Method for ethical polarity discrimination based on behavior purpose relationship

Abstract

The invention relates to the technical field of artificial intelligence ethics, in particular to an ethical polarity discrimination method based on a behavior objective relationship, which uses a crowdsourcing method to remark an open source data set FG-News, adds more fine-grained attributes for the data set, can provide more features for model training, and improves the ethical polarity discrimination efficiency. Then a pre-training language model BERT with a bidirectional Transform structure is utilized, knowledge information is coded through Wikipedia corpora and a training process, meanwhile, for downstream tasks, the model subjected to fine adjustment can quickly adapt to the current task, target relation reasoning and polarity judgment are conducted on specific behavior words, and the behavior words are taken as the center, so that the target relation reasoning and polarity judgment are conducted on the specific behavior words; the target words, the relation between the words and the relation between the contexts are captured, the accuracy and recall ratio of the system are effectively improved, and a more complete and deeper analysis result is provided.

Description

technical field [0001] The present invention relates to the technical fields of deep learning, natural language processing and artificial intelligence ethics, and in particular to a method for discriminating ethical polarity based on behavior-purpose relationship. Background technique [0002] With the development of computers and technologies such as deep learning, artificial intelligence systems have progressed from computational intelligence to perceptual intelligence, and then to cognitive intelligence. Artificial intelligence systems have gradually penetrated into all aspects of our lives, such as unmanned driving, smart homes, service robots and many other fields. In these applications, due to the continuous development of human-computer interaction methods, the interaction between artificial intelligence systems and humans and society has become more and more prominent, and ethical issues have been highlighted continuously. [0003] Existing artificial intelligence s...

AI Technical Summary

Problems solved by technology

[0003] Existing artificial intelligence systems have the following defects: weak adaptability to environmental changes, difficult to deal with emergencies; low learning efficiency, which requires a lot of manpower and computing resources; deep learning models rely heavily on training data and are easily trained Issues such as the impact of social bias in the data

For the above points, it is easy to cause ethical problems.

[0004] Current methods for identifying ethical behaviors of artificial intelligence usually use natural language processing technology to detect the overall polarity of ethical scenarios, without considering a specific behavior or goal, so it is easy to make wrong predictions about ethical polarity

Especially when a sentence contains multiple behaviors, it is difficult for the above method to perform fine-grained analysis on each behavior

Images