• Korean Journal of Ecology and Environment
• /
• v.55 no.1
• /
• pp.60-75
• /
• 2022

The Bayesian algorithm model is a model algorithm that calculates probabilities based on input data and is mainly used for complex disasters, water quality management, the ecological structure between living things or living-non-living factors. In this study, we analyzed the main factors affected Korean Estuary Trophic Diatom Index (KETDI) change based on the Bayesian network analysis using the diatom community and physicochemical factors in the domestic estuarine aquatic ecosystem. For Bayesian analysis, estuarine diatom habitat data and estuarine aquatic diatom health (2008~2019) data were used. Data were classified into habitat, physical, chemical, and biological factors. Each data was input to the Bayesian network model (GeNIE model) and performed estuary aquatic network analysis along with the nationwide and each coast. From 2008 to 2019, a total of 625 taxa of diatoms were identified, consisting of 2 orders, 5 suborders, 18 families, 141 genera, 595 species, 29 varieties, and 1 species. Nitzschia inconspicua had the highest cumulative cell density, followed by Nitzschia palea, Pseudostaurosira elliptica and Achnanthidium minutissimum. As a result of analyzing the ecological network of diatom health assessment in the estuary ecosystem using the Bayesian network model, the biological factor was the most sensitive factor influencing the health assessment score was. In contrast, the habitat and physicochemical factors had relatively low sensitivity. The most sensitive taxa of diatoms to the assessment of estuarine aquatic health were Nitzschia inconspicua, N. fonticola, Achnanthes convergens, and Pseudostaurosira elliptica. In addition, the ratio of industrial area and cattle shed near the habitat was sensitively linked to the health assessment. The major taxa sensitive to diatom health evaluation differed according to coast. Bayesian network analysis was useful to identify major variables including diatom taxa affecting aquatic health even in complex ecological structures such as estuary ecosystems. In addition, it is possible to identify the restoration target accurately when restoring the consequently damaged estuary aquatic ecosystem.

• Journal of Nutrition and Health
• /
• v.10 no.4
• /
• pp.43-58
• /
• 1977

Kojedo is the second largest island in Korea and a total population of 115,500 is living on the island of 394.69 sq. km. Under the direction of three nutrition professors, nutrition surveys in two villages in Kojedo, namely Siljun Ri in Hachung Myon and Soowol Ri in Shinhyun Myon, were carried by 30 college senior students majoring in nutrition from August to 20 August 1977. From a total of 176 households of the two villages, 67 households were randomly selected and 390 family members of the households were subjcets of the nutrition surveys. The precise weighing method was used in evaluating the kinds of foods and nutrient intakes of the subjects for three consecutive days. Thirty-seven pre-school children aged between 3 to 6 years and 27 fertile women were examined for biochemical findings and physical status. The main purposes of the surveys are to provide baseline data on nutrition in Kojedo Island for the Kojedo Community Development Project and to compare the nutritional status of the villages of Siljun Ri and Soowol Ri. Siljun Ri is located in the pilot project area of the Koiedo Community Health Project sponsored since December 1970 by the Christian Medical Commission of the World Council of Churches. While Soowol Ri is a control village for comparison. The results obtained are summarized as follows: Food Intake The average food intake per person per day in Siljun Ri, 1064 grams (91.7% in vegetable foods and 7.6% in animal foods) was 90 grams more than that of Soowol Ri, 974 grams (92.8% in vegetable foods and 5.9% in animal foods). However, the food intake per pre-school child in Siljun Ri, 485 grams (92.6% from vegetable foods and 6.4% from animal foods) was 21 grams lower than that of the Soowol Ri, 506 grams (88.5% from vegetable foods and 6.5% from animal foods). The average intake of beans was 16 grams(1.5% out of the total food intake) in Siljun Ri and 21 grams(2.2% of the total food intake)in Soowol Ri. The villagers should be guided for more consumption of soybeans to improve the quality of protein intake from vegetable foods. Nutrient Intake The adult intake in Siljun Ri and Soowol Ri were 2,529 kcal and 2,511 kcal respectively. The average energy intake of pre-school childen in Siljun Ri was 948 kcal and that for adult and 1,500 kcal for childen aged between 4 to 6 years-given by the Korea FAO Association, the diets in both villages were not adequate. Average daily protein intake of the subjected adult in Siljun Ri was 78.4 grams and that of Soowol Ri was 76.2 grams, while pre-school children took 30.7 grams in the former village and 31.7 grams in the latter village per child per day. The protein intake in both villages were lower than the recommended allowances, 80 grams for adult and 45 grams for $4{\sim}6$ years childen, and animal protein intake of the all subjects was very much lower than the RDA. The main charecter of the diet has been found low in quality of protein and high in carbohydrate. The calcium intakes of the pre-school children in both villages, 251.9 milligrams in Siljun Ri and 218.8 milligrams in Soowol Ri, were very much lower than the recommended allowance of 500 milligrams per day. It is apparent that the diet for children should be supplemented with calcium. Among the vitamin group, the daily average intakes of vitamin A and $B_{2}$(thiamine), $B_{2}$(riboflavin), C(ascorbic acid), and niacin were not adequate for the children in both villages. Especially the intake of riboflavin, 0.4 milligrams in both village children, was much lower than the RDA, 0.9 milligrams per day. Physical Characteristics Average height, weight, chest and head circumference of the pre-school children in both villages were similar to those of the Korean standard given by the Korean Paediatrics Association except that the average height of pre-school boys in Siljun Ri was 8 cm higher than the Korean standard of 105 cm. The mean values of upper arm circumference and skinfold thickness of pre-school boys in both villages were the same, 15.4 cm for upper arm circumference and 6.8 mm for skinfold thickness, but the mean values of those of the girls in Siljun Ri were higher than those of pre-school grils in Soowol Ri. Biochemical Findings Avera ge hemogobin value of boys and girls in both villages was the same, 11.1 grams per 100 ml of blood. The incidence of anemia (Hb value below 11g/100ml) was similar in both viltagesr 36.4% for boys and 50% for girls in Siljun Ri and 37.5% for boys and 50% for girls in Soowol Ri. Average hemoglobin values of fertile women were 10.7g% in Siljun Ri and 10.8% in Soowor Ri. The incidences of anemia(Hb valre brlow 12g/100ml) were 100% in Siljun Ri and 86.7% in Soowol Ri. The anemia of these subjects may be caused mainty low intake of good quality protein and iron intake from vegetable food. Recommendation In general, the nutritional status of a community health pilot village is not higher than that of control village due to the lack of nutrition improvement guldance services. Nutrition education should be delivered to the villagers as a main part of the health education artivities. The emphasis should be on building better health through bttter food habits and better food production as well as on preventing malnutrition and diseasrs. It can be an invaluable part of community developnent. Since nutrition is considered to be at least one-half of MCH care, no village or home visits should be made without careful provision for teaching and demoastrating something simple and practical on nutrition. The nurse, midwife, and village health worker should be the chief promoters of nutrition.

• The Korean Journal of Nuclear Medicine Technology
• /
• v.14 no.1
• /
• pp.83-89
• /
• 2010

Purpose: The Wide Beam Reconstruction (WBR) algorithms that UltraSPECT, Ltd. (U.S) has provides solutions which improved image resolution by eliminating the effect of the line spread function by collimator and suppression of the noise. It controls the resolution and noise level automatically and yields unsurpassed image quality. The aim of this study is WBR of whole body bone scan in usefulness of clinical application. Materials and Methods: The standard line source and single photon emission computed tomography (SPECT) reconstructed spatial resolution measurements were performed on an INFINA (GE, Milwaukee, WI) gamma camera, equipped with low energy high resolution (LEHR) collimators. The total counts of line source measurements with 200 kcps and 300 kcps. The SPECT phantoms analyzed spatial resolution by the changing matrix size. Also a clinical evaluation study was performed with forty three patients, referred for bone scans. First group altered scan speed with 20 and 30 cm/min and dosage of 740 MBq (20 mCi) of $^{99m}Tc$-HDP administered but second group altered dosage of $^{99m}Tc$-HDP with 740 and 1,110 MBq (20 mCi and 30 mCi) in same scan speed. The acquired data was reconstructed using the typical clinical protocol in use and the WBR protocol. The patient's information was removed and a blind reading was done on each reconstruction method. For each reading, a questionnaire was completed in which the reader was asked to evaluate, on a scale of 1-5 point. Results: The result of planar WBR data improved resolution more than 10%. The Full-Width at Half-Maximum (FWHM) of WBR data improved about 16% (Standard: 8.45, WBR: 7.09). SPECT WBR data improved resolution more than about 50% and evaluate FWHM of WBR data (Standard: 3.52, WBR: 1.65). A clinical evaluation study, there was no statistically significant difference between the two method, which includes improvement of the bone to soft tissue ratio and the image resolution (first group p=0.07, second group p=0.458). Conclusion: The WBR method allows to shorten the acquisition time of bone scans while simultaneously providing improved image quality and to reduce the dosage of radiopharmaceuticals reducing radiation dose. Therefore, the WBR method can be applied to a wide range of clinical applications to provide clinical values as well as image quality.

• Radiation Oncology Journal
• /
• v.25 no.4
• /
• pp.242-248
• /
• 2007

Purpose: To study the effect of recombinant human epidermal growth factor (rhEGF) on oral mucositis induced by cisplatin and radiotherapy in a mouse model. Materials and Methods: Twenty-four ICR mice were divided into three groups-the normal control group, the no rhEGF group (treatment with cisplatin and radiation) and the rhEGF group (treatment with cisplatin, radiation and rhEGF). A model of mucositis induced by cisplatin and radiotherapy was established by injecting mice with cisplatin (10 mg/kg) on day 1 and with radiation exposure (5 Gy/day) to the head and neck on days $1{\sim}5$. rhEGF was administered subcutaneously on days -1 to 0 (1 mg/kg/day) and on days 3 to 5 (1 mg/kg/day). Evaluation included body weight, oral intake, and histology. Results: For the comparison of the change of body weight between the rhEGF group and the no rhEGF group, a statistically significant difference was observed in the rhEGF group for the 5 days after day 3 of. the experiment. The rhEGF group and no rhEGF group had reduced food intake until day 5 of the experiment, and then the mice demonstrated increased food intake after day 13 of the of experiment. When the histological examination was conducted on day 7 after treatment with cisplatin and radiation, the rhEGF group showed a focal cellular reaction in the epidermal layer of the mucosa, while the no rhEGF group did not show inflammation of the oral mucosa. Conclusion: These findings suggest that rhEGF has a potential to reduce the oral mucositis burden in mice after treatment with cisplatin and radiation. The optimal dose, number and timing of the administration of rhEGF require further investigation.

• Food Science of Animal Resources
• /
• v.30 no.6
• /
• pp.975-988
• /
• 2010

Two experiments were conducted to determine whether water extracts from Artemisia capillaries (A. capillaries) and Camellia sinensis (C. sinensis) could be used as alternatives to antibiotic growth promoters in broiler feed. The experiment 1 was verified their chemical composition, extracts yields, total phenolic compounds concentration, antioxidant activity, antimicrobial activity, and chicken splenocytes proliferation through in vitro test. The extract yields of A. capillaries and C. sinensis were 26.5 and 16.8%, respectively. Total phenolic compounds concentrations of them expressed as gallic acid equivalent were 15.28 and 26.74 mg/mL, respectively. Electron donating abilities of them expressed as $SC_{50}$ showing 50% DPPH radical scavenging were 0.30 and 0.06 mg, respectively. Bacterial inhibitory rates of them against Escherichia coli, Staphylococcus aureus, and Salmonella Typhimurium were ranged from 42.1 to 52.3% and from 21.6 to 33.7%, respectively. And, these extracts increased proliferation of chicken splenocytes. Especially, A. capillaris was more excellent than Echinacea and Concanavalin A known as T-cell stimulator. The experiment 2 was investigated their effects on growth performance, relative organ weight, cecal microflora, blood biochemical parameters, and splenic cytokines mRNA expression in broiler chicks. Four hundred eighty 1-day-old male broiler chicks (Ross 308) were divided in to 4 treatment groups with 4 replicates of 30 birds in each group: NC (control, no antibiotics), PC (avilamycin, 10 ppm; salinomycin, 60 ppm), AC (A. capillaries, 100 ppm), and CS (C. sinensis, 100 ppm); treatments were administered through water supplementation. Final body weight was significantly higher in all treated groups than in NC (p<0.05). Cecal Salmonella numbers were significantly or somewhat decreased in all treated groups than in NC (p<0.05). The relative weights and lengths of the small intestine were more significantly decreased in the PC and AC groups than in the other groups. Cecal Salmonella numbers were significantly or somewhat decreased in all treated groups than in the NC group (p<0.05). The contents of total cholesterol, aspatate aminotransferase, and alanine aminotransferase in blood serum were more significantly decreased in all treated groups than in NC (p<0.05). In conclusion, these results suggested the possibility that these extracts could serve as alternatives for antibiotic growth promoters.

• Journal of Intelligence and Information Systems
• /
• v.23 no.2
• /
• pp.71-88
• /
• 2017

Language models were originally developed for speech recognition and language processing. Using a set of example sentences, a language model predicts the next word or character based on sequential input data. N-gram models have been widely used but this model cannot model the correlation between the input units efficiently since it is a probabilistic model which are based on the frequency of each unit in the training set. Recently, as the deep learning algorithm has been developed, a recurrent neural network (RNN) model and a long short-term memory (LSTM) model have been widely used for the neural language model (Ahn, 2016; Kim et al., 2016; Lee et al., 2016). These models can reflect dependency between the objects that are entered sequentially into the model (Gers and Schmidhuber, 2001; Mikolov et al., 2010; Sundermeyer et al., 2012). In order to learning the neural language model, texts need to be decomposed into words or morphemes. Since, however, a training set of sentences includes a huge number of words or morphemes in general, the size of dictionary is very large and so it increases model complexity. In addition, word-level or morpheme-level models are able to generate vocabularies only which are contained in the training set. Furthermore, with highly morphological languages such as Turkish, Hungarian, Russian, Finnish or Korean, morpheme analyzers have more chance to cause errors in decomposition process (Lankinen et al., 2016). Therefore, this paper proposes a phoneme-level language model for Korean language based on LSTM models. A phoneme such as a vowel or a consonant is the smallest unit that comprises Korean texts. We construct the language model using three or four LSTM layers. Each model was trained using Stochastic Gradient Algorithm and more advanced optimization algorithms such as Adagrad, RMSprop, Adadelta, Adam, Adamax, and Nadam. Simulation study was done with Old Testament texts using a deep learning package Keras based the Theano. After pre-processing the texts, the dataset included 74 of unique characters including vowels, consonants, and punctuation marks. Then we constructed an input vector with 20 consecutive characters and an output with a following 21st character. Finally, total 1,023,411 sets of input-output vectors were included in the dataset and we divided them into training, validation, testsets with proportion 70:15:15. All the simulation were conducted on a system equipped with an Intel Xeon CPU (16 cores) and a NVIDIA GeForce GTX 1080 GPU. We compared the loss function evaluated for the validation set, the perplexity evaluated for the test set, and the time to be taken for training each model. As a result, all the optimization algorithms but the stochastic gradient algorithm showed similar validation loss and perplexity, which are clearly superior to those of the stochastic gradient algorithm. The stochastic gradient algorithm took the longest time to be trained for both 3- and 4-LSTM models. On average, the 4-LSTM layer model took 69% longer training time than the 3-LSTM layer model. However, the validation loss and perplexity were not improved significantly or became even worse for specific conditions. On the other hand, when comparing the automatically generated sentences, the 4-LSTM layer model tended to generate the sentences which are closer to the natural language than the 3-LSTM model. Although there were slight differences in the completeness of the generated sentences between the models, the sentence generation performance was quite satisfactory in any simulation conditions: they generated only legitimate Korean letters and the use of postposition and the conjugation of verbs were almost perfect in the sense of grammar. The results of this study are expected to be widely used for the processing of Korean language in the field of language processing and speech recognition, which are the basis of artificial intelligence systems.

• Asia pacific journal of information systems
• /
• v.24 no.2
• /
• pp.211-232
• /
• 2014

Recently, there is a growing interest toward social commerce using SNS(Social Networking Service), and the size of its market is also expanding due to popularization of smart phones, tablet PCs and other smart devices. Accordingly, various studies have been attempted but it is shown that most of the previous studies have been conducted from perspectives of the users. The purpose of this study is to derive user-centered CSF(Critical Success Factor) of social commerce from the previous studies and analyze the CSF perception gap between social commerce service providers and users. The CSF perception gap between two groups shows that there is a difference between ideal images the service providers hope for and the actual image the service users have on social commerce companies. This study provides effective improvement directions for social commerce companies by presenting current business problems and its solution plans. For this, This study selected Korea's representative social commerce business Ticket Monster, which is dominant in sales and staff size together with its excellent funding power through M&A by stock exchange with the US social commerce business Living Social with Amazon.com as a shareholder in August, 2011, as a target group of social commerce service provider. we have gathered questionnaires from both service providers and the users from October 22, 2012 until October 31, 2012 to conduct an empirical analysis. We surveyed 160 service providers of Ticket Monster We also surveyed 160 social commerce users who have experienced in using Ticket Monster service. Out of 320 surveys, 20 questionaries which were unfit or undependable were discarded. Consequently the remaining 300(service provider 150, user 150)were used for this empirical study. The statistics were analyzed using SPSS 12.0. Implications of the empirical analysis result of this study are as follows: First of all, There are order differences in the importance of social commerce CSF between two groups. While service providers regard Price Economic as the most important CSF influencing purchasing intention, the users regard 'Trust' as the most important CSF influencing purchasing intention. This means that the service providers have to utilize the unique strong point of social commerce which make the customers be trusted rathe than just focusing on selling product at a discounted price. It means that service Providers need to enhance effective communication skills by using SNS and play a vital role as a trusted adviser who provides curation services and explains the value of products through information filtering. Also, they need to pay attention to preventing consumer damages from deceptive and false advertising. service providers have to create the detailed reward system in case of a consumer damages caused by above problems. It can make strong ties with customers. Second, both service providers and users tend to consider that social commerce CSF influencing purchasing intention are Price Economic, Utility, Trust, and Word of Mouth Effect. Accordingly, it can be learned that users are expecting the benefit from the aspect of prices and economy when using social commerce, and service providers should be able to suggest the individualized discount benefit through diverse methods using social network service. Looking into it from the aspect of usefulness, service providers are required to get users to be cognizant of time-saving, efficiency, and convenience when they are using social commerce. Therefore, it is necessary to increase the usefulness of social commerce through the introduction of a new management strategy, such as intensification of search engine of the Website, facilitation in payment through shopping basket, and package distribution. Trust, as mentioned before, is the most important variable in consumers' mind, so it should definitely be managed for sustainable management. If the trust in social commerce should fall due to consumers' damage case due to false and puffery advertising forgeries, it could have a negative influence on the image of the social commerce industry in general. Instead of advertising with famous celebrities and using a bombastic amount of money on marketing expenses, the social commerce industry should be able to use the word of mouth effect between users by making use of the social network service, the major marketing method of initial social commerce. The word of mouth effect occurring from consumers' spontaneous self-marketer's duty performance can bring not only reduction effect in advertising cost to a service provider but it can also prepare the basis of discounted price suggestion to consumers; in this context, the word of mouth effect should be managed as the CSF of social commerce. Third, Trade safety was not derived as one of the CSF. Recently, with e-commerce like social commerce and Internet shopping increasing in a variety of methods, the importance of trade safety on the Internet also increases, but in this study result, trade safety wasn't evaluated as CSF of social commerce by both groups. This study judges that it's because both service provider groups and user group are perceiving that there is a reliable PG(Payment Gateway) which acts for e-payment of Internet transaction. Accordingly, it is understood that both two groups feel that social commerce can have a corporate identity by website and differentiation in products and services in sales, but don't feel a big difference by business in case of e-payment system. In other words, trade safety should be perceived as natural, basic universal service. Fourth, it's necessary that service providers should intensify the communication with users by making use of social network service which is the major marketing method of social commerce and should be able to use the word of mouth effect between users. The word of mouth effect occurring from consumers' spontaneous self- marketer's duty performance can bring not only reduction effect in advertising cost to a service provider but it can also prepare the basis of discounted price suggestion to consumers. in this context, it is judged that the word of mouth effect should be managed as CSF of social commerce. In this paper, the characteristics of social commerce are limited as five independent variables, however, if an additional study is proceeded with more various independent variables, more in-depth study results will be derived. In addition, this research targets social commerce service providers and the users, however, in the consideration of the fact that social commerce is a two-sided market, drawing CSF through an analysis of perception gap between social commerce service providers and its advertisement clients would be worth to be dealt with in a follow-up study.

• The Korean Journal of Nuclear Medicine Technology
• /
• v.19 no.2
• /
• pp.87-92
• /
• 2015

Purpose When the patients takes myocardial perfusion SPECT using $^{201}Tl$, the operator gives the patients an injection of $^{201}Tl$. But the uniformity correction map in SPECT uses $^{99m}Tc$ uniformity correction map. Thus, we want to compare the image quality when it uses $^{99m}Tc$ uniformity correction map and when it uses $^{201}Tl$ uniformity correction map. Materials and Methods Phantom study is performed. We take the data by Asan medical center daily QC condition with flood phantom including $^{201}Tl$ 21.3 kBq/mL. After postprocessing with this data, we analyze CFOV integral uniformity(I.U) and differential uniformity(D.U). And we take the data with Jaszczak ECT Phantom by American college of radiology accreditation program instruction including $^{201}Tl$ 33.4 kBq/mL. After post processing with this data, we analyze spatial Resolution, Integral Uniformity(I.U), coefficient of variation(C.V) and Contrast with Interactive data language program. Results In the flood phantom test, when it uses $^{99m}Tc$ uniformity correction map, Flood I.U is 3.6% and D.U is 3.0%. When it uses $^{201}Tl$ uniformity correction map, Flood I.U is 3.8% and D.U is 2.1%. The flood I.U is worsen about 5%, but the D.U is improved about 30% inversely. In the Jaszczak ECT phantom test, when it uses $^{99m}Tc$ uniformity correction map, SPECT I.U, C.V and contrast is 13.99%, 4.89% and 0.69. When it uses $^{201}Tl$ uniformity correction map, SPECT I.U, C.V and contrast is 11.37%, 4.79% and 0.78. All of data are improved about 18%, 2%, 13% The spatial resolution was no significant changes. Conclusion In the flood phantom test, Flood I.U is worsen but Flood D.U is improved. Therefore, it's uncertain that an image quality is improved with flood phantom test. On the other hand, SPECT I.U, C.V, Contrast are improved about 18%, 2%, 13% in the Jaszczak ECT phantom test. This study has limitations that we can't take all variables into account and study with two phantoms. We need think about things that it has a good effect when doctors decipher the nuclear medicine image and it's possible to improve the image quality using the uniformity correction map of other radionuclides other than $^{99m}Tc$, $^{201}Tl$ when we make other nuclear medicine examinations.

• Journal of Intelligence and Information Systems
• /
• v.25 no.2
• /
• pp.141-166
• /
• 2019

Recently, channels like social media and SNS create enormous amount of data. In all kinds of data, portions of unstructured data which represented as text data has increased geometrically. But there are some difficulties to check all text data, so it is important to access those data rapidly and grasp key points of text. Due to needs of efficient understanding, many studies about text summarization for handling and using tremendous amounts of text data have been proposed. Especially, a lot of summarization methods using machine learning and artificial intelligence algorithms have been proposed lately to generate summary objectively and effectively which called "automatic summarization". However almost text summarization methods proposed up to date construct summary focused on frequency of contents in original documents. Those summaries have a limitation for contain small-weight subjects that mentioned less in original text. If summaries include contents with only major subject, bias occurs and it causes loss of information so that it is hard to ascertain every subject documents have. To avoid those bias, it is possible to summarize in point of balance between topics document have so all subject in document can be ascertained, but still unbalance of distribution between those subjects remains. To retain balance of subjects in summary, it is necessary to consider proportion of every subject documents originally have and also allocate the portion of subjects equally so that even sentences of minor subjects can be included in summary sufficiently. In this study, we propose "subject-balanced" text summarization method that procure balance between all subjects and minimize omission of low-frequency subjects. For subject-balanced summary, we use two concept of summary evaluation metrics "completeness" and "succinctness". Completeness is the feature that summary should include contents of original documents fully and succinctness means summary has minimum duplication with contents in itself. Proposed method has 3-phases for summarization. First phase is constructing subject term dictionaries. Topic modeling is used for calculating topic-term weight which indicates degrees that each terms are related to each topic. From derived weight, it is possible to figure out highly related terms for every topic and subjects of documents can be found from various topic composed similar meaning terms. And then, few terms are selected which represent subject well. In this method, it is called "seed terms". However, those terms are too small to explain each subject enough, so sufficient similar terms with seed terms are needed for well-constructed subject dictionary. Word2Vec is used for word expansion, finds similar terms with seed terms. Word vectors are created after Word2Vec modeling, and from those vectors, similarity between all terms can be derived by using cosine-similarity. Higher cosine similarity between two terms calculated, higher relationship between two terms defined. So terms that have high similarity values with seed terms for each subjects are selected and filtering those expanded terms subject dictionary is finally constructed. Next phase is allocating subjects to every sentences which original documents have. To grasp contents of all sentences first, frequency analysis is conducted with specific terms that subject dictionaries compose. TF-IDF weight of each subjects are calculated after frequency analysis, and it is possible to figure out how much sentences are explaining about each subjects. However, TF-IDF weight has limitation that the weight can be increased infinitely, so by normalizing TF-IDF weights for every subject sentences have, all values are changed to 0 to 1 values. Then allocating subject for every sentences with maximum TF-IDF weight between all subjects, sentence group are constructed for each subjects finally. Last phase is summary generation parts. Sen2Vec is used to figure out similarity between subject-sentences, and similarity matrix can be formed. By repetitive sentences selecting, it is possible to generate summary that include contents of original documents fully and minimize duplication in summary itself. For evaluation of proposed method, 50,000 reviews of TripAdvisor are used for constructing subject dictionaries and 23,087 reviews are used for generating summary. Also comparison between proposed method summary and frequency-based summary is performed and as a result, it is verified that summary from proposed method can retain balance of all subject more which documents originally have.