Bacillus licheniformis와 Lactobacillus casei 2단 발효에 의해 조제된 효소식품

강조은,문기성 한국유산균·프로바이오틱스학회 2016 Current Topic in Lactic Acid Bacteria and Probioti Vol.4 No.1

1-B-12 strain presenting α-amylase and protease activity was isolated from a local Doenjang product. The strain was identified as Bacillus licheniformis by 16S rRNA gene sequencing and named B. licheniformis 1-B-12. A combination of raw materials with unpolished rice (50%), rice bran (30%), and soybean (20%) was best for the enzyme activities. A lactic acid bacterium Lactobacillus casei GW140 was applied for co-culture with B. licheniformis 1-B-12 aiming at the production of a function-added enzyme food. Two-step fermentation, where 24 h fermentation with B. licheniformis 1-B-12 and additional 12 h fermentation with L. casei GW140 were performed, resulted in maintenance of both of the enzyme activities and high level of lactic acid bacteria cell count. Furthermore the freeze-dried enzyme food sample was stable at 25℃ for 45 d which was the last day we tested.

A Novel Lactobacillus casei LP1 Producing 1,4-Dihydroxy-2-Naphthoic Acid, a Bifidogenic Growth Stimulator

강조은,김태정,문기성 한국식품영양과학회 2015 Preventive Nutrition and Food Science Vol.20 No.1

1,4-Dihydroxy-2-naphthoic acid (DHNA) is a bifidogenic growth stimulator (BGS) and could be a functional food ingredient since bifidobacteria are beneficial for human health. For that reason, lactic acid bacteria producing DHNA have been screened. A lactic acid bacterium LP1 strain isolated from a natural cheese was confirmed to produce DHNA, analyzed by a HPLC method. The strain was identified as Lactobacillus casei by 16S rRNA gene sequence analysis. The cell-free supernatant of fermented whey produced by L. casei LP1 presented the BGS activity for three bifidobacterial strains such as Bifidobacterium longum subsp. infantis KCTC 3127, Bifidobacterium bifidum KCTC 3202, and Bifidobacterium breve KCTC 3220 which were human-originated. To the best of our knowledge, a L. casei strain which can produce DHNA was firstly identified in this study.

생성 AI의 혐오 화행 이해 평가를 위한 기초 연구

강조은 ( Kang Joeun ),김유진 ( Kim Yujin ),김한샘 ( Kim Hansaem ) 한국텍스트언어학회 2023 텍스트언어학 Vol.55 No.-

Hate speech comes in many forms within online comments, and detecting implicit hate speech is more difficult than explicit hate speech. The primary objective of AI-based hate speech research is to identify implicit and high context hate speech, demanding a precise understanding of the fundamental conditions that define hate speech. However, the ever evolving and replicating nature of hate speech limits explicit judgment studies relying on pattern and dictionary methods. In this study, the Felicity condition of hate speech was used as a measure to evaluate the model's understanding of hate speech. Accurate detection of hate speech, continually evolving, necessitates an intrinsic understanding of the concept of hate. In this study, hate is defined as expressions that rationalize and propagate discrimination through insulting, demeaning, belittling, threatening, or inciting violence based on innate attributes that individuals cannot choose. Additionally, we analyze the Felicity condition of hate speech, which has not been explicitly defined in previous studies, categorizing them into propositional content conditions, preparatory preconditions, sincerity conditions, and essential conditions based on Searle's theory. Two experiments were designed to evaluate the language model's ability to comprehend hate speech. In the first experiment, open-ended prompt engineering was used to evaluate the output of the Felicity conditions for hateful sentences. The results showed that open-ended prompt engineering was not appropriate for evaluating the model's understanding of the Felicity condition. In the second experiment, an assessment item design method for educational evaluation was adopted, incorporating true-false, linked, short-answer, and essay questions based on the analyzed the Felicity conditions. Results indicated that prompt engineering did not enhance the output of hate speech. The experimental results were encouraging, highlighting that every type of assessment item yielded scores exceeding 50 points. Notably, while short-answer and essay questions demand more advanced cognitive skills and consequently recorded relatively lower scores, they too surpassed the baseline threshold. These results suggest that a variety of evaluation questions and techniques should be applied for a more in-depth assessment of the model's comprehension capabilities, particularly the essential components of hate speech.

Bacillus licheniformis와 Lactobacillus casei 2단 발효에 의해 조제된 효소식품

강조은(Jo-Eun Kang),문기성(Gi-Seong Moon) 한국유산균프로바이오틱스학회 2016 Current Topic in Lactic Acid Bacteria and Probioti Vol.4 No.2

1-B-12 strain presenting α-amylase and protease activity was isolated from a local Doenjang product. The strain was identified as Bacillus licheniformis by 16S rRNA gene sequencing and named B. licheniformis 1-B-12. A combination of raw materials with unpolished rice (50%), rice bran (30%), and soybean (20%) was best for the enzyme activities. A lactic acid bacterium Lactobacillus casei GW140 was applied for co-culture with B. licheniformis 1-B-12 aiming at the production of a function-added enzyme food. Two-step fermentation, where 24 h fermentation with B. licheniformis 1-B-12 and additional 12 h fermentation with L. casei GW140 were performed, resulted in maintenance of both of the enzyme activities and high level of lactic acid bacteria cell count. Furthermore the freeze-dried enzyme food sample was stable at 25℃ for 45 d which was the last day we tested.

빈카 마이너 추출물의 항균 및 면역활성 연구

김준섭,강조은,유일환,정경환,문기성,이향렬 한국유화학회 2015 한국응용과학기술학회지 Vol.32 No.1

EtOH 추출법과 열수 추출법을 사용하여 빈카민 등의 빈카 알칼로이드 화합물을 함유한 빈카 마이너 추출물을 얻었다. 이들 추출물을 사용하여 비듬균, 효모, 그람양성 및 음성세균에 대한 항균활성을 측정 하였다. 비듬균(Malassezia furfur)에 대하여 표준물질인 빈카민과 EtOH 추출물은 항균활성을 보이 지 않은 반면 열수 추출물은 항균활성을 나타내었다. 또한 열수 추출물은 그람양성 및 음성세균에 속하는 바실러스(Bacillus sp.)와 대장균(Escherichia coli)에 대해서도 항균활성을 나타내었다. 추출물의 세포독성 은 HaCaT 세포(인간 케라티노사이트 세포), HT-29 세포(인간 결직장 선종암 세포), Raw 세포(인간 대 식세포)를 대상으로 측정하였으며, 해당 세포들에 대해 특별한 세포독성은 발견 되지 않았다. 또한 형광 기반의 탐색물질인 디클로로플루오신 디아세테이트(DCFDA)를 이용한 활성산소종을 측정하였다. 그 결과 HaCaT 세포와 HT-29 세포군에서 활성산소종의 형성이 약 20% 정도 증가하는 것을 알 수 있었다. Vinca alkaloid compounds including vincamine from Vinca minor L. have been extracted by ethanol and hot water extraction methods. Antimicrobial properties of those extracts have been investigated against dandruff causing microorganism Malassezia furfur, yeast, Gram positive and negative bacteria. Vincamine standard and ethanol extract showed no sign of antimicrobial activity, whereas hot-water extract had the activity against M. furfur. Furthermore hot water extract had antimicrobial activity against Gram positive Bacillus sp. and Gram negative Escherichia coli. Cytotoxic properties of those extracts have also been investigated with HaCaT cell (human keratinocyte), HT-29 cell (human colorectal adenocarcinoma cell) and Raw cell, showing no significant cytotoxic effects. We also measured the ROS using dichlorofluorescein diacetate (DCFDA), a popular fluorescence-based probe for reactive oxygen species detection. The result showed the increasement of reactive oxygen species formation (20%) in HaCaT and HT-29 cell lines.

빈카 마이너 추출물에 함유된 알칼로이드들의 분광학적 분석 및 항산화 효능 연구

김준섭,주지혜,강조은,장선동,정경환,문기성,이향렬 한국유화학회 2015 한국응용과학기술학회지 Vol.32 No.1

빈카마이너 L.의 알칼로이드는 HPLC의 PDA 검출기를 이용하여 측정하고 이를 표준시료인 빈카민의 자외선/가시광선 흡수 스펙트럼과 비교하여 분석하였다. 플라보노이드류를 포함하여 추출물에서 9개 이상의 화합물에 대한 자외선/가시광선 흡수 스펙트럼을 얻었다. 추출방법에 따라 에탄올 추출물과 열수 추출물을 얻고 이를 DPPH 라디칼 소거법에 의해 항산화 효과를 조사하였다. 그 결과 에탄올 추출 물과 열수 추출물 모두 비교군인 케르세틴과 비타민 C 대비 약 14~15%의 항산화 효과를 보여 추출법에 상관없이 유사한 결과를 보였다. Alkaloids of Vinca minor L. were characterized by HPLC equipped with the photo diode array detector and compared their UV/vis absorption spectra with an index material such as vincamine. UV/vis absorption spectra of more than 9 compounds belong to various classes such as flavonoids were shown and characterized. Extracts with ethanol or hot water were tested against the DPPH (2,2-diphenyl-1-picrylhydrazyl) free-radical to investigate their antioxidant activity. Based on the results, those extracts show about 14-15% of antioxidant activity of quercetine and vitamin C used as standards.

언어모델의 편향 개선을 위한 프롬프트 엔지니어링 연구 : ChatGPT를 활용한 정치인 감성분석 말뭉치를 중심으로

김유진,강조은,김한샘 한국코퍼스언어학회 2023 Corpus Linguistics Research Vol.8 No.1

This study aims to identify and address the inherent political bias in ChatGPT by utilizing a sentiment analysis task. We set up representative figures from each political faction, asked chatgpt to write about the politicians using various prompts, and then sentimentally analyzed their outputs to determine the bias of ChatGPT. We found that ChatGPT is more positively biased toward liberal politicians in South Korea. We also found that ChatGPT's bias can be reduced by combining general narratives that encourage neutral writing or by refining the prompts with variables such as tone and writing style. This study provides important insights into the responsible use of AI and how to improve its bias.

빈카 마이너 추출물의 항균 및 면역활성 연구

김준섭,강조은,유일환,정경환,문기성,이향렬,Kim, Jun-Sub,Kang, Jo-Eun,Yu, Il-Hwan,Jung, Kyung-Hwan,Moon, Gi-Seong,Lee, Hyang-Yeol 한국응용과학기술학회 2015 한국응용과학기술학회지 Vol.32 No.1

EtOH 추출법과 열수 추출법을 사용하여 빈카민 등의 빈카 알칼로이드 화합물을 함유한 빈카 마이너 추출물을 얻었다. 이들 추출물을 사용하여 비듬균, 효모, 그람양성 및 음성세균에 대한 항균활성을 측정 하였다. 비듬균(Malassezia furfur)에 대하여 표준물질인 빈카민과 EtOH 추출물은 항균활성을 보이지 않은 반면 열수 추출물은 항균활성을 나타내었다. 또한 열수 추출물은 그람양성 및 음성세균에 속하는 바실러스(Bacillus sp.)와 대장균(Escherichia coli)에 대해서도 항균활성을 나타내었다. 추출물의 세포독성은 HaCaT 세포(인간 케라티노사이트 세포), HT-29 세포(인간 결직장 선종암 세포), Raw 세포(인간 대식세포)를 대상으로 측정하였으며, 해당 세포들에 대해 특별한 세포독성은 발견 되지 않았다. 또한 형광기반의 탐색물질인 디클로로플루오신 디아세테이트(DCFDA)를 이용한 활성산소종을 측정하였다. 그 결과 HaCaT 세포와 HT-29 세포군에서 활성산소종의 형성이 약 20% 정도 증가하는 것을 알 수 있었다. Vinca alkaloid compounds including vincamine from Vinca minor L. have been extracted by ethanol and hot water extraction methods. Antimicrobial properties of those extracts have been investigated against dandruff causing microorganism Malassezia furfur, yeast, Gram positive and negative bacteria. Vincamine standard and ethanol extract showed no sign of antimicrobial activity, whereas hot-water extract had the activity against M. furfur. Furthermore hot water extract had antimicrobial activity against Gram positive Bacillus sp. and Gram negative Escherichia coli. Cytotoxic properties of those extracts have also been investigated with HaCaT cell (human keratinocyte), HT-29 cell (human colorectal adenocarcinoma cell) and Raw cell, showing no significant cytotoxic effects. We also measured the ROS using dichlorofluorescein diacetate (DCFDA), a popular fluorescence-based probe for reactive oxygen species detection. The result showed the increasement of reactive oxygen species formation (20%) in HaCaT and HT-29 cell lines.

빈카 마이너 추출물에 함유된 알칼로이드들의 분광학적 분석 및 항산화 효능 연구

김준섭,주지혜,강조은,장선동,정경환,문기성,이향렬,Kim, Jun-Sub,Joo, Ji-Hye,Kang, Jo-Eun,Jang, Sun-Dong,Jung, Kyung-Hwan,Moon, Gi-Seong,Lee, Hyang-Yeol 한국응용과학기술학회 2015 한국응용과학기술학회지 Vol.32 No.1

빈카마이너 L.의 알칼로이드는 HPLC의 PDA 검출기를 이용하여 측정하고 이를 표준시료인 빈카민의 자외선/가시광선 흡수 스펙트럼과 비교하여 분석하였다. 플라보노이드류를 포함하여 추출물에서 9개 이상의 화합물에 대한 자외선/가시광선 흡수 스펙트럼을 얻었다. 추출방법에 따라 에탄올 추출물과 열수 추출물을 얻고 이를 DPPH 라디칼 소거법에 의해 항산화 효과를 조사하였다. 그 결과 에탄올 추출물과 열수 추출물 모두 비교군인 케르세틴과 비타민 C 대비 약 14~15%의 항산화 효과를 보여 추출법에 상관없이 유사한 결과를 보였다. Alkaloids of Vinca minor L. were characterized by HPLC equipped with the photo diode array detector and compared their UV/vis absorption spectra with an index material such as vincamine. UV/vis absorption spectra of more than 9 compounds belong to various classes such as flavonoids were shown and characterized. Extracts with ethanol or hot water were tested against the DPPH (2,2-diphenyl-1-picrylhydrazyl) free-radical to investigate their antioxidant activity. Based on the results, those extracts show about 14-15% of antioxidant activity of quercetine and vitamin C used as standards.

GPT-4를 활용한 인간과 인공지능의 한국어 사용 양상 비교 연구

박서윤,김한샘,강예지,강조은,김유진,이재원,정가연,최규리 국어국문학회 2024 국어국문학 Vol.0 No.206

본 논문은 인간과 인공지능 모델이 생성한 텍스트 간 차이점을 밝히고자 에세이를 중심으로 형태론적, 통사론적, 사회언어학적 특징 등 언어학적 층위에 따라 연구하였다. 고등학교 1학년〜3학년 학생들이 작성한 에세이를 인간 생성 에세이 데이터셋으로 사용하였으며, 인공지능 모델 생성 데이터로는 GPT-4가 생성한 에세이를 사용하였다. 형태론적 층위 분석을 위해 품사 빈도와 어휘 다양도, 어휘 밀도 그리고 어휘 정교도를 사용하였으며, 카이 제곱 검정, 윌콕슨의 부호-서열 검정 등을 활용하여 인간과 GPT-4의 어휘 사용 양상 간 차이의 유의미성을 검정하였다. 검정 결과 품사 빈도, 어휘 다양도, 어휘 밀도에서는 유의 확률(p-value)이 모두 0.05 이하로 나타나 실제로 어휘 사용 양상에 통계적으로 유의미한 차이가 있음을 확인할 수 있었다. 어휘 정교도에서는 인간과 GPT-4 간 통계적으로 유의미한 차이가 드러나지 않았다. 통사론적 층위 분석은 문장 기초 특성과 의존 구문 분석 결과를 바탕으로 진행하였으며, 분석 결과 인간이 GPT-4보다 각 특성별로 편차와 분포가 넓은 양상을 보여 다양한 문장을 사용하여 에세이를 작성하고 있음을 확인할 수 있었다. 한편 구문 분석을 수행한 결과, 인간이 생성한 문장에는 VP와 AP가 상대적으로 빈번하게 등장하였고 GPT-4 생성 문장의 경우 NP와 수식 관련 성분이 자주 등장하였다. 언어 사용자 시각에서의 주체별 언어적 특징 분석을 위해 성별어 분석을 진행한 결과, GPT-4는 성별별 기호어, 혐오어, 그리고 성별 공통 혐오어를 거의 사용하지 않고 성별 공통 기호어는 자주 사용하는 경향을 보였다. 이에 따라 GPT-4는 성별 편향 혹은 혐오 표현에 대해 회피하도록 학습되었음을 확인하였다. 감성 분석의 경우에는 GPT-4가 생성한 에세이가 인간의 에세이보다 긍정의 극성을 가지는 비중이 높아, 학습 데이터 구성 시 부정적인 감성이 드러나는 데이터를 필터링하였거나 생성 시에 극성에 대한 조정이 가해졌을 가능성이 있음이 드러났다. This paper aims to enlighten the differences between texts which were generated by human and AI model, focusing on ‘essays’. We analyzed Korean usage patterns according to linguistic layers such as morphological, syntactic, and sociolinguistic features. In an aspect of dataset, we used essays written by high school students in the 1st to 3rd grades as ‘human-generated essays’, and essays generated by GPT-4 as data for ‘GPT-4 generated essays’. For morphological feature analysis, we used part-of- speech tag frequency, Type Token Ratio (TTR), lexical density, and lexical sophistication. We found that human and GPT-4 have different patterns of Part-Of-Speech by using statistical analysis. Also we observed that there are differences between human and GPT-4 in lexical features except lexical sophistication. The syntactic analysis was based on basic sentence characteristics and dependency parsing. The results showed that humans’ sentences have a wider range of deviations and distributions for each sentence characteristic than GPT-4s’, which indicates that humans write a variety of sentences than AI model. On the other hand, the dependency parsing showed that ‘VP’ and ‘AP’ appeared relatively frequently in human- generated sentences, while ‘NP’ and modifier(MOD) appeared frequently in GPT-4-generated sentences. To examine the linguistic features in the sight of language users, we analyzed gender-biased words, and found that GPT-4 rarely used gender-specific preferred words, gender-aversive words, and gender-common aversive words but frequently used gender-common preferred words. This suggests that GPT-4 has been pre-trained to avoid gender-biased or hate speech. In the case of sentiment analysis, GPT-4s showed more positive emotions than humans, and this also revealed that as same as gender bias results, GPT-4 has been pre-trained to generate positive sentences that negative sentences.