A multi-modal personalized emotion processing method based on long-short-term memory mechanism

Abstract

The embodiment of the present invention discloses a multi-modal personalized emotion processing method based on the long-short-term memory mechanism. Taking music videos as "stimuli", a multi-modal personalized processing method based on the long-short-term memory mechanism is designed to automatically perceive And extract the characteristics of "stimuli", understand and remember the stimuli, and then simulate the process of human emotion transfer to realize the expression of emotion. The invention belongs to the field of artificial intelligence and relates to an artificial emotion simulation system, which can be applied to the development of emotional and intelligent new human-computer interaction products. In addition, starting from the brain-like mechanism, the present invention can provide a reference idea for simulating, extending and expanding the natural brain-like emotion of intelligent robots.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to an artificial emotion simulation system, which can be applied to the development of an emotional and intelligent new human-computer interaction method. Background technique [0002] Emotion is considered to be the final standard that distinguishes human behavior from all other biological behaviors. To make computers have higher and more comprehensive intelligence, computers must be endowed with the ability to recognize, understand, express and adapt to human emotions to establish a harmonious human being. machine environment. This is the ultimate goal in the field of affective computing, and it is also the only way to go from weak artificial intelligence to strong artificial intelligence. With the continuous deepening of research, the research content gradually expands and extends. People are no longer limited to mechanically calculating emotions, but systematic research o...

AI Technical Summary

Problems solved by technology

However, in these emotion studies, there are generally the following three problems: (1) For emotion cognition, researchers pay more attention to the classification accuracy, but lack of consideration of the uncertainty of emotion

(2) For emotion understanding, few people have studied it, especially in multimedia video content; (3) For emotion expression, since most simulation experiments do not use real "stimuli", this will lead to a lack of simulation effect good practical sense

[0007] 1. The classification method of the classification results with the priority of accuracy rate is not reasonable, especially in complex situations such as uncertainty and emotion

[0008] 2. When studying emotion classification or emotion calculation, visual neurons, auditory neurons and some neurons generated by experience in the brain are ignored. The importance of the combination of these three neurons, so the final classification effect of the existing model, It's all one-sided

[0009] 3. The influence of the emotion of the "stimulus" on the emotion transfer matrix is ​​not considered, that is, the relationship between the external "stimulus" and the emotion of the human brain is not established

However, current researchers do not view emotional decision-making from this objective perspective.

[0012] 2. Current researchers often study emotion models in a single modality, often ignoring the correlation with other modality data, resulting in a one-sided cognitive effect

[0013] 3. Although the input of "stimuli" is one of the key influencing factors of the state transition matrix, there are few studies at present

In addition, a small number of researchers who study the emotion of "stimuli" have not proposed a personality-emotional state transition model suitable for real fuzzy stimulus input, nor have they considered the input of "stimuli" as an equivalent problem-emotion classification. solve

Images