• Pediatric Gastroenterology, Hepatology & Nutrition
• /
• v.8 no.1
• /
• pp.21-30
• /
• 2005

Purpose: The aim of this study was to identify the clinical features and natural history of dietary protein induced proctocolitis (DPIPC) and to detect the causative foods of DPIPC, and to evaluate the effect of elimination of the foods on the course of the disease. Methods: Between March 2003 and July 2004, data from 30 consecutive patients with DPIPC who were followed for over 6 months, was reviewed. The diagnostic criterion used for DPIPC was an increase in the number of eosinophils in the lamina propria (${\geq}60per$ 10 high-power fields). In breast feeding mothers, 5 highly allergenic foods were eliminated from the maternal diet for 7 days, namely, allergenic food groups such as dairy products, eggs, nuts and soybean, fish and shellfish, and wheat and buckwheat. We observed the disappearance or appearance of hematochezia after elimination or challenge with the offending foods. Results: Before diagnosis infants were breast-fed (93.3%) or formula-fed (6.7%). Mean age at symptom onset was $11.5{\pm}5.1$ (5~24) weeks, and mean age at diagnosis was $17.8{\pm}9.5$ (8~56) weeks. Duration from symptom onset to diagnosis was $6.3{\pm}6.7$ (0~36) weeks. Mean peripheral blood eosinophil count was $478{\pm}320$ (40~1,790)/$mm^3$ and eosinophilia (> $250/mm^3$) was observedin 90.0% of patients. None of patients were found to have an increased serum IgE level. Of 15 patients that received sigmoidoscopy, nodular hyperplasia with erosion was observed in 93.3%. Of 27 patients whose mother ate the diet eliminated the 5 food groups, hematochezia diappeared in 74.1% of patients. Offending foods were identified as dairy products (37.5%), wheat and buckwheat (27.5%), fish and shellfish (20.0%), nuts and soybean (7.5%) and eggs (7.5%). A free maternal diet without patient's clinical symptoms was achieved at $29.4{\pm}8.7$ (9~44) weeks of patient's age, and a free baby diet without blood in stools was achieved at $37.5{\pm}9.7$ (12~56) weeks of age. Conclusion: DPIPC commonly occurs in exclusively breast-fed babies. Elimination of the above-mentioned 5 hyper-allergenic food groups from the maternal diet for 7days enables the detection of the offending foods. DPIPC is a transient disorder and 96.0% of patients can tolerate the offending foods at 12 months of age.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.55 no.1
• /
• pp.76-82
• /
• 2010

Experiments were conducted to investigate seed persistence of Chinese milk vetch(CMV) in naturally reseeded rice field in 2007~2009. The seed and pods with seeds were buried in rice field at 0, 5 and 10 cm depths and retrieved from the field at one to three month intervals from July to the following year March and determined change of seed dormancy and viability. In the second experiment, persistence of the CMV seeds in the naturally reseeded rice field at different tillage methods and soil depths were also investigated after rice harvest in autumn. Burial depths and durations affected recovery rate, dormancy and viability of CMV seed. The viability loss was faster and greater in the seed than the pod with seeds and on the soil surface than the 5 or 10 cm burial depths. The recovery rate of CMV seed was decreased starting from one month as seed burial and it was significantly decreased to 52~65% for the seed in September. However, unlike the seed burial, the nearly 100% CMV seeds were recovered for burial as pod with seeds even after four months burial in both 0 and 5 cm depths. However, the recovery rate was sharply declined to below 30% at October in 2007 in both seed and pods with seeds and in the 2008/2009 experiment. the 15~47% of CMV seeds still remained even after October. The CMV had high seed dormancy of 95%, showing only 4~5% germination at the beginning in June but the seed germination increased to 25 to 35% in seed and 55 to 61% in pod with seeds in September due to breakage of hard seed dormancy. The viability loss was faster in the seed than in the pod with seeds regardless of depths of placement in the soil base on decayed seeds. Also the seed placed on the soil surface lost viability faster than the 5~10 burial depths. On the other hand, field observation in the naturally reseeded CMV rice field showed that as many as 917~2,185 CMV seeds $m^2$ were from the 0~15 cm soil depth in the rotary tillage and 250~10,105 CMV seeds in minimum tillage treatmints. The recovered seed germinated 25~33%, 23~43% but still had high percentage of hard seed having 64~72% and 51~77% even after rice harvest in autumn. These results indicate that freshly harvested CMV seeds had high level of primary dormancy and the dormancy was gradually broken in soil with time during rice cultivation periods and appreciable number of CMV seeds remained even 4 month after burial in soil. CMV plant regenerated naturally from the remained seed bank at rice harvest time in autumn. The CMV seedling still emerged even after 2 years of continuous destructive killing of emerged CMV plant by rotary tillage in naturally reseeded CMV plant in rice field, indicating that CMV seeds do persistent as least two years in soil.

• Korean Journal of Ecology and Environment
• /
• v.47 no.3
• /
• pp.135-145
• /
• 2014

The mountain forest vegetation of Gyebangsan (1,577 m) in Odaesan National Park is classified into deciduous broad-leaved forest, mountain valley forest, coniferous forest, subalpine coniferous forest, subalpine deciduous forest, plantation forest, and other vegetation which includes Actinidia argute community and agricultural land. As for the number of communities distributed in the each forest vegetation which were categorized by the physiognomy classification, deciduous broad-leaved forest had 33 communities, mountain valley forest 41 communities, coniferous forest 8 communities, subalpine coniferous forest 4 communities, subalpine deciduous forest 2 communities, plantation forest 6 communities and other vegetation 4 communities. Regarding the distribution rate of communities in the vegetation, in the deciduous broad-leaved forest. Quercus mongolica community accounted for 80.226% with $30,909,942.967m^2$, followed by Quercus variabilis community of 2.771% with $1,067,479.335m^2$. 55.463% of deciduous broad-leaved forest in the Gyebangsan had Quercus mongolica as a dominant or second dominant species. In the mountain valley forest, Fraxinus rhynchophylla - Juglans mandshurica community accounted for 10.955%. And there were ten mixed communities having Fraxinus rhynchophylla and upper layer at a similar level of coverage, taking up 32.776%. In the coniferous forest, Pinus densiflora and the community living with Pinus densiflora accounted for 100%, showing that the coniferous forest has the community with Pinus densiflora as a dominant species at upper layer. For other vegetation, subalpine coniferous forest had a total of four communities including Abies holophylla - Quercus mongolica community, and accounted for 4.980% of vegetation area of Odaesan National Park. Two communities including Betula ermani - Cornus controversa community were found in the subalpine deciduous forest, taking up 0.006% of total vegetation area of Odaesan National Park. Regarding plantation forest, Larix leptolepis was planted the most with 51.652%, followed by Betula platyphylla var. japonica with 38.975%, and Pinus koraiensis with 7.969%. These three species combined accounted for 98.565%. In conclusion, the forest vegetation found in the Gyebangsan of Odaesan National Park has Quercus mongolica as a dominant species at the top layer. A lot of other communities related with this species are expected to be quickly replaced due to vegetation succession and climatic causes. Therefore, Quercus mongolica is expected to become the main species in the deciduous broad-leaved forest, Fraxinus rhynchophylla, Juglans mandshurica and Fraxinus mandshurica in the mountain valley forest. Around the border line between deciduous broad-leaved forest and mountain valley forest, highly humid valley area is expected to be quickly taken up by Cornus controversa and Fraxinus mandshurica, and the slope area by Quercus mongolica. However, in the subalpine coniferous forest, the distribution rate of deciduous broad-leaved trees is expected to increase due to climate warming.

• Tuberculosis and Respiratory Diseases
• /
• v.50 no.4
• /
• pp.450-461
• /
• 2001

Background : The underlying pathogenesis of radiation-induced lung fibrosis (RTLF) has not been very well defined. However, the role of TGF-$\beta$ in the generation of RTLF has been a major focus because there is an increase in the expression of both the TGF-${\beta}m$-RNA and its protein preceding RTLF lesions. The down stream signal after a TGF-$\beta$ stimulated lung fibrosis includes the activation of many mediators such as Smad and c-Jun N-terminal kinase (JNK) through TAK1. It is we hypothesized that JNK activation may play a pivotal role in RTLF pathogenesis through increased transcription of the fibrogenic cytokines. The present study evaluates JNK activity in alveolar macrophages after irradiation and the relationship between JNK activity and the amount of collagen in the lung tissues. Methods : C57BL/6 mice(20-25 gr, males) received chlorotetracycline(2g/L) in their drinking water 1 week prior to irradiation and continuously there after. The mice were irradiated once with 1400 cGy of $60CO{\gamma}$-ray over the whole chest. The cellular composition of the whole lung bronchoalveoalr lavage fluids(BALF), elastin expression in the lung tissues, the level of hydroxyproline in lung tissues, and an in vitro JNK assay was measured before irradiation and one, four, and eight weeks after irradiation (RT). Results : The volumes of BALF retrieved from instilled 4 mL of saline with 2% heparin were 3.7-3.8 mL for each group. The cell numbers were similar before($4.1{\times}10^4{\pm}0.5{\times}10^4/mL$) and 1 week($3.1{\times}10^4{\pm}0.5{\times}10^4/mL$) after RT. At four and eight weeks after RT, the cell number reached to $14.0{\times}10^4{\pm}1.5{\times}10^4mL$ and $10.0{\times}10^4{\pm}1.3{\times}10^4/mL$, respectively. There we no changes in the lymphocytes and neutrophils population observed in the BALF after RT. The H-E stain of the lung tissues did not show any structural and fibrotic change in the lung tissues at 4 and 8 weeks after RT. In addition, the amount of elastin and collagen were not different on Verhoeff staining of the lung tissues before RT to eight weeks after RT. The hydroxyproine content was measured with the left lung dissected from the left main bronchus. The lung were homogenized and hydrolyzed with 6 N Hel for 12 hours at $110^{\circ}C$ then measured as previously described. The content of hydroxyproline, standardized with a lung protein concentration, reached a peak 4 weeks after RT, and thereafter showed a plateau. AnIn vitro JNK assay using c-$Jun_{1-79}$-GST sepharose beads were performed with the alveolar macrophages obtained from the BAL. JNK activity was not detected prior to RT, However, the JNK activity increased from one week after RT and reached a peak four weeks after RT. Conclusion : JNK may be involved in the pathogenesis because the JNK activity showed similar pattern observed with the hydroxyproine content. However, it is necessary to clarify that the JNK increases the transcription of fibrogenic cyiokines through the transcription factor.

• Journal of Korean Society of Forest Science
• /
• v.78 no.3
• /
• pp.287-301
• /
• 1989

A tree of Ailanthus altissima Swingle was fastened with a plastic band, 19mm wide, around the stem 180cm above ground level and was left to grow under this condition for one year, By removal of this band the tumorous tissue gradually developed and the tree bearing distinct tumorous tissue, an overgrowth surrounding the stem, was harvested two years after the band removal. For the investigation of this tumorous part and its comparison with adjacent normal parts in the anatomical features of individual elements, the tumorous part and parts directly and 40cm above and below the tumorous part were obtained from the tree. The tumor wood having remarkably wider growth increment occurred in the 3rd growth ring the first year after removal of the fastened band, and the barrier zone which delimited the discolored wood from the normal-colored wood inwards appeared u1 the intra-2nd growth ring produced during the fastened period in the tumorous part and the false ring-like zones equivalent to barrier Zone were shown in the normal-colored 2nd growth rings of the parts directly and 40cm above and below the tumorous part, as well. The tumor wood, the 3rd growth ring, and proportion of the 2nd growth ring formed after barrier zone in the tumorous part shared common characteristics in the irregular growth ring boundary, misshapen and shorter individual fibers and vessel elements, and large ray widths and heights. The springwood pores were smaller in diameter in the tumor wood, and the larger radial and smaller tangential diameters of summerwood solitary pores and individual pores consisting of pore multiples in proportion of the 2nd growth ring formed after the barrier zone were transformed into near-isodiametric in the tumor wood, the 3rd growth ring, in the tumorous part. Only in proportion of the 2nd growth ring formed after the barrier zone were transformed into near-isodiametric in the tumor wood, the 3rd growth ring, in the tumorous part, ray densities greatly increased. And the massive tumor wood was caused not by cell size but by cell number because the radial and tangential diameters of fibers in the tumor wood, the 3rd growth ring, in the tumorous part were not sufficiently different from those in the same aged growth rings of the directly and 40cm above and below the tumorous part.

• Journal of Oral Medicine and Pain
• /
• v.31 no.4
• /
• pp.283-296
• /
• 2006

Along with the development of scientific technologies, health care has been growing remarkably, and as the social life quality improves with increasing interest in health, the demand for medical service is rapidly increasing. However, medical accident and medical dispute also are rapidly increasing due to various factors such as, increasing sense of people's right, lack of understanding in the nature of medical practice, over expectation on medical technique, commercialize medical supply system, moral degeneracy and unawareness of medical jurisprudence by doctors, widespread trend of mutual distrust, and lack of systematized device for solution of medical dispute. This study analysed 30 cases of civil suit in the year between 1994 to 2004, which were selected among the medical dispute cases in dental field with the judgement collected from organizations related to dentistry and department of oral medicine, Yonsei university dental hospital. The following results were drawn from the analyses: 1. The distribution of year showed rapid increase of medical dispute after the year 2000. 2. In the types of medical dispute, suit associated with tooth extraction took 36.7% of all. 3. As for the cause of medical dispute, uncomfortable feeling and dissatisfaction with the treatment showed 36.7%, death and permanent damage showed 16.7% each. 4. Winning the suit, compulsory mediation and recommendation for settlement took 60.0% of judgement result for the plaintiff. 5. For the type of medical organization in relation to medical dispute, 60.0% was found to be the private dental clinics, and 30.0% was university dental hospitals. 6. For the level of trial, dispute that progressed above 2 or 3 trials was of 30.0%. 7. For the amount of claim for damage, the claim amounting between 50 million to 100 million won was of 36.7%, and that of more than 100 million won was 13.3%, and in case of the judgement amount, the amount ranging from 10 million to 30 million won was of 40.0%, and that of more than 100 million won was of 6.7%. 8. For the number of dentist involved in the suit, 26.7% was of 2 or more dentists. 9. For the amount of time spent until the judgement, 46.7% took 11 to 20 months, and 36.7% took 21 to 30 months. 10. For medical malpractice, 46.7% was judged to be guilty, and 70% of the cases had undergone medical judgement or verification of the case by specialists during the process of the suit. 11. In the lost cases of doctors(18 cases), 72.2% was due to violence of carefulness in practice and 16.7% was due to missing of explanation to patient. Medical disputes occurring in the field of dentistry are usually of relatively less risky cases. Hence, the importance of explanation to patient is emphasized, and since the levels of patient satisfaction are subjective, improvement of the relationship between the patient and the dentist and recovery of autonomy within the group dentist are essential in addition to the reduction of technical malpractice. Moreover, management measure against the medical dispute should be set up through complement of the current doctors and hospitals medical malpractice insurance which is being conducted irrationally, and establishment of system in which education as well as consultation for medical disputes lead by the group of dental clinicians and academic scholars are accessible.

• Journal of Global Scholars of Marketing Science
• /
• v.17 no.2
• /
• pp.111-131
• /
• 2007

As a brand becomes a core asset in creating a corporation's value, brand marketing has become one of core strategies that corporations pursue. Recently, for customer relationship management, possession and consumption of goods were centered on brand for the management. Thus, management related to this matter was developed. The main reason of the increased interest on the relationship between the brand and the consumer is due to acquisition of individual consumers and development of relationship with those consumers. Along with the development of relationship, a corporation is able to establish long-term relationships. This has become a competitive advantage for the corporation. All of these processes became the strategic assets of corporations. The importance and the increase of interest of a brand have also become a big issue academically. Brand equity, brand extension, brand identity, brand relationship, and brand community are the results derived from the interest of a brand. More specifically, in marketing, the study of brands has been led to the study of factors related to building of powerful brands and the process of building the brand. Recently, studies concentrated primarily on the consumer-brand relationship. The reason is that brand loyalty can not explain the dynamic quality aspects of loyalty, the consumer-brand relationship building process, and especially interactions between the brands and the consumers. In the studies of consumer-brand relationship, a brand is not just limited to possession or consumption objectives, but rather conceptualized as partners. Most of the studies from the past concentrated on the results of qualitative analysis of consumer-brand relationship to show the depth and width of the performance of consumer-brand relationship. Studies in Korea have been the same. Recently, studies of consumer-brand relationship started to concentrate on quantitative analysis rather than qualitative analysis or even go further with quantitative analysis to show effecting factors of consumer-brand relationship. Studies of new quantitative approaches show the possibilities of using the results as a new concept of viewing consumer-brand relationship and possibilities of applying these new concepts on marketing. Studies of consumer-brand relationship with quantitative approach already exist, but none of them include sub-dimensions of consumer-brand relationship, which presents theoretical proofs for measurement. In other words, most studies add up or average out the sub-dimensions of consumer-brand relationship. However, to do these kind of studies, precondition of sub-dimensions being in identical constructs is necessary. Therefore, most of the studies from the past do not meet conditions of sub-dimensions being as one dimension construct. From this, we question the validity of past studies and their limits. The main purpose of this paper is to overcome the limits shown from the past studies by practical use of previous studies on sub-dimensions in a one-dimensional construct (Naver & Slater, 1990; Cronin & Taylor, 1992; Chang & Chen, 1998). In this study, two arbitrary groups were classified to evaluate reliability of the measurements and reliability analyses were pursued on each group. For convergent validity, correlations, Cronbach's, one-factor solution exploratory analysis were used. For discriminant validity correlation of consumer-brand relationship was compared with that of an involvement, which is a similar concept with consumer-based relationship. It also indicated dependent correlations by Cohen and Cohen (1975, p.35) and results showed that it was different constructs from 6 sub-dimensions of consumer-brand relationship. Through the results of studies mentioned above, we were able to finalize that sub-dimensions of consumer-brand relationship can viewed from one-dimensional constructs. This means that the one-dimensional construct of consumer-brand relationship can be viewed with reliability and validity. The result of this research is theoretically meaningful in that it assumes consumer-brand relationship in a one-dimensional construct and provides the basis of methodologies which are previously preformed. It is thought that this research also provides the possibility of new research on consumer-brand relationship in that it gives root to the fact that it is possible to manipulate one-dimensional constructs consisting of consumer-brand relationship. In the case of previous research on consumer-brand relationship, consumer-brand relationship is classified into several types on the basis of components consisting of consumer-brand relationship and a number of studies have been performed with priority given to the types. However, as we can possibly manipulate a one-dimensional construct through this research, it is expected that various studies which make the level or strength of consumer-brand relationship practical application of construct will be performed, and not research focused on separate types of consumer-brand relationship. Additionally, we have the theoretical basis of probability in which to manipulate the consumer-brand relationship with one-dimensional constructs. It is anticipated that studies using this construct, which is consumer-brand relationship, practical use of dependent variables, parameters, mediators, and so on, will be performed.

• Journal of Intelligence and Information Systems
• /
• v.19 no.3
• /
• pp.25-44
• /
• 2013

Accessing movie contents has become easier and increased with the advent of smart TV, IPTV and web services that are able to be used to search and watch movies. In this situation, there are increasing search for preference movie contents of users. However, since the amount of provided movie contents is too large, the user needs more effort and time for searching the movie contents. Hence, there are a lot of researches for recommendations of personalized item through analysis and clustering of the user preferences and user profiles. In this study, we propose recommendation system which uses ontology based knowledge base. Our ontology can represent not only relations between metadata of movies but also relations between metadata and profile of user. The relation of each metadata can show similarity between movies. In order to build, the knowledge base our ontology model is considered two aspects which are the movie metadata model and the user model. On the part of build the movie metadata model based on ontology, we decide main metadata that are genre, actor/actress, keywords and synopsis. Those affect that users choose the interested movie. And there are demographic information of user and relation between user and movie metadata in user model. In our model, movie ontology model consists of seven concepts (Movie, Genre, Keywords, Synopsis Keywords, Character, and Person), eight attributes (title, rating, limit, description, character name, character description, person job, person name) and ten relations between concepts. For our knowledge base, we input individual data of 14,374 movies for each concept in contents ontology model. This movie metadata knowledge base is used to search the movie that is related to interesting metadata of user. And it can search the similar movie through relations between concepts. We also propose the architecture for movie recommendation. The proposed architecture consists of four components. The first component search candidate movies based the demographic information of the user. In this component, we decide the group of users according to demographic information to recommend the movie for each group and define the rule to decide the group of users. We generate the query that be used to search the candidate movie for recommendation in this component. The second component search candidate movies based user preference. When users choose the movie, users consider metadata such as genre, actor/actress, synopsis, keywords. Users input their preference and then in this component, system search the movie based on users preferences. The proposed system can search the similar movie through relation between concepts, unlike existing movie recommendation systems. Each metadata of recommended candidate movies have weight that will be used for deciding recommendation order. The third component the merges results of first component and second component. In this step, we calculate the weight of movies using the weight value of metadata for each movie. Then we sort movies order by the weight value. The fourth component analyzes result of third component, and then it decides level of the contribution of metadata. And we apply contribution weight to metadata. Finally, we use the result of this step as recommendation for users. We test the usability of the proposed scheme by using web application. We implement that web application for experimental process by using JSP, Java Script and prot$\acute{e}$g$\acute{e}$ API. In our experiment, we collect results of 20 men and woman, ranging in age from 20 to 29. And we use 7,418 movies with rating that is not fewer than 7.0. In order to experiment, we provide Top-5, Top-10 and Top-20 recommended movies to user, and then users choose interested movies. The result of experiment is that average number of to choose interested movie are 2.1 in Top-5, 3.35 in Top-10, 6.35 in Top-20. It is better than results that are yielded by for each metadata.

• Journal of Intelligence and Information Systems
• /
• v.26 no.3
• /
• pp.149-169
• /
• 2020

"The Urban Renewal New Deal project", one of the government's major national projects, is about developing underdeveloped areas by investing 50 trillion won in 100 locations on the first year and 500 over the next four years. This project is drawing keen attention from the media and local governments. However, the project model which fails to reflect the original characteristics of the area as it divides project area into five categories: "Our Neighborhood Restoration, Housing Maintenance Support Type, General Neighborhood Type, Central Urban Type, and Economic Base Type," According to keywords for successful urban regeneration in Korea, "resident participation," "regional specialization," "ministerial cooperation" and "public-private cooperation", when local governments propose urban regeneration projects to the government, they can see that it is most important to accurately understand the characteristics of the city and push ahead with the projects in a way that suits the characteristics of the city with the help of local residents and private companies. In addition, considering the gentrification problem, which is one of the side effects of urban regeneration projects, it is important to select and implement urban regeneration types suitable for the characteristics of the area. In order to supplement the limitations of the 'Urban Regeneration New Deal Project' methodology, this study aims to propose a system that recommends urban regeneration types suitable for urban regeneration sites by utilizing various machine learning algorithms, referring to the urban regeneration types of the '2025 Seoul Metropolitan Government Urban Regeneration Strategy Plan' promoted based on regional characteristics. There are four types of urban regeneration in Seoul: "Low-use Low-Level Development, Abandonment, Deteriorated Housing, and Specialization of Historical and Cultural Resources" (Shon and Park, 2017). In order to identify regional characteristics, approximately 100,000 text data were collected for 22 regions where the project was carried out for a total of four types of urban regeneration. Using the collected data, we drew key keywords for each region according to the type of urban regeneration and conducted topic modeling to explore whether there were differences between types. As a result, it was confirmed that a number of topics related to real estate and economy appeared in old residential areas, and in the case of declining and underdeveloped areas, topics reflecting the characteristics of areas where industrial activities were active in the past appeared. In the case of the historical and cultural resource area, since it is an area that contains traces of the past, many keywords related to the government appeared. Therefore, it was possible to confirm political topics and cultural topics resulting from various events. Finally, in the case of low-use and under-developed areas, many topics on real estate and accessibility are emerging, so accessibility is good. It mainly had the characteristics of a region where development is planned or is likely to be developed. Furthermore, a model was implemented that proposes urban regeneration types tailored to regional characteristics for regions other than Seoul. Machine learning technology was used to implement the model, and training data and test data were randomly extracted at an 8:2 ratio and used. In order to compare the performance between various models, the input variables are set in two ways: Count Vector and TF-IDF Vector, and as Classifier, there are 5 types of SVM (Support Vector Machine), Decision Tree, Random Forest, Logistic Regression, and Gradient Boosting. By applying it, performance comparison for a total of 10 models was conducted. The model with the highest performance was the Gradient Boosting method using TF-IDF Vector input data, and the accuracy was 97%. Therefore, the recommendation system proposed in this study is expected to recommend urban regeneration types based on the regional characteristics of new business sites in the process of carrying out urban regeneration projects."

• Journal of Intelligence and Information Systems
• /
• v.26 no.2
• /
• pp.79-104
• /
• 2020

Recently, as deep learning has attracted attention, the use of deep learning is being considered as a method for solving problems in various fields. In particular, deep learning is known to have excellent performance when applied to applying unstructured data such as text, sound and images, and many studies have proven its effectiveness. Owing to the remarkable development of text and image deep learning technology, interests in image captioning technology and its application is rapidly increasing. Image captioning is a technique that automatically generates relevant captions for a given image by handling both image comprehension and text generation simultaneously. In spite of the high entry barrier of image captioning that analysts should be able to process both image and text data, image captioning has established itself as one of the key fields in the A.I. research owing to its various applicability. In addition, many researches have been conducted to improve the performance of image captioning in various aspects. Recent researches attempt to create advanced captions that can not only describe an image accurately, but also convey the information contained in the image more sophisticatedly. Despite many recent efforts to improve the performance of image captioning, it is difficult to find any researches to interpret images from the perspective of domain experts in each field not from the perspective of the general public. Even for the same image, the part of interests may differ according to the professional field of the person who has encountered the image. Moreover, the way of interpreting and expressing the image also differs according to the level of expertise. The public tends to recognize the image from a holistic and general perspective, that is, from the perspective of identifying the image's constituent objects and their relationships. On the contrary, the domain experts tend to recognize the image by focusing on some specific elements necessary to interpret the given image based on their expertise. It implies that meaningful parts of an image are mutually different depending on viewers' perspective even for the same image. So, image captioning needs to implement this phenomenon. Therefore, in this study, we propose a method to generate captions specialized in each domain for the image by utilizing the expertise of experts in the corresponding domain. Specifically, after performing pre-training on a large amount of general data, the expertise in the field is transplanted through transfer-learning with a small amount of expertise data. However, simple adaption of transfer learning using expertise data may invoke another type of problems. Simultaneous learning with captions of various characteristics may invoke so-called 'inter-observation interference' problem, which make it difficult to perform pure learning of each characteristic point of view. For learning with vast amount of data, most of this interference is self-purified and has little impact on learning results. On the contrary, in the case of fine-tuning where learning is performed on a small amount of data, the impact of such interference on learning can be relatively large. To solve this problem, therefore, we propose a novel 'Character-Independent Transfer-learning' that performs transfer learning independently for each character. In order to confirm the feasibility of the proposed methodology, we performed experiments utilizing the results of pre-training on MSCOCO dataset which is comprised of 120,000 images and about 600,000 general captions. Additionally, according to the advice of an art therapist, about 300 pairs of 'image / expertise captions' were created, and the data was used for the experiments of expertise transplantation. As a result of the experiment, it was confirmed that the caption generated according to the proposed methodology generates captions from the perspective of implanted expertise whereas the caption generated through learning on general data contains a number of contents irrelevant to expertise interpretation. In this paper, we propose a novel approach of specialized image interpretation. To achieve this goal, we present a method to use transfer learning and generate captions specialized in the specific domain. In the future, by applying the proposed methodology to expertise transplant in various fields, we expected that many researches will be actively conducted to solve the problem of lack of expertise data and to improve performance of image captioning.