No abstract

In the last few years thousands of scientific papers have investigated sentiment analysis, several startups that measure opinions on real data have emerged and a number of innovative products related to this theme have been developed. There are multiple methods for measuring sentiments, including lexical-based and supervised machine learning methods. Despite the vast interest on the theme and wide popularity of some methods, it is unclear which one is better for identifying the polarity (i.e., positive or negative) of a message. Accordingly, there is a strong need to conduct a thorough apple-to-apple comparison of sentiment analysis methods, as they are used in practice, across multiple datasets originated from different data sources. Such a comparison is key for understanding the potential limitations, advantages, and disadvantages of popular methods. This article aims at filling this gap by presenting a benchmark comparison of twenty-four popular sentiment analysis methods (which we call the state-of-the-practice methods). Our evaluation is based on a benchmark of eighteen labeled datasets, covering messages posted on social networks, movie and product reviews, as well as opinions and comments in news articles. Our results highlight the extent to which the prediction performance of these methods varies considerably across datasets. Aiming at boosting the development of this research area, we open the methods' codes and datasets used in this article, deploying them in a benchmark system, which provides an open API for accessing and comparing sentence-level sentiment analysis methods.

Several messages express opinions about events, products, and services, political views or even their author's emotional state and mood. Sentiment analysis has been used in several applications including analysis of the repercussions of events in social networks, analysis of opinions about products and services, and simply to better understand aspects of social communication in Online Social Networks (OSNs). There are multiple methods for measuring sentiments, including lexical-based approaches and supervised machine learning methods. Despite the wide use and popularity of some methods, it is unclear which method is better for identifying the polarity (i.e., positive or negative) of a message as the current literature does not provide a method of comparison among existing methods. Such a comparison is crucial for understanding the potential limitations, advantages, and disadvantages of popular methods in analyzing the content of OSNs messages. Our study aims at filling this gap by presenting comparisons of eight popular sentiment analysis methods in terms of coverage (i.e., the fraction of messages whose sentiment is identified) and agreement (i.e., the fraction of identified sentiments that are in tune with ground truth). We develop a new method that combines existing approaches, providing the best coverage results and competitive agreement. We also present a free Web service called iFeel, which provides an open API for accessing and comparing results across different sentiment methods for a given text.

[1] A new Global Positioning System (GPS)-derived velocity field for the Andes mountains (26°-36°S) allows analysis of instantaneous partitioning between elastic and anelastic deformation at the orogen's opposing sides. Adding an ''Andes'' microplate to the traditional description of Nazca-South America plate convergence provides the kinematic framework for nearly complete explanation of the observed velocity field. The results suggest the oceanic Nazca boundary is fully locked while the continental backarc boundary creeps continuously at $4.5 mm/yr. The excellent fit of model to data (1.7 mm/yr RMS velocity misfit), and the relative aseismicity of the upper crust in the interior Andean region in comparison with its boundaries, supports the notion that the mountains are not currently accruing significant permanent strains. Additionally, the model implies permanent deformation is not accumulating throughout the backarc contractional wedge, but rather that the deformation is accommodated only within a narrow deformational zone in the backarc.

Data from a digitally recording seismic network in San Juan, Argentina, provide the first images of crustal scale basement faults beneath the lyecordillera. This seismicity is near the boundary between the lyecordillera (a thin-skinned thrust belt) and the Sierras Pampeanas (a region of thickskinned basement deformation), two seismically active tectonic provinces of the Andean foreland. The seismicity data support models for this region in which crustal thickening, rather than magmatic addition or thermal uplift, plays the dominant mountain building role. The lYecordillera seismicity occurs in three segments distributed north to south. The southern segment is an area of diffuse activity extending across the Precordillera and eastward into the Sierras Pampeanas that shows no patterns in map or cross section. The northern and central segments have well-defined dipping planes that define crustal scale faults extending from 5 to 35 km depth. It is clear from the relative fault geometries that the overlying Precordillera is not simply related to the basement activity. The seismicity here may result from reactivation of an ancient suture between the lYecordillera and Pampeanas terranes or be occurring in basement of unknown affinity west of the suture. The seismicity provides the first constraints on basement fault geometries, and we present models integrating this information with the surface geology. These basement faults may have been responsible for the 1944 Ms 7.4 earthquake that destroyed the city of San Juan. The imaging of these faults suggests that seismic risk estimates for San Juan made on the basis of surface geologic studies may be too low. Copyfight 1993 by the Amefican Geophysical Union. Paper number 92TC01108. 0278-7407/93/92TC-01108510.00 by subduction of the oceanic Nazca plate beneath the continental South American plate. The modern Andes are due to Neogene tectonics characterized by along-strike segmentation of major features of the subducted and overriding plates [Barazangi and Isacks, 1976; Jordan et al., 1983] ( Figures 1 and 2). In addition, the along-strike features of the two plates are correlated, both spatially and in their temporal development. Between 15øS and 24øS the Altiplano-Puna plateau, the Eastern Cordillera and Subandean Zone thrust belts, and an active volcanic arc are located above a moderately dipping Wadati-Benioff zone (WBZ). In contrast, a narrow thin-skinned thrust belt (lyecordillera), a wide zone of basement uplifts (Sierras Pampeanas), and an extinct are are located above an intermediate-depth subhorizontally subducting WBZ between 28øS and 33øS. Shutdown of volcanism, onset of Pampean and Precordilleran crustal deformation, and development of flat subduction are thought to coincide at 10-15 Ma [Jordan et al., 1983]. The Sierras Pampeanas and the lYecordillera are thought to have a tectonic relationship similar to that of the Mesozoic Sevier and Laramide provinces of North America, which are also thought to have developed over a region of flat subduction [Dickenson and Sn...

S U M M A R YThe convergence between the Nazca and South America tectonic plates generates a seismically active backarc region near 31 • S. Earthquake locations define the subhorizontal subducted oceanic Nazca plate at depths of 90-120 km. Another seismic region is located within the continental upper plate with events at depths <35 km. This seismicity is related to the Precordillera and Sierras Pampeanas and is responsible for the large earthquakes that have caused major human and economic losses in Argentina. South of 33 • S, the intense shallow continental seismicity is more restricted to the main cordillera over a region where the subducted Nazca plate starts to incline more steeply, and there is an active volcanic arc. We operated a portable broadband seismic network as part of the Chile-Argentina Geophysical Experiment (CHARGE) from 2000 December to 2002 May. We have studied crustal earthquakes that occurred in the back arc and under the main cordillera in the south-central Andes (29 • S-36 • S) recorded by the CHARGE network. We obtained the focal mechanisms and source depths for 27 (3.5 < M w < 5.3) crustal earthquakes using a moment tensor inversion method. Our results indicate mainly reverse focal mechanism solutions in the region during the CHARGE recording period. 88 per cent of the earthquakes are located north of 33 • S and at middle-to-lower crustal depths. The region around San Juan, located in the western Sierras Pampeanas, over the flat-slab segment is dominated by reverse and thrust fault-plane solutions located at an average source depth of 20 km. One moderate-sized earthquake (event 02-117) is very likely related to the northern part of the Precordillera and the Sierras Pampeanas terrane boundary. Another event located near Mendoza at a greater depth (∼26 km) (event 02-005) could also be associated with the same ancient suture. We found strike-slip focal mechanisms in the eastern Sierras Pampeanas and under the main cordillera with shallower focal depths of ∼5-7 km. Overall, the western part of the entire region is more seismically active than the eastern part. We postulate that this is related to the presence of different pre-Andean geological terranes. We also find evidence for different average crustal models for those terranes. Better-fitting synthetic seismograms result using a higher P-wave velocity, a smaller average S-wave velocity and a thicker crust for seismic ray paths travelling through the crust of the western Sierras Pampeanas (Vp = 6.2-6.4 km s −1 , Vp/Vs > 1.80, th = 45-55 km) than those of the eastern Sierras Pampeanas (Vp = 6.0-6.2 km s −1 , Vp/Vs < 1.70, th = 27-35 km). In addition, we observed an apparent distribution of reverse crustal earthquakes along the suture that connects those terranes. Finally, we estimated average P and T axes over the CHARGE period. The entire region showed P-and T-axis orientations of 275 • and 90 • , plunging 6 • and 84 • , respectively.

S U M M A R YBecause earthquakes on large active thrust or reverse faults are not always accompanied with surface rupture, paleoseismological estimation of their associated seismic hazard is a difficult task. To improve the seismic hazard assessments in the Andean foreland of western Argentina (San Juan Province), this paper proposes a novel approach that combines structural geology, geomorphology and exposure age dating. The Eastern Precordillera of San Juan is probably one of the most active zones of thrust tectonics in the world. We concentrated on one major regional active reverse structure, the 145 km long Villicúm-Pedernal thrust, where this methodology allows one to: (1) constrain the Quaternary stress regime by inversion of geologically determined slip vectors on minor or major fault planes; (2) analyse the geometry and the geomorphic characteristics of the Villicúm-Pedernal thrust; and (3) estimate uplift and shortening rates through determination of in situ-produced 10 Be cosmic ray exposure (CRE) ages of abandoned and uplifted alluvial terraces. From a structural point of view, the Villicúm-Pedernal thrust can be subdivided into three thrust portions constituting major structural segments separated by oblique N40• E-trending fault branches. Along the three segments, inversion of fault slip data shows that the development of the Eastern Precordillera between 31• S and 32• S latitude is dominated by a pure compressive reverse faulting stress regime characterized by a N110• ± 10 • E-trending compressional stress axis (σ 1 ). A geomorphic study realized along the 18 km long Las Tapias fault segment combined with CRE ages shows that the minimum shortening rate calculated over the previous ∼20 kyr is at least of the order of 1 mm yr −1 . An earthquake moment tensor sum has also been used to calculate a regional shortening rate caused by seismic deformation. This analysis of the focal solutions available for the last 23 yr shows that the seismic contribution may be three times greater than the shortening rate we determined for the Las Tapias fault (i.e. ∼3 mm yr −1 ), suggesting that the San Juan region may have experienced a seismic crisis during the 20th century. Moreover, the ramp that controls the development of the Eastern Precordillera appears to be one of the main seismic sources in the San Juan area, particularly the 65 km long Villicúm-Las Tapias segment. A first-order evaluation of the seismic hazard parameters shows that this thrust segment can produce a maximum earthquake characterized by a moment magnitude of ∼7.3 (±0.1) and a recurrence interval of 2.4 (±1.5) kyr. This part of the Villicúm-Pedernal ramp may have ruptured during the M s = 7.4, 1944 San Juan earthquake producing very few surface ruptures and only distributed flexural slip deformation on to the Neogene foreland bedding planes between the Eastern Precordillera and Pie de Palo.

The Andean foreland of western Argentina (28°S–33°S) corresponds to back arc deformations associated with the ongoing flat subduction of the Nazca plate beneath the South American lithosphere. In this paper, short‐ and long‐term tectonic regimes for this Andean region are investigated and compared to NUVEL‐1 and the new GPS‐derived velocity field for the Andes Mountains. From this comparison, it appears that (1) the slight convergence obliquity at the plate boundary is accommodated by the subduction zone itself, precluding any deformation partitioning at lithospheric scale, (2) clockwise rotation of σ1 and P axes within the Andean foreland of western Argentina suggests that Plio‐Quaternary deformation partitioning may occur in the upper plate at crustal scale, and (3) comparison of seismic rates of shallow deformations determined from moderate to large earthquakes localized between 22°S and 36°S shows that the amount of shortening is about the same (2–4 cm yr−1) between the forearcs of the 30° dipping slab segments and the back arc of the flat segment. All together, those observations are in close agreement with the idea that interplate coupling in flat subduction settings may play a key role in the present‐day transfer of plate boundary forces. Consequently, the Andean back arc of western Argentina might be regarded as an obliquely converging foreland where Plio‐Quaternary deformations are partitioned between strike‐slip and thrust motions that are localized in the east verging thin‐skinned Argentine Precordillera and west verging thick‐skinned Pampean Ranges, respectively.