On August 11, Lei Jun, founder and CEO of Xiaomi, unveiled the company’s latest AI product: CyberOne, a full-scale humanoid robot. It is reported that CyberOne is 177CM tall and weighs 52KG. The humanoid—whose nicknamed "Metal Bro", can perceive human emotions, has keen vision, and can achieve bipedal posture balance. This full-scale humanoid bionic robot can also perceive 45 kinds of human semantic emotion and has a depth information accuracy of 1% within 8 meters. Lei Jun said that CyberOne takes artificial intelligence as the core and standard humanoid as the carrier. What amazes me most about this robot is its 45 kinds of human semantic emotion perception ability, which makes this product no longer a splicing of cold metal materials, but a "warm" and perceptible mind.

Semantic Sentiment Analysis

Semantic sentiment analysis belongs to the category of knowledge mining -- including information extraction, opinion mining, label construction, graph construction and other tasks. Sentiment analysis, also known as propensity analysis, or opinion mining, is the process of analyzing, processing, summarizing and reasoning on subjective texts with emotional colors. Using the ability of sentiment analysis, it can automatically judge the positive and negative sentiment tendency of the text and give the corresponding results for the natural language text with subjective description. Its main purpose is the process of analyzing, processing, summarizing and reasoning on subjective texts with emotional colors.

Mainstream Technology

One is sentiment analysis based on sentiment dictionary, which refers to the text processing of the text to be analyzed to extract sentiment words according to the constructed sentiment dictionary, and the sentiment tendency of the text is calculated. The final classification effect depends on the perfection of the sentiment dictionary.

The other is sentiment analysis based on machine learning, which refers to selecting sentiment words as feature words, matrixing the text, and using logistic regression, Naive Bayes (Naive Bayes), support vector machine (SVM) and neural network methods. Classification. The final classification effect depends on the choice of training text and the correct sentiment annotation. At present, some researchers combine the two. For example, the texts in some fields are not marked, and the sentiment dictionary in this field is not perfect, and manual annotation requires a lot of cost. When the data collection is much smaller than the labor cost; some texts can be selected. , using the basic sentiment dictionary method to roughly calculate the sentiment scores of these texts, and select the texts with high or low scores as the marked training texts. Combined with machine learning methods for analysis. According to the granularity of semantic analysis, it can be divided into document-level, sentence-level, object-level, and vocabulary-level sentiment analysis.

Development Trend

In recent years, the SOTA models in the NLP field are basically based on pre-training. At present, most sentiment analysis tasks are carried out on the basis of pre-training models. The review article "S. Poria, D. Hazarika, N. Majumder, and R. Mihalcea. Beneath the tip of the iceberg: Current challenges and" lists the indicators of different methods for IMDB, SST-2/5, and Semeval datasets (as shown in the figure below). The results of the SOTA sentiment analysis model are all based on predictions. Train the model. Although the industry SOTA model can achieve more than 95% accuracy on some coarse-grained emotional datasets, the current effect is far less than human level on fine-grained and complex tasks. At the same time, the migration ability of sentiment analysis models in multiple scenarios Poor, more scholars are needed to study multi-domain adaptive algorithms to solve this problem. For application scenarios, with the development of e-commerce, intelligent customer service, public opinion media, social media, humanoid robots and other technologies, sentiment analysis can be implemented in these directions.

Maybe, one day in the future, you will find that the person who understands you best in the world may be a humanoid robot.

Recommendations:

Chinese Chatting Text Corpus

Mandarin Chinese Speech Corpus for TTS