An impression label extraction method and device

Abstract

The invention discloses an impression label extraction method and device. According to the method, a first binary group and a second binary group which are obtained through combination are combined into an impression label according to position information of attributes in a document. Therefore, according to the method, the impression label existing in the document can be extracted from the document without training data. The method comprises the following steps of: extracting a preset product attribute from a document through character string matching; evaluating the impression language and apreset product type; Therefore, the method can meet a certain recall rate, and related product types, product attributes and product attribute evaluation impressions are associated with one another in the document, so that the first binary group and the second binary group are combined into the impressions labels by using the position information of the product attributes in the document as a bridge. And the extracted impression label can meet a certain accuracy rate. The invention further discloses a storage medium and a processor.

Description

technical field [0001] The present application relates to the technical field of data processing, in particular to a method and device for extracting impression tags. Background technique [0002] Impression tags are used for impression views about a certain attribute of a certain product. Generally speaking, the impression tag is a triplet composed of three elements, which can be specifically <target product type (target), product attribute (aspect), evaluation impression language (opinion)>, and the extraction of impression tags is done in text analysis In , the impression of a certain attribute of a product is extracted from a large amount of user evaluation data. For example: the triplet that needs to be extracted from the text "Ford Maverick looks stylish and wild" is <Ford Maverick, shape, stylish and wild>; the triplet that needs to be extracted from the text "Volkswagen Polo is narrow" is < VW Polo, space, cramped >. [0003] Existing image tag e...

AI Technical Summary

Problems solved by technology

Among them, the existing rule-based extraction methods mainly use dictionaries to match words that appear in sentences. If three elements appear in a short sentence in the text at the same time: target product type, product attribute, and evaluation impression, then these three elements Elements are extracted as an impression label. However, the target product, attribute, and evaluation impression that appear simultaneously in a short sentence may not have any subordination or modification relationship with each other. Therefore, the existing rule-based extraction of impression labels Although the method has a high recall rate, its precision rate is low

However, the existing algorithm-based extraction methods mainly use training data to train extraction models such as CRF or LSTM models. This method requires a large amount of training data when training the extraction model, and the recall rate of this method is low.

Images