In this chapter, the authors propose to use contextual Word2Vec model for understanding OOV (out of vocabulary). The OOV is extracted by using left-right entropy and point information entropy. They choose to use Word2Vec to construct the word vector space and CBOW (continuous bag of words) to obtain the contextual information of the words. If there is a word that has similar contextual information to the OOV, the word can be used to understand the OOV. They chose the Weibo corpus as the dataset for the experiments. The results show that the proposed model achieves 97.10% accuracy, which is better than Skip-Gram by 8.53%.
This paper presents a system, namely, the abnormal-weather monitoring and curation service (AWMC), which provides people with a better understanding of abnormal weather conditions. The service can analyze a set of multivariate weather datasets (i.e., 7 meteorological datasets from 18 cities in Korea) and show (i) which dates are mostly abnormal in a certain city, and (ii) which cities are mostly abnormal on a certain date. In particular, the dynamic graph-embedding-based anomaly detection method was employed to measure anomaly scores. We implemented the service and conducted evaluations. Regarding the results of monitoring abnormal weather, AWMC shows that the average precision was approximately 90.9%, recall was 93.2%, and F1 score was 92.1% for all the cities.
Our study aims to detect multilingual Out-of-Vocabulary (OOV) and matching among multilingual OOV. Based on the original OOV issue, many multilingual OOVs also emerged at the same time. In order to solve this problem, this paper proposes a graph embedding-based matching among multilingual OOV. The method is divided into two parts. The first part is to extract OOV from the network corpus and understand it. In the second part, the OOV in the first part is taken as the target node, and the understood part is taken as the feature node of the target node to construct the graph and embed the graph. Our study uses Chinese, Korean, and Japanese for the experiment. The result of the method that Our study proposed is that F1- score reached 93.94%. Our study also compares this method with other embedding algorithms, and the F1-score is higher than the average of other algorithms F1-score by 9.62%.