Jianqiang Hao

Mining social media data for opinion polarities about electronic cigarettes

Background There is an ongoing debate about harm and benefit of e-cigarettes, usage of which has rapidly increased in recent years. By separating non-commercial (organic) tweets from commercial tweets, we seek to evaluate the general publics attitudes towards e-cigarettes. Methods We collected tweets containing the words ‘e-cig’, ‘e-cigarette’, ‘e-liquid’, ‘vape’, ‘vaping’, ‘vapor’ and ‘vaporizer’ from 23 July to 14 October 2015 (n=757 167). A multilabel Naïve Bayes model was constructed to classify tweets into 5 polarities (against, support, neutral, commercial, irrelevant). We further analysed the prevalence of e-cigarette tweets, geographic variations in these tweets and the impact of socioeconomic factors on the public attitudes towards e-cigarettes. Results Opinions from organic tweets about e-cigarettes were mixed (against 17.7%, support 10.8% and neutral 19.4%). The organic—against tweets delivered strong educational information about the risks of e-cigarette use and advocated for the general public, especially youth, to stop vaping. However, the organic—against tweets were outnumbered by commercial tweets and organic—support tweets by a ratio of over 1 to 3. Higher prevalence of organic tweets was associated with states with higher education rates (r=0.60, p<0.0001), higher percentage of black and African-American population (r=0.34, p=0.01), and higher median household income (r=0.33, p=0.02). The support rates for e-cigarettes were associated with states with fewer persons under 18 years old (r=−0.33, p=0.02) and a higher percentage of female population (r=0.3, p=0.02). Conclusions The organic—against tweets raised public awareness of potential health risks and could aid in preventing non-smokers, adolescents and young adults from using e-cigarettes. Opinion polarities about e-cigarettes from social networks could be highly influential to the general public, especially youth. Further educational campaigns should include measuring their effectiveness.

Geographic density and proximity of vape shops to colleges in the USA

Background Vape shops have been spreading rapidly in the USA since 2008, catering to the fast-growing market for electronic cigarettes. Little is known about the geographic density and proximity of vape shops near colleges. Methods Names and addresses of vape shops were collected from 3 online directories: Yelp.com, Yellowpages.com and Guidetovaping.com. We identified the prevalence of US-based vape shops and their density and proximity to colleges using a geographic information system. General linear model and negative binomial regression were performed to examine the factors associated with proximity and density of vape shops near colleges. Results We identified 9945 vape shops in the USA as of December 2015, a nearly threefold increase from 2013. Among the 2755 colleges included in this study, 66.5% had at least 1 vape shop within a 3-mile radius. The median proximity of the nearest vape shop to each college/university was 1.8 miles. Proximity increased by student population, private as compared to public institutions, and location (city vs rural). Within a 1-mile radius, colleges with smoke-free campus policies had a lower density of vape shops (RR=0.6, p=0.002) than those without smoke-free campus policies. Private institutions had a higher density of vape shops (RR=7.8, p<0.0001) than did public institutions. Colleges with campus housing had a lower density of vape shops (RR=0.4, p<0.0001) than those without campus housing, and colleges located in cities had a much higher density of vape shops than those located in rural areas (RR=6.6, p<0.0001). Smoke-free and e-cigarette-free campus policies had significant interactions with college type (private vs public) and campus housing in reducing vape shop density. Conclusions Vape shops are more likely to be located near private institutions and colleges in cities as opposed to rural areas. Smoke-free and e-cigarette-free campus policies have had significant effects in reducing the density of vape shops but have not reduced the proximity of vape shops to colleges. Regulations on the sale and advertisement of e-cigarettes to youth and young adults are critically needed.

Mining social media data on marijuana use for Post Traumatic Stress Disorder

BackgroundWe seek to evaluate factors that could potentially impact the publics attitudes to PTSD related marijuana use on Twitter. MethodsWe collected tweets that contained the PTSD and Post Trauma Stress Disorder from August 1, 2015 to April 15, 2016 (n=1,253,872 tweets). A Nave Bayes model was constructed to classify tweets into two opinion polarities (support vs. neutral/against marijuana use for PTSD). ResultsThe marijuana related tweets were predominated by the supporting opinions (89.6%). The public opinions about marijuana use for PTSD on Twitter were significantly associated with state-level legislation. States that legalized medical and recreational marijuana use had the highest prevalence of support tweets (1.30.6), followed by the states that legalized medical but not recreational use (0.50.3) and the states that had no laws legalizing marijuana (0.20.1, p<0.0001). A higher prevalence of support tweets was associated with states with a lower proportion of youth (r=0.35, p=0.01) and a higher education rate (r=0.38, p=0.006). ConclusionTwitter data suggest a proliferation of supporting marijuana use for PTSD treatments, especially in the states that legalized medical and/or recreational use of marijuana. We found a proliferation of PTSD tweets that support marijuana use on Twitter.The prevalence of supporting marijuana use was correlated with state legal statues.Age, education and income were associated with the prevalence of support tweets.Educational campaigns on adverse effects of marijuana use are critically needed.

Predicting Asthma Prevalence by Linking Social Media Data and Traditional Surveys

Asthma is one of the most common chronic diseases that has a profound impact on people’s well-being and our society. In this study, we link multiple large-scale data sources to construct an epidemiological model to predict asthma prevalence across geographic regions. We use: (1) the Social Media Monitoring (SMM) data from Twitter (N = 500 million tweets/day), (2) the 2014 Behavioral Risk Factor Surveillance System (BRFSS) (N = 464,664), and (3) the 2014 American Community Survey (ACS) conducted by the U.S. Census Bureau (N = 3.5 million per year). We predict asthma prevalence in the traditional survey (BRFSS) using social media information collected from Twitter and socioeconomic factors collected from ACS. The evidence suggests that monitoring asthma-related tweets may provide real-time information that can be used to predict outcomes from traditional surveys.

Social media content and sentiment analysis on consumer security breaches

Purpose Security breaches have been arising issues that cast a large amount of financial losses and social problems to society and people. Little is known about how social media could be used a surveillance tool to track messages related to security breaches. This paper aims to fill the gap by proposing a framework in studying the social media surveillance on security breaches along with an empirical study to shed light on public attitudes and concerns. Design/methodology/approach In this study, the authors propose a framework for real-time monitoring of public perception to security breach events using social media metadata. Then, an empirical study was conducted on a sample of 1,13,340 related tweets collected in August 2015 on Twitter. By text mining a large number of unstructured, real-time information, the authors extracted topics, opinions and knowledge about security breaches from the general public. The time series analysis suggests significant trends for multiple topics and the results from sentiment analysis show a significant difference among topics. Findings The study confirms that social media monitoring provides a supplementary tool for the traditional surveys which are costly and time-consuming to track security breaches. Sentiment score and impact factors are good predictors of real-time public opinions and attitudes to security breaches. Unusual patterns/events of security breaches can be detected in the early stage, which could prevent further destruction by raising public awareness. Research limitations/implications The sample data were collected from a short period of time on Twitter. Future study could extend the research to a longer period of time or expand key words search to observe the sentiment trend, especially before and after large security breaches, and to track various topics across time. Practical implications The findings could be useful to inform public policy and guide companies responding to consumer security breaches in shaping public perception. Originality/value This study is the first of its kind to undertake the analysis of social media (Twitter) content and sentiment on public perception to security breaches.