Gaoyan Ou's research while affiliated with Peking University and other places

... Community detection plays a vital role in analyzing the effect of some real-world happenings. Ou et al. (2017) examined the emotion of an event that occurred in the real world-the proposed algorithm finds a community, detects the community emotion, aggregates the community emotion, and detects any community emotion burst. In other cases, most people are comfortable with the brand they use for a given product. ...

... Following key methods are carried out: (i) topic tracking-active learning method (7) and topical phrases learning method (8) were employed to detect the posts related to TCM topics; (ii) topic mining-we utilize LDA topic mining method (9,10) to explore the topic categories in posts; (iii) sentiment analysis-opinion aware knowledge graph (11) and sentiment oriented maximum entropy classification method (12) are used to calculate the posters' opinions on TCM; (iv) factor analysis-Lasso (Least Absolute Shrinkage and Selection Operator) Regression (13) and Bayesian network (14) are employed to calculate the dependencies between the growing visits of TCM and the major factors. The above big data platform and machine learning methods can support us to do such objective quantitative analysis: (i) How intense is the criticism of TCM, and the impacts on the actual TCM visiting population? ...

... Hwang et al. divided the total waiting time at a fixed location by the number of taxi rides and made this value as the waiting time for passengers at this location [15]. Qiu et al. utilized the road, climate, and Non-homogeneous Poisson process (NPPCRW) to forecast waiting time [16]. Jing et al. merged the factors of speed, climate, weekdays, and weekends to construct the T wait and T delay formulas and regarded their sum as the final passenger waiting time [17]. ...

... In advanced logistic regression models, like the one implemented here, biased parameter estimates due to the rare events problem, (small-sample bias), are corrected. For more on logistic modeling of rare events see Heinze and Schemper (2002), Heinze and Puhr (2010), Gao and Shen (2007), Lee et al. (2006), Maalouf and Trafalis (2011), Qiu et al. (2013). ...

... SA techniques based on unsupervised machine learning involve learning algorithms that develop without labeled training samples. In this method, the similarities of the texts are first measured according to keyword lists of categories, and then the texts are clustered into multiple groups according to the similarities of the texts, as exemplified in the work of [30], [66], [21], and [47]. SA techniques based on semi-supervised machine learning utilize learning algorithms that are partially trained on labeled data, as demonstrated by [38] and related works. ...

... Second, these features are processed using some typical machine learning approaches to determine whether or not the text has a sarcastic intent. Liu et al. [25] adopted lexical sequence features and semantic imbalance features to detect the sarcastic properties of sentences. Joshi et al. [11] classified the incongruity theory of sarcasm in linguistics into implicit and explicit incongruity features. ...

... However, it did not handle more types of features such as adjectives and verbs, and did not consider implicit features. Besides, in 2013, authors [25] proposed two novel APSM and ME-APSM models to extract aspects and aspect-specific polarity-aware sentiments from online reviews. However, the results showed that the model still needed improvements in terms of the aspect-level sentiment classification. ...

... It can be replaced by its alternatives like ELMo [30] or XLNet [42]. 8 On average, each variable node contains only 2-5 words in different verticals. 5% 4% 3% 2% 1% 0% Figure 9: Heatmap denoting the performance lifts per F1 score from the out-of-domain knowledge. ...