Text conceptualization and network representation fused opinion retrieval system and method

Abstract

The invention relates to a text conceptualization and network representation fused opinion retrieval system and method. The system comprises a text conceptualization module, a network representation module, an opinion retrieval score calculation module, a relevant retrieval unifying module and a sorting learning retrieval module. The text conceptualization module is used for judging concepts, in specific contexts, of entities corresponding to multiple concepts; the network representation module projects vocabulary nodes into a low-dimension semantic space; the opinion retrieval score calculation module calculates query and text topic similarity scores and to-be-queried text opinion scores to obtain text composite scores; the relevant retrieval unifying module obtains a query opinion retrieval result according to a descending order of the text composite scores and outputs the query opinion retrieval result; the sorting learning retrieval module integrates text conceptualization characteristics, network representation characteristics and the text opinion score characteristics into an existing sorting learning framework to classify and outputs texts sequentially. By the text conceptualization and network representation fused opinion retrieval system and method, concept level abstraction of text knowledge can be efficiently realized, query and text semantic relation can be analyzed, and user query related texts with query opinions can be retrieved.

Description

technical field [0001] The present invention relates to the field of opinion retrieval, and more specifically, to an opinion retrieval system and method that integrates text conceptualization and network representation, which can efficiently retrieve documents related to a given query from social media and other document collections in combination with knowledge graphs, and at the same time A list of related documents with an opinion on the query. Background technique [0002] Currently, there are many technical approaches available for opinion retrieval. The traditional opinion retrieval method divides the opinion retrieval into two stages. Firstly, the traditional information retrieval model or language model is used to obtain the document collection related to the topic, and then the opinion classifier is used to obtain the documents with opinions from the topic-related documents. The focus of method work is on the second stage, but it lacks theoretical basis and the eff...

AI Technical Summary

Problems solved by technology

Generally speaking, using ranking learning algorithm for opinion retrieval can often achieve higher accuracy, but because it requires a large amount of manually labeled data to build a training set, the application scenarios of this method are relatively limited compared with the first two methods

[0005] However, in social media, the user's query input is usually short, often just a simple word, and it is difficult to accurately express the information needs of the user's query, and the current opinion retrieval systems and methods often fail to integrate words according to the context during the retrieval process. Carry out abstraction at the conceptual level, and ignore the semantic connection between words at the semantic level, and lack the ability to generalize at the tendentious level

Images