Query-based Answer Extraction using Korean Dependency Parsing (의존 구문 분석을 이용한 질의 기반 정답 추출)
-
- Journal of Intelligence and Information Systems
- /
- v.25 no.3
- /
- pp.161-177
- /
- 2019
In this paper, we study the performance improvement of the answer extraction in Question-Answering system by using sentence dependency parsing result. The Question-Answering (QA) system consists of query analysis, which is a method of analyzing the user's query, and answer extraction, which is a method to extract appropriate answers in the document. And various studies have been conducted on two methods. In order to improve the performance of answer extraction, it is necessary to accurately reflect the grammatical information of sentences. In Korean, because word order structure is free and omission of sentence components is frequent, dependency parsing is a good way to analyze Korean syntax. Therefore, in this study, we improved the performance of the answer extraction by adding the features generated by dependency parsing analysis to the inputs of the answer extraction model (Bidirectional LSTM-CRF). The process of generating the dependency graph embedding consists of the steps of generating the dependency graph from the dependency parsing result and learning the embedding of the graph. In this study, we compared the performance of the answer extraction model when inputting basic word features generated without the dependency parsing and the performance of the model when inputting the addition of the Eojeol tag feature and dependency graph embedding feature. Since dependency parsing is performed on a basic unit of an Eojeol, which is a component of sentences separated by a space, the tag information of the Eojeol can be obtained as a result of the dependency parsing. The Eojeol tag feature means the tag information of the Eojeol. The process of generating the dependency graph embedding consists of the steps of generating the dependency graph from the dependency parsing result and learning the embedding of the graph. From the dependency parsing result, a graph is generated from the Eojeol to the node, the dependency between the Eojeol to the edge, and the Eojeol tag to the node label. In this process, an undirected graph is generated or a directed graph is generated according to whether or not the dependency relation direction is considered. To obtain the embedding of the graph, we used Graph2Vec, which is a method of finding the embedding of the graph by the subgraphs constituting a graph. We can specify the maximum path length between nodes in the process of finding subgraphs of a graph. If the maximum path length between nodes is 1, graph embedding is generated only by direct dependency between Eojeol, and graph embedding is generated including indirect dependencies as the maximum path length between nodes becomes larger. In the experiment, the maximum path length between nodes is adjusted differently from 1 to 3 depending on whether direction of dependency is considered or not, and the performance of answer extraction is measured. Experimental results show that both Eojeol tag feature and dependency graph embedding feature improve the performance of answer extraction. In particular, considering the direction of the dependency relation and extracting the dependency graph generated with the maximum path length of 1 in the subgraph extraction process in Graph2Vec as the input of the model, the highest answer extraction performance was shown. As a result of these experiments, we concluded that it is better to take into account the direction of dependence and to consider only the direct connection rather than the indirect dependence between the words. The significance of this study is as follows. First, we improved the performance of answer extraction by adding features using dependency parsing results, taking into account the characteristics of Korean, which is free of word order structure and omission of sentence components. Second, we generated feature of dependency parsing result by learning - based graph embedding method without defining the pattern of dependency between Eojeol. Future research directions are as follows. In this study, the features generated as a result of the dependency parsing are applied only to the answer extraction model in order to grasp the meaning. However, in the future, if the performance is confirmed by applying the features to various natural language processing models such as sentiment analysis or name entity recognition, the validity of the features can be verified more accurately.
The purpose of this study is to suggest the basic information for diverse and balanced research and development in this field with understanding research trends related to safety education in home economics. In order to so, this study makes population and sampling by targeting cases which refer to 'safety' on 15 papers of academic journals related to home economics registered in the National Research Foundation from 2001 to 2015, 244 papers related to safety education area and 179 master doctorate thesis by searching keyword as 'safety'. Analysis contents are research trends of papers related to safety education by year and by subject and research trends of safety education by area and by research method. As a result of the study, first, the number of research papers related to safety education by year on home economics curriculum repeated increase and decrease and there have been consistent studies conducted on safety education with 14-52 papers per every year and yearly average 28.2 papers. On the other hand, the most number of studies conducted in 2015 with 52 papers which are twice as much of 26 papers in 2014. This seems to be affected by the announcement of safety comprehensive countermeasures from government and the emphasis of safety subject on 2015 curriculum revision of the Ministry of Education. Second, with regards to research trends by topic, 137 papers are related to safety education (29%), 336 papers are related to safety actual condition (71%). Accidents and recognition had a greater percentage in a paper before 2009 (74.4%) and studies are increased after 2009 (from 21 papers to 53 papers) in terms of development or evaluation of safety education program, development of education materials, development of education method etc. Subject area dealt with the most on the research of safety actual condition is regarding safety accidents or effective variables (23.2%). Subject regarding the variables are researches related to factors influencing family violence, internet addiction, spouse violence, willingness to purchase unsafe food, age harassment, or suicidal attempt etc. Next, researches related to safety recognition (13.9%), safety knowledge and attitude (7.4%), safety behaviors (6.3%), safety consciousness (2.3%) show in sequence. Subject area dealt with the most on the researches regarding safety education is development and evaluation of safety education program (11%) and this appears the most in 2015 by year (21.5%). Third, with regards to eight areas of safety education, there are 143 papers regarding public safety (33.8%), 106 papers regarding violence and personal safety (25.1%), 93 papers regarding general subject on safety or whole safety area (22%) and 58 papers regarding drug and internet addiction (13.7%) in sequence. And there is no paper related to first aid and 1 paper is related to occupational safety (0.2%). Occupational safety area is less researched nevertheless its included in home economic curriculum as relative chapter. First aid does not directly correlate with home economics curriculum but should be studied in preparation for accident which could happen in practical class. Forth, with regards to research trends by research method, quantitative research (89.1%) is mostly used and both research study (70.4%) and experimental research (18.7%) are used the most frequently. In particular, researches on the actual condition of safety education and experimental studies for effectiveness verification take most of research method. As qualitative studies, there are phenomenological study (3.1%) and case study (3.1%) related to actual conditions of safety accidents. 10 papers (2.4%) are mixture of quantitative and qualitative research and some research conducted research study and experimental research at the same time (0.9%). With regards to subject of study, human environments (87.5%) are more than physical environments (12.5) and students (48.4%) are more than teachers and school parents (20.6%). As the subject of physical environments, school (6.5%) is the most but home environment is none. As a result of the study, research for the development of evaluation tool for evaluating safety education, occupational safety and lifelong education should be conducted from this time forward. In addition, the object of study shall be expanded to both human environments in terms of entire life and physical environments for home. An in-depth qualitative research should be needed by observing and meeting with each student.
Until recently, as we recognize the significance of sustainable growth and competitiveness of small-and-medium sized enterprises (SMEs), governmental support for tangible resources such as R&D, manpower, funds, etc. has been mainly provided. However, it is also true that the inefficiency of support systems such as underestimated or redundant support has been raised because there exist conflicting policies in terms of appropriateness, effectiveness and efficiency of business support. From the perspective of the government or a company, we believe that due to limited resources of SMEs technology development and capacity enhancement through collaboration with external sources is the basis for creating competitive advantage for companies, and also emphasize value creation activities for it. This is why value chain network analysis is necessary in order to analyze inter-company deal relationships from a series of value chains and visualize results through establishing knowledge ecosystems at the corporate level. There exist Technology Opportunity Discovery (TOD) system that provides information on relevant products or technology status of companies with patents through retrievals over patent, product, or company name, CRETOP and KISLINE which both allow to view company (financial) information and credit information, but there exists no online system that provides a list of similar (competitive) companies based on the analysis of value chain network or information on potential clients or demanders that can have business deals in future. Therefore, we focus on the "Value Chain Network System (VCNS)", a support partner for planning the corporate business strategy developed and managed by KISTI, and investigate the types of embedded network-based analysis modules, databases (D/Bs) to support them, and how to utilize the system efficiently. Further we explore the function of network visualization in intelligent value chain analysis system which becomes the core information to understand industrial structure ystem and to develop a company's new product development. In order for a company to have the competitive superiority over other companies, it is necessary to identify who are the competitors with patents or products currently being produced, and searching for similar companies or competitors by each type of industry is the key to securing competitiveness in the commercialization of the target company. In addition, transaction information, which becomes business activity between companies, plays an important role in providing information regarding potential customers when both parties enter similar fields together. Identifying a competitor at the enterprise or industry level by using a network map based on such inter-company sales information can be implemented as a core module of value chain analysis. The Value Chain Network System (VCNS) combines the concepts of value chain and industrial structure analysis with corporate information simply collected to date, so that it can grasp not only the market competition situation of individual companies but also the value chain relationship of a specific industry. Especially, it can be useful as an information analysis tool at the corporate level such as identification of industry structure, identification of competitor trends, analysis of competitors, locating suppliers (sellers) and demanders (buyers), industry trends by item, finding promising items, finding new entrants, finding core companies and items by value chain, and recognizing the patents with corresponding companies, etc. In addition, based on the objectivity and reliability of the analysis results from transaction deals information and financial data, it is expected that value chain network system will be utilized for various purposes such as information support for business evaluation, R&D decision support and mid-term or short-term demand forecasting, in particular to more than 15,000 member companies in Korea, employees in R&D service sectors government-funded research institutes and public organizations. In order to strengthen business competitiveness of companies, technology, patent and market information have been provided so far mainly by government agencies and private research-and-development service companies. This service has been presented in frames of patent analysis (mainly for rating, quantitative analysis) or market analysis (for market prediction and demand forecasting based on market reports). However, there was a limitation to solving the lack of information, which is one of the difficulties that firms in Korea often face in the stage of commercialization. In particular, it is much more difficult to obtain information about competitors and potential candidates. In this study, the real-time value chain analysis and visualization service module based on the proposed network map and the data in hands is compared with the expected market share, estimated sales volume, contact information (which implies potential suppliers for raw material / parts, and potential demanders for complete products / modules). In future research, we intend to carry out the in-depth research for further investigating the indices of competitive factors through participation of research subjects and newly developing competitive indices for competitors or substitute items, and to additively promoting with data mining techniques and algorithms for improving the performance of VCNS.