We Collected Data From 1990 To 2010 On Proposals For Grants Supported By U – Communications Of The Acm

Computer Department Science does not award final grades of CR in any CSCI courses to be used ward the big, from CSCI 13500 through the 400 level courses.

You might be expected to maintain a minimum grade point average of 0 in all courses required for the huge.

Required Each and elective courses which count ward the huge must be completed with a grade of C or better. Then the following credits, you must complete key Education Requirements as set forth in the Hunter College Undergraduate Catalog. Landscape of CS research.

For the remaining categories, the fastest growing were publications in information systems.

The author defined keywords contributing to the write were control theory and logic.

After 1994 publications fraction in the mathematics of computing category shrank considerably, a lot of ACM records ‘20092010’ For ACM. Publications in every category increased year over year.

Likewise, the IEEE dataset showed the fastestgrowing research area was information science and information retrieval.

We looked at the CS evolution research landscape 19902010″. The most frequently used ‘authordefined’ keywords were Internetrelated. Usually, We attribute the write to a shift of focus from fundamental problems to challenges specific to an area with which such publications are increasingly associated. We conclude while funding ain’t essential in the initial growth in a CS research topic, So it’s essential for maintaining research momentum and researcher interest. While contributing to the observed pattern, since novelty has usually been prized in publications and grant applications, authors tend to stress novel facts of their work in abstracts and keywords. Notice that opposite pattern is usually at least twice less frequent. Same pattern was reflected in grants number awarded for every pic every year.

While overall trends provide a clear direction picture any pic was always taking, publications fraction on every pic oscillates from year to year to point the direction of improvements in one year successively reversed in subsequent year. Therefore in case a research pic bursts looking at the NSF grants first, look, that’s, Undoubtedly it’s possibly to burst in publications within a few years, We looked for strong evidence of money preceding research. While confirming that sustained NSF funding probably was essential for maintaining interest in a given topic, During a NSF burstiness period, publication burstiness scores were more going to increase than decrease. Needless to say, in reverse case, nearly identical to in ACM dataset. All delays have been one year longer than in ACM NSF pair, resulting, we conjecture, from a larger ratio of computer engineering pics in IEEE than in ACM and presumably to a larger fraction of support for IEEE publications from ‘non NSF’ sources. Accordingly, research number papers published in CS conferences and journals was increasing rapidly for past 3 decades. We aim here to identify more precisely relationships betwixt funding and publications about newest topics, even though plenty of funded programs are probably developed in reliable collaboration with leading researchers. As a result, scientific research is increasingly influenced by funding potentials, with growing emphasis on externally funded research in most universities. Computer science is an expanding research field driven by emerging application domains and enhancing hardware and software that eliminate old enough bottlenecks as they create newest challenges and prospects for CS research. Ok, and now one of most crucial parts. The analysis as well revealed data mining has probably been more broadly used than information retrieval.

Text mining is temporally about one and the other information retrieval and data mining.

For instance, wireless sensor networks are always temporally about simulation, security, and clustering in bursty order periods, Further analysis identified keywords related to any bursty period that burst together.

Latest is always used mainly with ‘Web related’ topics, former has been used with computational science, Web mining, time series mining, and security. Known This order corresponds to the location temporal evolution that primarily focused on simulation of networks, so on security, and ultimately on clustering algorithms. That’s where it starts getting practically serious. Top 20 and bottom 20 trends 1990 2010 and ‘20062010’ from ACM and IEEE datasets.

Figures 3a, 3b, 3c and 3d.

First, all ‘realtime’ systems and parallel processing were associated with scheduling, later expanding to genetic algorithms and embedded systems.

In 1999 bursty period, scheduling correlated with genetic algorithms, parallel processing, performance evaluation, embedded systems, approximation algorithm, multimedia, quality of service, optimization, and heuristics. Fact, whenever scheduling as well correlated with multimedia, online algorithm, and fairness, In last few years of its bursty periods. Multiple bursty periods for a keyword involve interesting temporally correlated terms.

Now look. There were 4 bursty periods for keyword scheduling. Besides, In the period ‘2001 2006’, such keywords, listed in identical order, were approximation algorithms, multimedia, online algorithms, real time, embedded systems, fairness, multiprocessor, quality of service, and genetic algorithms.

Trend analysis.

We fit trend lines to data from the preceding 2 to 6 years if you are going to predict keyword fractions for the following year.

We generated a trend line for every keyword fraction and used its slope for ranking. Here, we analyze research trends through linear regression trend line and changing popularity of pics depending on papers fraction including a given keyword in any year. An increase of 10percentage in published number papers in a given pic in the ACM dataset was followed with 75 probability of an increase in the overall amount of NSF grants awarded on quite similar topic. While approximately 1 core researchers were connected with every cluster, the evolutionary average length chain was five years.

Every 5 years or so, entirely a few stable researchers typically remained from an original research group.

Figure five plots communities number that survived from one year to next in the ACM and IEEE datasets.

We used authors networks represented as a bipartite graph in which any node representing a paper has edges to all nodes representing paper’s authors. Remember, the table lists average evolutionary chain length, average cluster size, average size of intersections of 2 to 5 consecutive clusters, and average relative density.h We searched with success for the recovered clusters had lofty average relative density of eight for all datasets. Using framework for analyzing community evolution communities developed by Goldberg et al,six we tracked evolution of CS researcher communities by searching for overlapping communities over consecutive time periods. Communities of CS researchers. This finding was consistent with the typical university team consisting of one or 2 stable faculty and 4 to 4 graduate students and postdocs joining and leaving continuously. Longer delay shows if NSF initiates a tally new area, the increase in publications has been delayed by the time researchers need to obtain grants and start research leading to publication.

Whenever becoming bursty in 1999 for NSF and in 2000 for ACM, For another 16 of cases, it was reverse, examples of bursts appearing first in the NSF dataset are probably data mining and search engine.

While, in 75 of such cases the keyword happened to be bursty in the NSF dataset unto it happened to be bursty in the ACM dataset.

Therefore if a keyword turned out to be bursty in ACM data first, in the reverse case. For a ACM NSF pair, it proven to be bursty in NSF four years later on average. Finally, reverse included bioinformatics and semantic Web. For any pair of datasets, we analyzed in which one a keyword’s bursty period begins first and hereupon how long it needs for the keyword to proven to be bursty in another. Now please pay attention. For keywords with more than one bursty period, we looked at their burstiness score in any bursty period, after that, tabulated cases percentage in which the later burstiness scores increased, decreased, or was unchanged.

Burstyperiod’ analysis.

To evaluate research influence funding on publications and vice versa, we extracted from ACM, d IEEE, and public Science Foundation datasets the bursty periods of author defined keywords depending on the burstiness score for a time period12 defined as where w usually was the keyword/topic of interest, t is a time period, dt has been a document created during time t, d is any document, and T was usually tal time over which all documents were created.

20 every segment with positive score corresponds to a bursty period, the burstiness score measures how frequently w is always in t compared to its occurrence in A positive score implies w appears more oftentimes throughout the bursty period t than over tal time We recovered burstiness maximal segments scores in sequence of documents using the ‘lineartime’ maximum sum algorithm.

In all datasets, we observed that if a trend on the basis of 1 data years had a positive slope, or publications fraction increased from the previous year to current year, therefore the subsequent year fraction declined.

We as well used trend line on the basis of the NSF dataset to predict fractions for following year in ACM and IEEE datasets.

All accuracy these models was less than 50percentage, results show trend line has been an unsuccessful predictor, as was usually using ACM and IEEE trends to predict the tal number of grants awarded by NSF. During 19902010, 87 research topics, including image analysis, data transmission, and operating system, were bound with up to mentioned 3 14 topics. Notice, whenever in the course of the period 1990 2010″, mostly 32 were persistent topics. That’s where it starts getting actually intriguing, right? Besides 3 most frequent topics, 11 others had persistent connections with multiple research pics nearly any year 1990 2010, including programming language, artificial intelligence, clustering, image processing, computer vision, network, distributed system, pattern recognition, robotics, software engineering, and integrated circuit.

Figure two includes the research pic subnetworks culled from ACM dataset by Map Generator software package4 for the security and the multimedia subnetworks looked for in 1995 and for World Wide Web and Internet subnetworks searched with success for in In 1995, Web was used as a keyword associated mostly with multimedia and information visualization, whereas information retrieval was used mostly with Internet.

Privacy and security have happen to be significant in Web context, while semantic Web, Web 0, Web service, and XML proven to be huge Internet topics, since 2005.

While Internet was used mostly with network, by later 2000s, Web was used mostly with data mining and information retrieval, protocol, and routing. See Hoonlor et al dot nine for details, Researchers in human computer interaction remain active longest, accompanied by researchers in computer architecture.

Whenever computing methodologies, and information systems tended to stay active in these categories for a longer time, Researchers in computer systems organization.

Unlike in, we looked with success for it ugh for researchers to publish in artificial intelligence and programming language year after year, say, ‘human computer’ interaction.

While computing milieu, and data keywords, indicating the authors in these categories were either mostly briefly involved in multiple research pics or mostly briefly collaborated with somebody else from these categories, besides a considerable first year write rate, we learned a relativelyrelatively shorter ‘half essence’ time notably for computer application. Figure Landscape of CS research fields, on the basis of conferences 1990 2010″, for ACM and IEEE datasets, including raw numbers and percentage of publications for every keyword every year. Write information impact systems, we extracted the p 25 research pics from ACM and IEEE and quantified the results in 1 ways.

See Hoonlor et al dot nine for detailed results, has been documents number and nk, d is the overall number of times k appears in d.

Most publications in collaboration, data mining, information retrieval, machine studying, privacy, and XML appeared 2000 2010 and showed noteworthy trends in CS research.

And therefore the related pics were present since late During period 1990 1997, 376 NSF grants and 10 IEEE papers mentioned NSFNET in their abstracts, but solely 3 ACM papers included it as a keyword, terms Internet and World Wide Web did not appear in any publication until 1995. Different terms appeared in NSF dataset before prodigy was bursty over the period 19911992″ and TCP/IP over the period ‘19901993’. Keep reading. In late 1990s, interest shifted to Web, information retrieval, and ‘computer supported’ cooperative work. Needless to say, ‘mid2000s’ saw strong interest in sensor network and later in wireless sensor network, Throughout the 2000s, the areas most connected to others were design, usability, and security.

While confirming our earlier observation that while a peculiar research pic can be crucial enough to be mentioned in an article’s abstract, it may not represent the article’s key research contribution, In ACM networks using author defined keywords, no persistent link appeared ‘1990 2010’.

Lack of link persistence was usually evident for algorithm and database topics.

Another example of lack of link persistence is the neural network node in IEEE and ACM networks. In ACM networks, it under no circumstances achieved this status. You see, In the former, neural network was a central node in virtually every year. In the later 1990s, user interface, scheduling, and multimedia were tied with lots of CS research fields. On p of this, For any such pair and every year 1990 2010″, we searched for year in which entries number changed compared to most of previous 5 years in first database.

For every rethink, we searched in other dataset for a corrections in majority of next 4 years.

The relative rethink values ranged from -five to 5, that we grouped into bins of size We counted review frequency in one dataset followed by a corrections in the next.

We analyzed NSF dataset versus ACM and IEEE datasets and vice versa. Remember, His big shift was from planning and Web intelligence to the semantic Web. In 2000s, he focused mainly on semantic Web and for the past few years on huge data and almost any 4 to ten years by broadening their scope and branching into modern applications. Less frequently, apparently once in a career, there’s a huge shift to a completely new area. While keeping algorithm as a node greatly lowered separation degree betwixt research pics and created a central node, or one with biggest tal weight of its edges, dominating other research topics, For IEEE dataset. We extracted 1 keywords sets. We were able to monitor when connections betwixt 1 fields occurred or changed, since we looked back over the period ‘1990 2010’. We performed the network analysis on the algorithm pic first, hereafter removed the algorithm node from the network being that term was used in virtually any CS research paper to describe how data was probably processed. On p of this, Network of CS research. Faculty member has been oftentimes active in more than one area.

Since novelty has always been prized, authors tend to pursue newest directions in their research, as reflected in an article’s abstract and keywords, further contributing to the observed pattern.

Authors tend to publish in very similar fundamental research category for at most entirely a few years.

Most authors manage to publish at most once a year in a particular research field. With last routinely changing pics after leaving a team, This solidary with an academic model research team in which permanent faculty represent entirely a tiny overall fraction team of faculty. Thus postdocs. Simply a fraction of them continues to publish in identical field year after year for a long time. Anyways, Topics with this particular increase included data mining, information extraction, and wireless network.

For NSF dataset versus either ACM or the IEEE dataset, a 10 or greater increase in NSF number grants awarded for a given pic from previous few years was followed by an increase in the actual number of published papers on the pic of at least 10 in next 2 years and 20 in the next 5 years. Examples were e government, groupware, and knowledge management, an increase of 10percent in published number papers in a given pic in the ACM dataset was followed with 75 probability of an increase in the overall number of NSF grants awarded on identical topic. Apirak Hoonlor was usually an instructor in Information Faculty and Communication Technology at Mahidol University, Bangkok. Using sequence mining,24 network extraction and visualization,18 bursty words detection,12 clustering with bursty keywords, c,ten and network evolution,six we investigated rethinking over time in the CS research landscape, interaction of CS research communities, similarities and dissimilarities betwixt research topics, and funding impact on publications and vice versa. See Hoonlor et al dot nine for results, We figured out a list of terms clustered gether with network connectivity in period 20062010″ though not connected in at least 1percentage of the documents dot 24 We examined p ten frequently used keywords at numerous degrees of separation.

Was not used to describe similar research project mostly enough, simulation was instead clustered with information retrieval. Therefore filtering. Machine practicing, and artificial intelligence.

Data mining was rarely used to describe research about mobile networks and its related research topics.

While During period ‘2006 2010’, despite they were either not used or used usually rarely by authors to describe their research in simulation. Let’s say, ‘human computer’ interaction focused mainly on interaction design, visual design, and ‘computer supported’ cooperative work in the 1990s and augmented reality, computer vision, human factors, and ubiquitous computing in the earlier 2000s, so ultimately shifting to public media, studying, computermediated communication, and tangible user interface in the late 2000s, Note while researchers could continue to publish in one area for a long time, this place itself evolves and may cover unusual pics during special time periods. From the late 1990s to mid 2000s, she focused on grid computing from a bottom up perspective.

Another researcher in this area, Francine Berman of Rensselaer Polytechnic Institute, Troy, NY, characterized her work in 1980s as ‘topdown’ mathematical modeling of mapping and scheduling troubles.

In the later 1990s, her papers used such keywords as data driven, performance, and algorithms.

She described this evolution as a broadening and branching approach. Whenever saying he investigates any subject in 4 year phases discovers an open field rather frequently associated with previous work, since 2002, he has investigated privacy and security problems, including cybersecurity. Whenever analyzing data, and extracting information from Web and from wireless networks, In earlier 2000s, he focused on target tracking. As a result, In late 1990s, George Cybenko of Dartmouth College, Hanover, NH, studied ‘big performance’ computing and classification by neural networks.

In the late 1990s, he shifted to mobile agents, mobile networks, and simulations. One exception was a fundamental shift in 1992 when moving from one university to another. Edge thickness represents strength of interaction, Figure Research clusters, or subnetworks, in 1995 and 2001. Oftentimes frequency is number of publications on any pic any year; fraction always was publications percentage on any pic any year, Figure 3 CS views research, ‘1990 2010’, on the basis of the ACM and and IEEE and datasets. See Hoonlor et al dot nine for categories complete list, we used the CS conferences listed in Wikipedia23 to categorize every paper in IEEE and ACM. With every pic representing a set of CS conferences, e when a conference covered 2 topics. Figure four indicates growth of approximately 11percent in publications for most CS pics year over year.

IEEE dataset did not show a considerable decrease between p and bottom trends as long as research pics appear in the abstracts longer than do ‘authordefined’ keywords, unlike the ACM dataset.

We could not statistically compare the growth in special areas due to immense differences in conferences number in any field and number of papers published in any conference.

Figure three includes the p 20 up and p 20 down trends for the period 19902010 and for the period 2006 2010″ for ACM and IEEE. For the IEEE dataset, it was 29 /64percent/7percentage, respectively, and for the ACM dataset, it was 12 /85percent/4. Understand if bursty period appeared first in IEEE dataset, for interleaved or overlapped bursty periods in the NSF and IEEE datasets, the following NSF bursty period had a higher/lower/equal burstiness score in 31percentage/22 /47 of the cases.

In reverse case, it was 36 /10 /55.

For a keyword with multiple bursty periods in the NSF dataset, the following bursty period had a higher/lower/equal burstiness score in 37 /51percentage/12percent of the cases.

In the reverse case, for following ACM bursty period, the numbers were 8 /8percentage/84, the ACM same analysis and NSF datasets showed the following NSF bursty period had higher/lower/equal burstiness score for 38 /14percentage/48percent of the cases. Later, we used 408 research pics included in 16 Wikipedia articles on CS research areas identified in the key Wikipedia CS article23 to classify 458395 papers in the IEEE dataset. IEEE Xplore dataset included identical information but lacked a pic classification like the ACM CCS.

For NSF dataset, we retrieved titles, start dates, and abstracts of 21687 funded grant proposals.

Reason for a great percentage of equal burstiness scores was probably that a bursty period in one dataset was oftentimes a bursty subset period in another.

Burstiness scores tend to decrease in the periods following a bursty period in NSF dataset. Usually, while confirming sustained NSF funding is essential for maintaining interest in a given topic, during a NSF burstiness period, publication burstiness scores were more gonna increase than decrease. Whenever contributing to observed pattern, since novelty is probably prized so extremely in publications, authors tend to stress newest facts of their work in abstracts and keywords. We used the ACM CCS and authordefined keywords to respectively study the broader and static versus finer and dynamic CS views landscape and trends. Whenever yielding smaller research groups than if we had used just ACM CCS alone, In another analysis, we used solely authordefined keywords to identify relationships betwixt researchers. We used ACM, IEEE, and NSF datasets from which we collected data on publications from 1990 to 2010.a For the ACM dataset, we extracted papers number listed in p 1998 categories ACM Computing Classification System, or CCS.b the ACM dataset included authors, title, abstract, year published, publication venue, author defined keywords, and ACM classification categories for every of the 116003 articles.

Trend analysis kinds of datasets types, from medical17 to weather15 to stock markets dot five lots of publications track research trends, analyze a particular impact paper on development of a field or topic, and study the relationships between exclusive research fields.

Porter and Rafols16 analyzed citation information to search for evidence of collaboration across fields in scientific research.

Science22 Web has collected data since 1900 on nearly 50 million publications in multiple scientific disciplines and analyzed it at a variety of levels of detail by looking at the overall trends and patterns of emerging fields of research and the influence of individual papers on related research areas. Commonly, Over past decade, besides Science Web, studies have as well investigated the overlap and evolution of community communities around a field or a topic. Thence, Rosvall and Bergstrom1819 explored methods and visualizations for scientific research and analyzed every impact research area quantified by collective ‘cross disciplinary’ citations of every paper. We looked with success for if an uncommonly lofty frequency of a specific pic has been included in publications, funding for pic in general increases.

Several studies have focused on challenges, directions, and landscapes in specific CS fields27 and on specific CS pics dot 821 here, we probably were interested in studying about CS evolution research.

Whenever finding usually a short fraction of authors attribute their work to identical research area for a long time, reflecting an emphasis on novelty and frequent correction in academic research teams, We analyzed CS researchers and communities.

While betwixt research topics, We collected data from 1990 to 2010 on proposals for grants supported by the international Science Foundation14 and on CS publications in the ACM Digital Library1 and IEEE Xplore digital library dot 11 We analyzed research communities. Relations betwixt awarded grants and progress in communities and trends. Remember, We attempt to capture that vibrant coevolution here. Naturaly, researchers proceed with their evolution artifacts by adjusting their research interests. CS is an atypical academic discipline in that its universe is evolving so rather fast, at a speed unprecedented for engineering., with its focus constantly moving to modern challenges due to newest technological developments, our work highlights the dynamic CS research landscape. With logic and control theories contributing most to decline, TFIDF and DF values showed information rise system contributed to the key interest in data mining. ‘Webrelated’ topics, whereas mathematics of computing continued to decrease year over year throughout the same period.

Authors thank Francine Berman and James Hendler of Rensselaer Polytechnic Institute, George Cybenko of Dartmouth, and Jack Dongarra of Tennessee University, Knoxville, for discussions on evolution of their research interests.

The government probably was authorized to reproduce and distribute reprints for government purposes notwithstanding any copyright notation here on.

The research was sponsored by the Army Research Laboratory and accomplished under Cooperative Agreement Number W911NF 09 2 views and conclusions here are authors those and shouldn’t be interpreted as representing the official policies, either expressed or implied, of Army Research Laboratory or government. The authors as well thank Katie Bahran for any pic a node and connected 3 nodes with a weighted edge representing abstracts number that mention all adjacent topics, to create research pic networks.

For the ACM and IEEE datasets, we created 3 data indexes authors and their publication venues and papers and their keywords/topics finding, in the analyzed period, publications number grew approximately 11percent yearly over those 20 years.

While pulling in a variety of researchers working on data mining, information retrieval, cloud computing, and networks, Our investigation as well searched for the Web has turned out to be an attractive source of data and application testbeds.

Even when concept was introduced shortly after standardization of TCP/IP protocol suite in late 1980s, Most research about Internet had been done since 2000.