Discovering Similar Cities Using Text Mining: A .
International Journal of Scientific Engineering and ScienceVolume 1, Issue 12, pp. 8-14, 2017.ISSN (Online): 2456-7361Discovering Similar Cities Using Text Mining: ARecommendation Application for TurkeyYunus Doğan1, Yunus Turdu21, 2Department of Computer Engineering, Dokuz Eylül University, Izmir, TurkeyAbstract— The purpose of this study is to show that it is possible to benefit from the use of text mining to capture alternative cities that aretargeted by a person on touristic journeys. The first process has been to collect the texts containing the descriptions of 100 cities and to convertthem into a dataset form for text mining. Secondly, K-Means and the density-based spatial clustering of applications with noise (DBSCAN)algorithms have been used and compared to obtain similar cities. Multi-layer perceptron, Naïve-Bayes, K-Nearest Neighbor, Decision Tree andSupport Vector Machine (SVM) algorithms have been used to classify these cities. Since Multi-Layer Perceptron yields over 70%, it has beendetermined to be the most successful algorithm for this purpose. The SOM (Self Organizing Map) algorithm has been used to obtain moreconsistent and accurate results of the distribution, and the clusters have been finalized. In analyzing of the application for Turkey, 28 cities inTurkey and 72 other cities have been evaluated and it has been possible to present the cities which have been similar to Turkey as alternatives.For this purpose, the obtained results from text mining have been visualized through a mobile application. The results of the analysis for themobile application have been recorded in a database and presented to the user on the Android platform using the Windows CommunicationFoundation (WCF) web service methods.Keywords— Text Mining, Natural Language Processing, Mobile Application, Web Service, Tourism.I.INTRODUCTIONSome city guide applications or websites use machine learningand data mining technique to present more correct results. Forexample, Triposo city guide application has sent advises as email similar cities according to user's previous presentsadversities according to user data and searches after logginginto the websites (tripadvisor.com.tr/ Tourism-g293969Turkey-Vacations.html). These applications have used usertransaction data and analyzed by using of data miningtechniques to make campaigns depend on these results.Tourism may improve for by promotions giving by advisorwebsites. There are many applications for promotions ofdifferent cities. The main goal of our study is that finding mostsimilar cities to the city selected by user with usage of textmining. For example, a user in Turkey wants to visit a city inAustralia. When the user searches cities according to his/herinputs in this application, the system returns cities in Turkeyas alternatives to the selected city in Australia. Thus, it canfind similar cities in Turkey and the idea of the user canchange not to go to a far city. For this purpose, the data ofdifferent cities have been collected; this data have been usedand analyzed by using of text mining techniques. Finally, ithas presented these results.Firstly, different cities data have been collected fromdifferent websites (wikitravel.org and wikivoyage.org). Afterthat, the operations of the natural language processing (NLP)have been implemented with a tool. It has been generated fordata preparation. Trivial words, punctuation marks and stopwords have been removed in text files by the tool. Theseprepared data have been used for data mining techniques. Theprogram has been decided the most similar cities according tothe results of clustering algorithm. As Microsoft SQL has beenused for data storage and user queries, Android SDK havebeen used for user interface and result representation. As aresult, the obtained clusters contain other cities including thecities in Turkey, and then the application has been found thesimilar cities in Turkey to the selected city and showed theresult set of cities in the mobile application.Weka tool have used for data mining Clustering algorithmmethodologies and the experimental studies and results havebeen explained in next sections in detail. Also, C# tools haveused for data preparation. WCF web service, Microsoft .NETand Microsoft SQL have been used for entity, facade andpresentation layer operations of the applications.II.RELATED WORKSLorenzi et al. [1] have implemented a tourismrecommender system based on textual analysis. Usually,travelers need some advices about where to go and what to do.The system use collected user data for text mining techniques.Some travel agencies use this system to help tourists. Allwords related to tourism called “Ontology” in the system,some words have been discarded in ontology (stop words,prepositions etc.). Travel agencies have created tourismcategories. For example, adventure, tropical, beach, vacation,historical tourism. After that the customer and travel agentstart answer and question in the system. Customers areclassified by applying text mining methods to customermessages. The system suggests the customer according tothese classes.Lau et al. [2] have a study about text mining for the hotelindustry. Business intelligence is important for hotel industry.Many studies propose text mining as a means of informationmanagement. There are many ways to collect data for textmining. Online collected data; different web pages, socialmedia accounts, e-mails, news-groups are web-based datasources. The other way of collecting information is inguidebooks. The hotel-related datasets and words are collectedin the database. Hotel profiles are analyzed based on textmining. There are features to consider when choosing a hotel8http://ijses.com/All rights reserved
International Journal of Scientific Engineering and ScienceVolume 1, Issue 12, pp. 8-14, 2017.(number of room, transportation, price, safety etc.). The mainpurpose of this project is to assist in hotel selection using textmining algorithms. This project was made for the hotels inHong Kong.In another study, Berezina et al. [3] have analyzed onlinehotel reviews by using text mining. The purpose of this articleis to investigate satisfied and unsatisfied hotel ��tripadvisor.com” website have been reviewed. Satisfied andunsatisfied customers have been compared. Some categorieshave identified for the positive review and the negativereviews recommendations. For example, a small distancebetween the beach and the hotel is a positive situation but it isa negative situation that the number of hotel employees is low.CATPAC, a text mining method of determining the frequencyof words has been used in the project. Both positive andnegative reviews have been important for text-mining processand they have been evaluated. For each word, they have beencalculated the number of positive and negative reviews. Thenext process has been Text-Link. Text-Link analysis hasidentified new means of word groups.The main purpose of the article of Segall et al. [4] is thatSAS Text Miner and Megaputer Polyanalyst especially haveapplied for hotel consumer survey data. The increased textualknowledge has caused another increasing for text miningapplications. There are a lot of text mining applications. Thisarticle has compared two selected applications according tohotel data. Hotels have been increased customer servicequality by using text mining methods. These text mining toolshas been applied in the different sectors.Claster et al. [5] have used seventy-million tweets asdataset. Contains a tweet; user information, comment, locationand date information. The article has been interested onlythree locations (Sri Lanka, Thailand and Mexico). Tweetshave been filtered according include terms about tourism.SOM and Naive-Bayes algorithms have been used in thestudy. According to specific dates and locations tweets havebeen analyzed. For example; November and January monthsare summer in Sri Lanka. Therefore, the tweets in thesemonths have been evaluated. Results of this study have beenused in tourism sector.Furthermore, it is possible to investigate some studiesabout tourism using data mining algorithms in literature (e.g.[6], [7] – [8]). For this aim, Dickinger et al. [9] have usedSVM, Guo et al. [10] have used Sequential Pattern MiningAlgorithms, Godnov et al. [11] have used SentimentalAnalysis, Pembeci [12] has used Regression and KernelDensity, Gassiot and Coromina [13] have used Multivariatestatistical analyses and Claster et al. ([14] – [15]) have usedSelf-Organizing Maps, Naïve Bayes and unsupervisedartificial neural networks.III.METHODOLOGIESThe study has kept to the process of “knowledge discoveryin database” (KDD). The first step has been for data selection,the second step has been for data pre-processing, the third stephas been for data transformation, the forth step has been fordata mining and the fifth step has been for patternISSN (Online): 2456-7361interpretation. In this section, the process of KDD is explainedgenerally.A. Data SelectionIn this step, a data collecting operation has been done. Forthis reason, the sites that provide information about the citieshave been searched and the data used in this study have beencollected from Wikitravel.com and Wikivoyage.com websites.B. Data Pre-processingIn this step, NLP operations have been done. The first onehas been the elimination of punctuation marks. Punctuationmarks may change the accuracy of the analysis. Such as“doing.” and “doing” words are not reduced into the samemeaning in the analysis. Therefore, the punctuation markshave deleted to make to word-based processing.The second operation has been the elimination of stopwords. Some of the words in the sentence does not make anysense, like “a, an, the, of called stop words. These wordswould prevent to the analysis to get the right results.Therefore, the stop words have been removed.The third operation has been the changing the words inirregular form into infinitive forms, because the meaning ofsentence may change according to the tense of the Englishverb. For example; “write” word should change as “wrote” or“written” according to related time. All verbs have reduced tothe infinitive forms. For these operations, a tool has beenimplemented in the study.C. Data Transformation and StorageIn this study, firstly, the corpus has been collected in adatabase. Microsoft Sql (MSSQL) has been used as a databasemanagement system. Each city name and information hassaved in this database. When users select any city, the systemsends the most similar city information.Secondly, a “.csv” file has been created for each city andthe words about the city have been saved. The words aboutcities have saved a different “.csv” file and all files have beencompared using some text-mining methodologies. The systemhas decided the most similar files according to result of textmining algorithms.D. Data MiningAt comparison phase, Weka tool has been used. The subfunctions have been different data mining algorithms such as;K-Nearest Neighbour, Multi-layer perceptron, Naïve Bayes,Decision Trees, Support Vector Machine (SVM), SelfOrganizing Map (SOM), K-Means and DBSCAN.E. Pattern InterpretationIn this phase, the outcomes from data mining algorithmshave been evaluated and validated. The success rates of thealgorithms have been compared and the pattern of the mostsuccessful algorithm has been assumed for the next phase ofthe mobile application.The clusters with the cities have been stored in the factdatabase and WCF web service methods, which have used thisdatabase to connect the database with the mobile application,have been implemented.9http://ijses.com/All rights reserved
International Journal of Scientific Engineering and ScienceVolume 1, Issue 12, pp. 8-14, 2017.Fig. 1 shows the architectural view of the system.Fig. 1. The architectural view of the study.IV.EXPERIMENTAL STUDIESIn order to compare the cities in the project and find similarones, the articles to introduce these cities have been found.The different sources have been searched, but each source hadto emphasize a different feature of cities. For example, whilethere is an introductory article concentrating on a rich Parisiancuisine, the other one is the fashion, culture and artisticfeatures of the Parisian city. This could have caused thedifferent results and illusions when comparing cities.Therefore, a resource had to offer all the features of the citiesunder certain headings and explain the different features.Furthermore, it had to contain all the city and tourist places inour database, it is possible to use “wikipedia.com” tourismrelated versions of “wikitravel.com” and “wikivoyage.com”.Finally, the related words have been collected and saved as".txt" files.For the selection of recorded cities, “telegraph.co.uk” hasbeen selected because this site shows the best 50 cities in theworld. citiesin-pictures/). Tourist attractions in Turkey have been alsoadded to the storage. The designated tourist destinations are asshown in Table I and Table II.Id1234567TABLE I. Cities in Turkey taken part in the system.CityIdCityIdCityIdCityAdana 8 Diyarbakır 15Hatay22KonyaAnkara 9Edirne16Istanbul23MarmarisAntalya 10 Erzurum17Kapadokya24SamsunBodrum 11 Eskisehir18Kas25TokatBursa 12Fethiye19Kastamonu26TrabzonCesme 13 Gaziantep 20Kayseri27UrfaDenizli 14 Hasankeyf 21Kemer28VanAfter the data have been collected, the data preparationphase, which is one of the most important stages of the datamining, has been implemented. The words in English textshave been translated into plain words. At this stage in turn: Elimination of non-English words. Example: Sultan,Ahmet, Cami’s, they must be removed. Elimination of stop words. Example: The, an, a, at, is,etc. Elimination of comparative. Example: “bigger”changed “big” as base form.ISSN (Online): 2456-7361Elimination of superlative. Example: “biggest” changed“big” as base form. Elimination of suffixes. Example: “worker” changed“work” as base form. Transformation from past tense, past participle andpresent participle to infinitive forms. Example:“worked”,” working”,” works” changed “work” as baseform.The first process has been that deleting the punctuationmarks in the texts. After that, the obtained words have beencompared to the words in the English dictionary ist/wordsEn.txt). Id123456789101112131415161718TABLE IIIII. Cities except Turkey taken part in the system.CityIdCityIdCityIdCityAbu Dhabi 19Chiang Mai37Lisbon55 Rio de JenarioAlaska20Cologne38 London 56 San FranciscoAmsterdam 21 Copenhagen 39 Luxembourg 57SantiagoAntarctica ngkok24Dresden42 Marrakech 60 ShanghaiBarcelona 25Dubai43Milano61 St PetersburgBeijing26Dublin44Minsk62 StockholmBelgrade27 East Bourne 45 Monaco 63 StrasbourgBerlin28Eindhoven46 Moscow 64SydneyBirmingham 29Florence47 Munich 1Helsinki49Napoli67 VancouverBoston32Jakarta50 New Delhi 68VeniceBudapest33Kabul51 New York 69ViennaCanberra34Krakow52Oslo70WarsawCancun35 Kuala Lumpur 53Paris71YerevanCape Town 36Kyiv54Prague72ZagrebThe words, which do not exist in the dictionary, have beendeleted from the data set. Furthermore, the stop-words inEnglish, which mean that they do not contain any meaning ifthey do not take part in the sentences, have been deleted. The“a” part in Fig. 2 shows these operations.In this phase, the text file we obtained has contained onlyEnglish words which are not stop-words. Suffixes of adverbs,adjectives, superlatives, comparatives, past simple and pastparticiple have been converted to the base forms of the words.These operations have been done as given in the “b” part inFig. 2 and finally, the last version of the text file has beenprepared for data mining operations.After cleaning the data set, the files with the extension".csv" have been needed for text mining. All city names had tobe represented as rows and the cleaned words had to berepresented as columns to storage specific frequencies in thedata set. The next operation has been to discover the breakpoint for the frequency in the data set, because the aim of datamining operations had to take the words which have countshigher than a certain frequency. The break point has beenassumed as 2 for frequencies. Thus, the words as features,which have more frequency than 2, would be able to be usedto compare 2 cities.10http://ijses.com/All rights reserved
International Journal of Scientific Engineering and ScienceVolume 1, Issue 1, pp. xx-xx, 2017.ISSN (Online): 2456-7361a)b)Fig. 2. a) The parsing and elimination of non-English words. b) The elimination of suffixes.In the next phase, the operations for the creation of “.csv”files, which have the words with different break points for thefrequency, have been performed.TABLE IVVVI. The standard deviation values by frequency.Freq.K 15K 10K 7K 5K 1514.18It has been to record the words higher frequency than acertain one after obtaining the general frequencies. Forexample, the files with frequencies of 60, 55, 50, 45, , 10,and 5 have been created for words separately. Finally, thewords with a certain frequency have been saved as separatefiles with ".csv" extensions.A common file consisting frequency information for eachcity has been created as given in Fig. 3. In this sample showsthe frequencies of each word for each city as a matrix.Respectively, the standard deviations of each file for thefrequency numbers of 2, 5, 10, 15, 20, 25, 30, 35, 40, 45, 50,55, and 60 after obtaining files using the Weka tool have beencalculated. Table III shows that the standard deviation value isthe minimum when the frequency is 45. Also, this resultmeans the maximum consistency. After this step,“frequency45.csv” file have been obtained. The data set hasbeen divided into certain k-valued clusters by K-Meansmethod using Weka tool. When the standard deviation values11http://ijses.com/All rights reserved
International Journal of Scientific Engineering and ScienceVolume 1, Issue 12, pp. 8-14, 2017.ISSN (Online): 2456-7361of this file have been compared with k number of 3, 5, 7, 10and15, it has been seen that the most consistent value has beenobtained for 15. This observation has pointed out that the idealdistribution would be obtained when the cities had to bedivided into 15 clusters.After the “frequency45.csv” file has been divided into 15clusters using the K-Means algorithm, the number of elementsfor each cluster is as shown in the first column in Table V.The centres of these clusters have been obtained with theWeka tool as given in Fig. 4. The Self Organizing Map (SOM)algorithm has been used to obtain a better distribution of theclusters, because the 5th cluster had only one city.SOM algorithm has been evaluated to discover the closestneighbour to merge the cluster with its closest neighbour. Inthis study, a recommendation mechanism has been the aim;therefore, all clusters had to have cities at least two. AfterSOM, the coordinates in Table IV have been obtained.IdXYTABLE VIIV. Coordinates of clusters according to SOM algorithm.0 1 2 3 4 5 6 7 8 9 10 11 12 13 140 0 2 3 0 0 3 2 0 0 310100 0 2 1 2 0 3 1 0 0 01033thThe 5 cluster with one element had to be included intoone of the other cluster in the same coordinate. When thedistances between the centres of the clusters have beenmeasured according to the Euclidean distance, it has been seenthat the closest cluster centre has been the centre of cluster 0.For this reason, the single element in the cluster 5 hastransferred into the cluster 0.The number of elements in the clusters after K-Means andSOM operations and their differences are shown in Table V.Before applying the SOM algorithm, the standard deviationvalue had been 3.75, which have been 2.70 after the SOMalgorithm has been applied. This shows that after theimplementation of the SOM algorithm, the number ofelements in the clusters has more consistent and betterdistribution.Fig. 3. A sample of “.csv” file.Finally, the calculation of the standard deviations of wordshas been done to find out which words in each cluster had tobe used. According to the result of the SOM algorithm, thefrequencies of the 0th and 5th clusters have been needed tocombine by a written code to calculate the weighted averageof the centroids. As a result, a decreasing has been observedfor the number of clusters from 15 to 14. In the next sections,the precision, recall and f-measure values are evaluated for thenumbers of words to obtain the optimum break pointaccording to the classification algorithms.Fig. 4. The centroids before SOM implementation.12http://ijses.com/All rights reserved
International Journal of Scientific Engineering and ScienceVolume 1, Issue 12, pp. 8-14, 2017.TABLE V. The number of elements in the clusters after K-Means and SOMCluster NameAfter K-Means,After SOM,the number ofthe number of elementselementsCluster 045Cluster 122Cluster 288Cluster 399Cluster 488Cluster 51Cluster 61010Cluster 777Cluster 822Cluster 955Cluster 1066Cluster 1188Cluster 1266Cluster 131616Cluster 1488Total100100Std. Deviation3.752.86V.ANALYSIS RESULTSAfter the process of the clustering, the pattern has beentested using classification algorithms. The cluster value foreach city has been used as the target attribute in the data setand new data set files have been created for this analysis. Toextend the confidence bounds, it has been assumed that as theobtained cluster values had to be stable, the words as theattributes in the data set had to be changed. Therefore, thebreak points according to the frequencies have changed and 8“.csv” files, which have contained 3, 5, 7, 10, 15, 20, 25 and30 words and have had a harmonious structure with a targetattribute for the classification algorithms, have been created.Subsequently, all files have been classified with differentclassification techniques. The used classification algorithmshave been as follows; Support Vector Machines (SMO),Multilayer Perceptron, Naïve Bayes, K-Nearest Neighbour(IBk), Decision Tree (J48)These algorithms have been compared using f-measure in(1), precision in (2) and recall in (3) terms. In Weka tool, theseterms are named as follows; Weighted Average Precision,Weighted Average Recall and Weighted Average F-measure.In Table VI, the precision, the recall and the f-measurevalues are listed for SMO algorithm. In Table VII, the valuesfor the same terms are listed for Multilayer Perceptron. InTable VIII, the values are listed for Naïve Bayes.In Table IX, the values are given for SMO algorithm andthe values are listed for J48 algorithm in Table X. All valueshave been obtained separately for each file. For example, thefirst file has 100 instances and 4 attributes with 3 words as thefeatures and a target attribute.ISSN (Online): 2456-7361TABLE VI. Precision, Recall, F-measure values for SMO algorithm.NumberWeighted Avg.Weighted Avg.Weighted Avg.of WordsPrecisionRecallF-Measure3 words0.0670.2000.0945 words0.2010.3000.2097 words0.3740.3800.31710 words0.5130.4400.39315 words0.5750.5100.49420 words0.6220.6200.59525 words0.6550.6500.63030 words0.6540.6600.637TABLE VII. Precision, Recall, F-measure values for Multilayer Perceptron.NumberWeighted Avg.Weighted Avg.Weighted Avg.of WordsPrecisionRecallF-Measure3 words0.0670.2000.0945 words0.2530.2900.2687 words0.2710.2900.27810 words0.4440.4300.43215 words0.5630.5400.54220 words0.6540.6500.64225 words0.7120.6900.68930 words0.6430.6400.628TABLE VIII. Precision, Recall, F-measure values for Naïve Bayes.NumberWeighted Avg.Weighted Avg.Weighted Avg.of WordsPrecisionRecallF-Measure3 words0.2060.2600.2265 words0.3150.3300.3157 words0.3650.2900.31110 words0.4100.3900.38015 words0.4110.4000.38220 words0.4010.4300.40525 words0.4070.4400.41530 words0.4120.4400.414TABLE IX. Precision, Recall, F-measure values for IBk algorithm.NumberWeighted Avg.Weighted Avg.Weighted Avg.of WordsPrecisionRecallF-Measure3 words0.1360.2000.1615 words0.1360.2000.1617 words0.3200.2800.28110 words0.4160.3700.36615 words0.3890.4200.38020 words0.5450.5300.50425 words0.4680.4800.45030 words0.4670.4800.452TABLE X. Precision, Recall, F-measure values for J48 algorithm.NumberWeighted Avg.Weighted Avg.Weighted Avg.of WordsPrecisionRecallF-Measure3 words0.1420.1900.1585 words0.1870.2200.1997 words0.1970.2200.20610 words0.2740.2800.27615 words0.2550.2900.26820 words0.4000.4000.39225 words0.3810.3700.36730 words0.3680.3500.349According to these results, it has been observed that the fmeasure values of the Multilayer Perceptron technique hasbeen obtained more successfully than the other techniques inthe previous 5 tables. Also, it has been said that the file with25 words has had the optimum number of the attributes.13http://ijses.com/All rights reserved
International Journal of Scientific Engineering and ScienceVolume 1, Issue 12, pp. 8-14, 2017.VI.MOBILE APPLICATIONThe mobile application has been implemented on Android5.0. This application contains 4 main operations; finding themost similar cities to a city, finding the most similar Turkishcities to a city, finding the cities that contain the selected wordin the introduction, and finding the Turkish cities that containthe selected word in the introduction.Fig. 5 shows two screenshots of the application. The firstscreenshot contains the menu of the operations and the secondone presents the cities list to discover alternative cities.ISSN (Online): 2456-7361of this study is to show that implementing text miningalgorithms on web texts about the tourism sector can bring outthe efficient applications containing mixing of an academicbackground and commercial software tools.REFERENCES[1][2][3][4][5][6][7][8]Fig. 5. The screenshots of the mobile applicationFig. 5 shows two screenshots of the application. The firstscreenshot contains the menu of the operations and the secondone presents the cities list to discover alternative cities. Thismobile application has been implemented to test the results oftext mining and obtain the visual information.VII. CONCLUSIONThe study has been developed to find similarities among acertain number of cities using various text mining methodsand present them to the user. In the study, 100 cities have beenprocessed and this number should be increased in the nextstudies. The city information containing 16820 words has beencollected from the different sources and they have been savedas files. NLP operations have been implemented on these filesand clean data sets have been obtained for text miningmethods. This study has been applied for a total of 100 cities,27 of which have been Turkish cities and 73 have been thecities in the other countries. The cities have been divided intospecific clusters using clustering algorithms. The SOMalgorithm has been used to make the distribution better and togain the balance of the distribution. Finally, 100 cities havebeen located in 14 clusters. Among the classification methods,Multilayer Perceptron had more consistent results withaveragely 70% for the precision, the recall and the f-measurevalues than other methods. It shows that a new city with thefrequency values for the certain 25 words can be located into acluster which is discovered by Multilayer Perceptron.In this study, the obtained results have been saved inMSSQL database and presented to the user in mobileenvironment using WCF web service. The main consequence[9][10][11][12][13][14][15]F. Lorenzi, R. Saldana, S. Loh and D. Litchnow, “A TourismRecommender System Based on Collaboration and Text Analysis”,Information Technology & Tourism, vol. 6(3), pp. 157-165, 2003.K. Lau, K. Lee and Y. Ho, “Text Mining for the Hotel Industry”,Cornell Hotel and Restaurant Administration Quarterly, vol. 46(3), pp.344-362, 2005.K. Berezina, F. Okumus, A. Bilgihan and C. Cobanoglu, “UnderstandingSatisfied and Dissatisfied Hotel Customers: Text Mining of Online HotelReviews”, Journal of Hospitality Marketing & Management, vol. 25(1),pp. 1-24, 2016.R. Segall, Q. Zhang, and M. Cao, "Web-Based Text Mining of HotelCustomer Comments Using SAS Text Miner and MegaputerPolyanalyst", SWDSI 2009 Proceedings, pp. 141-152, 2009.W. Claster, W., M. Cooper, M., K. Tajeddini and P. Pardo, “Tourism,travel and tweets: algorithmic text analysis methodologies in tourism”,Middle East J. Management, vol. 1(1), pp. 81-99, 2013W. J. Amadio and J. D. Procaccino, "Competitive analysis of onlinereviews using exploratory text mining", Tourism and HospitalityManagement, vol. 22(2), pp. 193-210, 2016.Y. Yifan, D. Junping, F. Dan and J. Lee, "Design and implementation oftourism activity recognition and discovery system", in Proc. IntelligentControl and Automation (WCICA), Guilin, China, pp. 781-786, 2016H. Ban, H. Kimura and T. Oyabu, "Feature extraction of Englishguidebooks for Hokuriku region in Japan", Journal of Global TourismResearch, vol. 1(1), pp. 71-76, 2016.A. Dickinger, D. Astrid, L. Lidija, M. Josef and M. Josef, "Exploring thegeneralizability of discriminant word items and latent topics in onlinetourist reviews", International Journal of Contemporary HospitalityManagement, vol. 29(2), pp. 803-816, 2017.T. Guo, B. Guo, Y. Ouyang, Z. Yu, J. C. Lam and V. O. Li,"CrowdTravel: scenic spot profiling by using heterogeneouscrowdsourced data", Journal of Ambient Intelligence and HumanizedComputing, pp. 1-10, 2017.U. Godnov and T. Redek, "Application of text mining in tourism: Caseof Croatia", Annals of Tourism Research, vol. 58, pp. 162-166. 2016.I. Pembeci, "Using Word Embeddings fo
1, 2Department of Computer Engineering, Dokuz Eylül University, Izmir, Turkey Abstract — The purpose of this study is to show that it is possible to benefit from the
Text text text Text text text Text text text Text text text Text text text Text text text Text text text Text text text Text text text Text text text Text text text
Correlation of Discovering Algebra, 2nd Edition, Discovering Geometry, 4th Edition, and Discovering Advanced Algebra, . importance of checking whether an answer to a real-world . students gain more experience Discovering Geometry Discovering Advanced Algebra . Correlation of Discovering Mathematics Key Curriculum Press June 2010 .
Discovering Advanced Algebra also completes the fully updated Discovering Mathematics series. All of the features that make Discovering Algebra and Discovering Geometry innovative and exciting are now incorporated into this book as well, to make a coherent and streamlined series. Investigations are at the heart of each book.
Enabling the VCS Control for Mobile and Remote Access 22 Discovering Unified Communications Servers and Services 24 Trusting the Certificates Presented to the VCS Control 24 Discovering IM and Presence Service Nodes 24 Discovering Unified CM Servers 25 Discovering Cisco Unity Connection Servers 26 Automatically Generated Zones and Search Rules 26
Enabling the Expressway-C for mobile and remote access 20 Discovering Unified Communications servers and services 22 Trusting the certificates presented to the Expressway-C 22 Discovering IM and Presence Service nodes 22 Discovering Unified CM servers 23 Discovering Cisco Unity Connection servers 24 Automatically generated zones and search rules 24
A. The Smarter Cities Challenge In 2010, IBM Corporate Citizenship launched the Smarter Cities Challenge to help 100 cities around the world over a three-year period become smarter through grants of IBM talent. Boston, Massachusetts, was selected through a competitive process as one of 33 cities to be awarded a Smarter Cities Challenge grant in .
Cities are moving from pledges to action, with 27 of the world's largest cities in the C40 Cities Initiative having achieved at least 10 percent lower emissions than their peak as they work towards decarbonization by 2050. Through this initiative, cities are starting to develop Paris Agreement Compatible Climate Action Plans. Seven C40 cities .
Discovering Human Sexuality FIFTH EDITION Welcome to Discovering Human Sexuality, 5E! To help you get off to a smooth start this term, this Student Quick Start Guide will cover essential information about using and accessing your text's digital resources within your school's learning management system (LMS) via Oxford Learning Link Direct.
