Interpretability-considered automatic recognition method and system for network false comments

Abstract

The invention belongs to the technical field of network comment information processing, and discloses an interpretability-considered automatic recognition method and system for network false comments.The method comprises the steps: constructing a language use behavior and psychological analysis framework of a network false comment writer, and constructing a Chinese network comment language clue index system; constructing a false comment text feature set in combination with text distributed representation; constructing a false comment classifier, and judging whether the network comments are true or false; quantitatively evaluating the importance of different explanatory indexes to obtain true and false comment clues of which the dimension mean values are significantly different; Accordingto the value of the candidate network comments on the explainable language clue dimension, comparing the value with the obtained mean value of the explainable features, and giving explanation of a judgment result of the false comment automatic recognition method. According to the method, accurate, automatic recognition and easy-to-accept, stable and reliable interpretation of the false network comments are realized, and the accuracy and interpretability of an existing automatic classification method for the network false comments are improved.

Description

technical field [0001] The invention belongs to the technical field of network comment information processing, and in particular relates to an automatic identification method and system for network false comments with interpretability in mind. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: the early fake review standards emphasized the difference between user ratings and mainstream ratings, and the requirements for user data scale are relatively high, and the results are often too rough. In recent years, some researchers have transformed the problem of review credibility into a binary machine learning classification task of distinguishing fake and real reviews. Among them, related research generally adopts supervised and semi-supervised machine learning methods, and relies on the statistical regularity characteristics of user behavior and ratings. However, most of the classifiers built in related models targ...

AI Technical Summary

Problems solved by technology

[0006] In general, in the context of Chinese network fake comments, the problems existing in the existing technology are: (1) How to integrate text features such as syntax, semantics and stylistic elements to achieve better classification accuracy and achieve better classification Accuracy remains a challenge

[0007] (2) How to explain the intrinsic motivation and vocabulary usage behavior of fake review writers from a unified cognitive framework, resulting in the lack of interpretation of relevant prediction methods and the inability to give reasons for judging real and fake online reviews

However, the existing automatic classification methods lack the realization ideas, operation steps and methods for this integration process.

[0010] To sum up, the problems existing in the existing technology are: (1) The existing technology does not have a unified cognitive framework to explain the intrinsic motivation and vocabulary usage behavior of the fake review writers, which leads to the lack of interpretability of the relevant prediction methods and cannot give Reasons for identifying real and fake online reviews

[0011] (2) The existing methods have not yet solved the fusion of artificial features such as syntax, semantics, and stylistic features in the above-mentioned language clues with the automatic features of the deep neural network, and cannot further enhance the performance of the automatic false review classification method

[0012] (3) It is currently impossible to solve the problem of mapping the language clues in online reviews and the psychological characteristics of reviewers

[0013] (4) The existing automatic classification methods lack the realization ideas, operation steps and method realization of the integration process of syntax, semantics and neural network features

At present, there is no solution to this problem in the related technologies of false review identification, which is extremely difficult for technological innovation

Images