have the following problems: (1) There is no screening mechanism in the process of introducing knowledge, and some core external knowledge cannot be introduced in a targeted manner, so some irrelevant noise may be introduced

(2) There will be a problem of insufficient feature extraction under the single attention mechanism, and it is impossible to take into account the features of different categories

(3) There is no distinction between various categories from the perspective of label embedding, and there is a lack of distinction between labels

Images