3. THE METHODOLOGY This section describes the overall process, which was followed in order to get the analysis result that was envi-sioned. The process starts with simplification of knowl-edge to generate “List of Topics” associated with the Sound & Music Computing community, secondly a hand-ful of “Keywords” were created to map each paper to one or the other topics in the list. Finally, statistical analyses of the papers were done. 3.1 Knowledge simplification To classify all the papers of the previous SMC Confer-ences on the basis of “Research Topics”, we started with the topics for call for papers for the upcoming Sound & Music Conference. Since these topics were very specific in nature, we classified them into broader topics. The classification, of the original topics into broader ones was done on the basis of “Similarity in Concepts”. This led to the construction of a hierarchical classifi-cation, constituting two levels of simplification: 1. “Middle Level Topics” 2. “High Level Topics” The mappings of these two levels of topics are presented in Table 2. 3.2 Keyword building Since the papers presented in the SMC Conference did not have “Keywords” in the full text, there were two options for building a keyword repository for the papers:
1. Automatic extraction of keywords from the ab-stract of the papers using a probabilistic mixture model as introduced in [5]. 2. Use of CSCW methods using Google docs as discussed in [6] to collaborate with other re-searchers of the field to have a consensus on a set of keyword for each topic. For the current work, we used the second method for building the keyword list for each middle level topic. The consensus was reached using cross validation of the key-words between each researcher in the second pass of the questionnaires. Furthermore, relevance of each keyword was checked by searching for the keyword in the SMC papers as well as searching for papers using the same keyword in Google scholar. 3.3 Search Mechanism The search mechanism attempts to assign a topic to each paper, based on the keyword. The entire process can be described as below: 1. Search every keywords of each topic in the ab-stract & title of each paper. 2. If a particular keyword is present in the abstract or title, we add that keyword & the associated topic as a contender for classifying that paper. 3. Once the keywords are mapped in a particular paper, we count the total number of occurrences of each keyword & the total presence of key-words from each topic.
Middle level topics High level topics Name ID 3D sound/music, Sound/music signal processing algo-rithms, Digital Audio Effects, Musical sound source separation Topic 1 Processing of sound and music signals Sound synthesis, Spectral modeling synthesis, Physical modeling for sound generation Topic 2 Music information retrieval, Musical pattern recogni-tion/modeling, Computational musicology, Technologies for the preservation, access and modeling of musical heritage, Automatic music transcription, Musical sound source separation and recognition Topic 3 Understanding and modeling sound and music Music and emotions, Sound/music and Neuroscience, psychology, psychoacoustics, Sound/music perception and cognition Topic 4 Interfaces for sound and music Interfaces for music creation and fruition, Gesture con-trolled audio systems, Mobile music, Interactive per-formance systems, Musical performance modeling Visu-alization of sound/music data, Sonic interaction design Topic 5 Web 2.0 and music, Networked music generation Topic 6 Assisted sound and music creation Computer environments for sound/music processing, Automatic music Topic 7 Table 2. Hierarchy between High & Middle Level Topics

4. The topic with the maximum number of key-words present in a paper is decided to be the relevant topic for that paper. 5. If more than one topic has equal presence in a paper, we classify the paper as ‘Multiple Topic’. 6. If none of the keywords could be mapped to a paper, we label the paper as ‘Unknown Topic’. This process is carried out to classify a paper to a Middle Level topic using the set of corresponding keywords for each topic. For the re-classification of each paper based on High Level Topics, we use the relationship between the Middle Level & High Level topics as depicted in Ta-ble 2. Furthermore, if there were discrepancies in the high level classification, we assigned those papers as ‘Unclas-sified’. Topic ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ALL Multiple 13 4 5 15 8 11 56 No topic 4 0 1 2 3 3 13 Topic 1 3 2 1 3 7 3 19 Topic 2 9 13 7 13 9 10 61 Topic 3 7 0 5 4 1 13 30 Topic 4 2 0 1 0 0 2 5 Topic 5 5 11 4 18 3 15 56 Topic 6 3 1 1 4 3 5 17 Topic 7 0 0 0 1 0 0 1 Table 3. Year wise distribution showing the absolute number of papers for each Middle Level Topic. Topic ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ALL Unclassi-fied 13 4 5 13 6 10 51 No topic 4 0 1 2 3 3 13 As-sisted… 3 1 1 5 3 5 18 Inter-faces… 5 11 4 18 3 15 56 Process-ing… 12 15 8 18 17 14 84 Under-stand-ing… 9 0 6 4 2 15 36 Table 4. Year wise distribution showing the absolute number of papers for each High Level Topic. 4. RESULTS The results that we obtained after the analysis are pre-sented in this section. For better aesthetics of the plots & charts, we have used aliases for each Middle Level Topic.
4.1 Participation ratio for each relevant topic Since each paper was classified as either a topic or multi-ple topic or unknown topic, we can deduce the distribu-tion of each topic in each year’s conference as well as in the overall conference till date. Figure 1 shows the distribution of each Middle Level Topic in the overall conference history, while Table 3 is used to visualize the distribution of each Middle Level Topic in each edition of the conference. Likewise, Figure 2 displays the distribution of the High Level Topics in all years of the conference taken together, whereas Table 4 is used to show the distribution of these topics in each conference year.
Figure 1. Publication distribution for Middle Level Topics for the overall conference till date.
Figure 2. Publication distribution for High Level topics for all years taken together. 4.2 Trends for each level of topics over the entire con-ference history The change in the number of papers for each topic over the years is presented both for Middle Level & High Level topics in Figure 3 and Figure 4 respectively.

Figure 3. Middle Level Topic trend.
Figure 4. High Level Topic trend.

4.3 Closeness between the topics Since we used decision based approach for assigning a paper a relevant topic based on the “presence of key-word”, we observed many papers which were classified as particular topic but had a fair amount of keywords of other topics were present as well. This can be used to deduce the closeness of a topic with others. The cross infiltration (presence) of each Middle Level topic in every other for the overall conference publications is showed in Table 5. Topic 1 2 3 4 5 6 7 Topic 1 58.55 18.42 3.95 0.66 13.82 4.61 0.00 Topic 2 9.37 59.10 7.21 1.26 17.66 5.23 0.18 Topic 3 6.48 10.65 60.19 4.63 16.20 1.85 0.00 Topic 4 12.20 12.20 12.20 43.90 17.07 2.44 0.00 Topic 5 15.43 9.88 3.09 2.78 51.23 17.44 0.15 Topic 6 2.26 15.04 2.26 3.76 24.06 52.63 0.00 Topic 7 0.00 0.00 0.00 0.00 0.00 0.00 100.00 Table 5. Presence of each topic in each other (Middle Level). 4.4 Keywords & their relevance As the keywords play a pivotal role in the overall proce-dure that we presented here, we found out the popularity of each individual keyword irrespective of the topic they represent in all the papers published in the SMC Confer-ence till date. A keyword cloud representing the popularity or pres-ence of these keywords is plotted below as Figure 5. The most frequent 50 keywords are shown with a font size that reflects this popularity. Frequency values range from 6 to 119 occurrences.
Figure 5. Keyword Cloud
5. CONCLUSION In this paper, we analyzed the proceedings of the past SMC Conferences, tried to categorize each published paper into one of the proposed 7 Middle Level & 4 High Level Topics so that the trend of the SMC Conferences could be identified and justified. To start with we noticed that 482 authors have par-ticipated in the SMC Conferences till 2009 and out of those, 68 authors have publications in more than one edi-tion of the conference. For e.g. we found out that Topic 2: Sound synthesis, Spectral modeling synthesis, Physical modeling for sound generation and Topic 5: Interfaces for music creation and fruition, Gesture controlled audio systems, Mobile music, Interactive performance systems, Musical performance modeling Visualization of sound/music data, Sonic inter-action design remains the most popular topic throughout the conference with a combined share of ~77% in SMC Conference 2005 and ~45% overall. Of all the confer-ences till date, the share of Topic 1: 3D sound/music, Sound/music signal processing algorithms, Digital Audio Effects, Musical sound source separation was highest in 2008 about 20% and Topic 3: MIR & others had a con-siderable share in the 2009’s conference with about 21% publications. From the participation ratio of each Middle Level topic in each year, we find the following trends in the evolution of some topics over the years: 1. Web 2.0 grows since 2005, this can be justified by the fact that web 2.0 evolved a lot since that time, so it attracted much research in the recent years. 2. Sound synthesis/ signal processing has a slight decline in percentage in the recent years this might be because the growing popularity of other fields. 3. Since the theme of the 2008 conference was "Sound in Space", the abrupt increase in the number of publications of the topic “3D Audio” for that year is justified. From the closeness analysis of each topic Vs the oth-ers, we could clearly see that Topic 1 & Topic 2 are closely related to each other, so our classification of grouping them together in the higher level of classifica-tion is fairly justified. Although we have grouped Topic 6 & Topic 7 together, this is not fairly justified by the data presented in Table 5. This is due to the fact that there is a hairline difference between the last two high level topics and thus Topic 5 & Topic 6 are also closely related as depicted in the same table. Alternatively, Topic 5, 6 & 7 could be regrouped to a new High Level topic as well. And finally looking at the Keyword cloud, we could see that the popular keywords from the set we had, are synthesis, analysis, instrument, realtime, voice, net, etc.

Since the overall methodology relied on text mining and knowledge simplification using which we classified a dataset of nearly 31000 words with a keyword set of 117 keywords, the evaluation of the system is tough. Moreo-ver, the evaluation is also hampered by the fact that there were no keywords provided by the authors in the SMC papers to cross check with. To conclude with, we would like to highlight that only 5% of the papers were of unknown topic and about 21% of the papers were of Multiple Topics (unclassified), this correlates to the fact that Sound & Music Computing is highly inter-disciplinary in nature. Another point to take into account is that these conclusions have been de-duced from the last 6 SMC proceedings, which might not represent enough data to support them. Also, we would like to continue to explore the presence of research groups of different universities in the SMC Conferences based on publications and how papers, authors & research topics could be classified together on the basis of co-authorship, citations and bibliographic links. 6. REFERENCES [1] A Roadmap for Sound and Music Computing. Version 1.0. , 2007. Available at: http://www.smcnetwork.org/roadmap [2] J.H. Lee, J.S. Downie and M.C. Jones: “An analysis of ISMIR proceedings: Patterns of authorship, topic, and citation”, 10th International Society for Music Information Retrieval Conference, pp. 57-62, 2009 [3] G. Widmer et al, : “The ISMIR Cloud: A decade of ISMIR Conferences at your fingertips”, 10th International Society for Music Information Retrieval Conference, pp. 63-68, 2009 [4] Official drupal website available at http://drupal.org [5] A. Velivelli, B.Yu and C. Zhai: “A cross-collection mixture for comparative text mining,” Proceedings of ACM-KDD, pp. 743–748, 2004. [6] S. Dekeyser and R. Watson: “Extending Google Docs to Collaborate on Research Papers”. Technical report, The University of Southern Queensland, Australia, 2006.