Source PaperYearLineSentence
N13-1017 2013 22
For the third con secutive year, a record number of new drugs (49) were detected in Europe in 2011 (EMCDDA, 2012)
N13-1017 2013 23
About two-thirds of these new drugs were synthetic cannabinoids (used as legal marijuana substitutes), which led to 11,000 hospitalizations in the U.S. in 2010 (SAMHSA, 2012)
N13-1017 2013 24
Treatment is complicated by the fact that novel substances like these may have unknown side effects and other properties.Accurate information on drug trends can be ob tained by speaking directly with users, e.g. focus groups and interviews (Reyes et al2012; Hout and Bingham, 2012), but such studies are slow and costly, and can fail to identify the emergence of new drug classes, such as mephedrone (Dunn et al., 2011)
N13-1017 2013 26
By (manually) analyzing YouTube videos, Drugs-Forum (dis cussed below), and other social media websites andonline communities, researchers have uncovered details about the use, effects, and popularity of a va riety of new and emerging drugs (Morgan et al 2010; Corazza et al2012; Gallagher et al2012),and comprehensive drug reviews now include non standard sources such as web forums in addition to standard sources (Hill and Thomas, 2011)
N13-1017 2013 10
Paul and Dredze (2012b) introduced factorialLDA (f-LDA), a general framework for multi dimensional text models that capture an arbitrarynumber of factors (explained in ?3)
N13-1017 2013 29
While topic models are a natural fit for corpus exploration (Eisenstein et al2012; Chaney and Blei, 2012), and have been used for similar public health applications(Paul and Dredze, 2011), online forums can be organized in many ways beyond topic
N13-1017 2013 35
With current information on a variety ofdrugs and an extensive archive, Drugs-Forum pro vides an ideal information source for public health researchers (Corazza et al2012)
N13-1017 2013 82
See Paul and Dredze (2012b) for more details
N13-1017 2013 113
An example of learned parameters is shown in Figure 2, illustrating the hierarchical process behind this model.Learning the Priors In various applications, pri ors can come from many different sources, such as labeled data (Jagarlamudi et al2012)
N13-1017 2013 218
See Paul and Dredze (2012a) for examples of parameters (the top words associated with various triples) learned by this model on this corpus
P14-1033 2014 56
Aspect extraction has been studied by many re searchers in sentiment analysis (Liu, 2012, Pang and Lee, 2008), e.g., using supervised sequence labeling or classification (Choi and Cardie, 2010, Jakob and Gurevych, 2010, Kobayashi et al, 2007,Li et al, 2010, Yang and Cardie, 2013) and us ing word frequency and syntactic patterns (Hu and Liu, 2004, Ku et al, 2006, Liu et al, 2013,Popescu and Etzioni, 2005, Qiu et al, 2011, So masundaran and Wiebe, 2009, Wu et al, 2009, Xu et al, 2013, Yu et al, 2011, Zhao et al, 2012, Zhou et al, 2013, Zhuang et al, 2006)
P14-1033 2014 60
To extract and group aspects simultaneously, topic models have been applied by researchers (Branavan et al, 2008, Brody and Elhadad, 2010, Chen et al, 2013b, Fang and Huang, 2012, He et al, 2011, Jo and Oh, 2011, Kim et al, 2013, Lazaridou et al, 2013, Li et al, 2011, Lin and He, 2009, Lu et al, 2009, Lu et al, 2012, Lu andZhai, 2008, Mei et al, 2007, Moghaddam and Es ter, 2013, Mukherjee and Liu, 2012, Sauper and Barzilay, 2013, Titov and McDonald, 2008, Wang et al, 2010, Zhao et al, 2010)
P14-1033 2014 10
Aspect extraction aims to extract target entities and their aspects (or attributes) that people have expressed opinions upon (Hu and Liu, 2004, Liu, 2012)
P14-1033 2014 76
In (Kang et al, 2012), labeled documents from source domains are transferred to the target domain to produce topic models with better fitting
P14-1033 2014 27
Other re lated works include (Andrzejewski et al, 2011,Chen et al, 2013a, Chen et al, 2013c, Mukher jee and Liu, 2012, Hu et al, 2011, Jagarlamudi et al., 2012, Lu et al, 2011, Petterson et al, 2010)
P14-1033 2014 259
To measure the results, we computePrecision@n (or p@n) based on the anno tations, which was also used in (Chen et al, 2013b, Mukherjee and Liu, 2012)
P14-3004 2014 127
pro or contra (like those of (Walker et al, 2012)or (Somasundaran and Wiebe, 2009) are not ap plicable in our case
P14-3004 2014 128
The same applies to the task of subgroup detection (as done by (AbuJbara et al., 2012), (Anand et al, 2011) or (Thomas et al, 2006)).In order to produce a finer-grained model of po sitions, we want to develop a model that places positions stated in text along a one-dimensional scale, as done by (Slapin and Proksch, 2008)with their system called Wordfish, (Gabel and Hu ber, 2000),(Laver and Garry, 2000), (Laver et al, 2003) or (Sim et al, 2013)
P14-3004 2014 57
We will evaluate the performance of this heuristic by checking a sample with human judges.Our extracted dataset covers the time period between March 2010 and December 2012 and con sists of 182 meetings
P14-3004 2014 89
Given that we have labeled data, a first solutionis to opt for a supervised approach to text clas sification, which has been successfully used for many tasks like topic detection ((Diermeier et al, 2012), (Husby and Barbosa, 2012), or sentiment analysis (Bakliwal et al, 2013), to name a few
P14-3004 2014 34
Some previous research looked at the relatedfield of opinion mining, also on political discus sion, as in (AbuJbara et al, 2012), (Anand et al., 2011) or (Somasundaran and Wiebe, 2009)
P14-3004 2014 104
There are variations of topic models that al low for influencing the creation of the topics, such as the systems of (Ramage et al, 2009) (LabeledLDA), (Andrzejewski and Zhu, 2009) or (Jagar lamudi et al, 2012)