Application cases

Application cases

AI Cloud Service

Linkage between KT GIGA Genie and ADAMS.ai

‘AI Speaker’ which allows people to experience AI technologies easily in everyday life has been introduced throughout the world. Beginning with ‘Echo’ introduced by Amazon based on its AI secretary Alexa in 2014, the relevant market is growing rapidly throughout the world.

In our country, three mobile carriers, including KT, SK Telecom and LG Uplus are leading the AI speaker market and this market is growing rapidly although the market is a latecomer in comparison to markets in major countries that are led by global Internet companies, countries such as North America. Among these mobile carriers, KT is the No. 1 company in the domestic AI speaker market with 1.5 million subscribers. After the introduction of AI speaker ‘GIGA Genie’ in January 2017, KT is diversifying its product lines into GIGA Genie LTE, GIGA Genie Buddy, GIGA Genie 2 and AI Makers Kit.

GIGA Genie operates in close linkage with each home service platform (IPTV, music, phone call, home IoT) of KT and is linked with the third parties’ platforms, composing an overall AI ecosystem. Among the third parties’ platforms, GIGA Genie is linked with Saltlux’s ADAMS.ai platform for advancing dialog quality between a user and AI, and GIGA Genie provides a lifelike Deep Q &A service that requires extensive knowledge within such a linkage.

ai01
cloud1

< Linkage between KT ‘GIGA Genie’ and ADAMS.ai >

① Main Content
  1. GIGA Genie provides a Q&A service that finds and delivers answers to users’ questions. GIGA Genie is linked with Saltlux’s ADAMS.ai for Deep Q &A in fields which include real-life information, general knowledge and expert knowledge that requires extensive background knowledge. When a user’s question submitted through GIGA Genie is delivered to the ADAMS.ai platform, ADAMS.ai finds an answer to the question through large-scale knowledge graph-based Deep Q &A technology. The answer found in such a way is delivered to GIGA Genie and the requested answer is provided to the user. In such a process, new knowledge vital to Q &A is added and learned continuously to improve the quality of Q &A.
② Applied technology
  1. Flexible Q&A using the ensemble technology
    KT ‘GIGA Genie’ finds answers to equations in fields which include real life information, general knowledge and expert knowledge using the Deep Q &A service provided by ADAMS.ai through the Open API. The Deep Q &A service provides answers to submitted questions to users by forming an ensemble of knowledge graph-based Q &A (KBQA), information search-based Q &A (IRQA), MRC-based Q &A (MRQA) and selecting the optimum solution according to the nature of each particular query.
  2. Continuous Q &A monitoring and knowledge curation
    Continuous Q &A monitoring across fields which include real life information, general knowledge and expert knowledge where Deep Q &A is applied is supported, and large-scale new knowledge in each field is added and established every day through knowledge curation. The established new knowledge is converted into data that can be used for Q &A easily through the management function of the platform.
③ Main performance
  1. It provides a complex reasoning speed of 500,000 knowledge units per second utilizing general knowledge of some 800 million items established as a knowledge base. It also delivers a 94% probability of choosing the correct answer in real-life information, general knowledge, and expert knowledge Q&A through its strength in complex reasoning and knowledge learning from 5 million documents, which is the quantity of data collected throughout each and every day.
Linkage between Woori Bank AI consultation system and ADAMS.ai

Recently, various customer service channels, including mobiles, have been expanded in financial fields and there is a growing demand for a service responding to latest AI-based technology trends. Woori Bank also saw the need to establish an AI consultation system in response to a paradigm shift in financial services for customers and employees.

ảnh 3

< AI Consultation System in Woori Bank >

Woori Bank initiated the ChatBot service ‘WebeeBot’ that enables real-time consultation with a customer using AI technologies in September 2017. ‘WebeeBot’ provides financial consultation services in real time 24 hours a day. ‘WebeeBot’ provides an answer in the form of assessing a questioner’s intent and chatting with the customer just as a consultant would, not the simple response selection activity of the previous question and answer process. It also provides general information besides financial information.

① Main Content
  1. Woori Bank’s AI Consultation System can be divided into the data area, AI platform area, and service area, and is linked with Saltlux’s ADAMS.ai, a huge knowledge base-based platform, in order to provide general information, such as weather and personal information, through Q &A. Knowledge of consultation services in financial fields is internally established in the data area, and such knowledge is used in consultation services, but in the case of general information that cannot be obtained within a short period of time, it is serviced through a link with an external open platform, ADAMS.ai.
② Applied technology
  1. Knowledge graph-based Deep Q &A
    Woori bank’s ‘WebeeBot’ finds answers to general knowledge questions as well as financial queries using the Deep Q &A service provided by ADAMS.ai through the Open API. The Deep Q &A service provides Q &A regarding current general knowledge through complex reasoning based on Asia’s largest knowledge graph.
  2. Prompt application of new knowledge
    The platform for managing information, such as dictionaries, knowledge, and indexes, is provided to enable Q &A processing by creating, learning and reasoning centering on new general knowledge where Deep Q &A is applied so that new knowledge can be integrated easily and corresponding Q &A can be implemented.
③ Main performance
  1. Woori Bank’s ‘WebeeBot’ is linked with ADAMS.ai that embeds knowledge graphs for general knowledge regarding approximately 800 million items, expanding the service range that can be provided through AI ChatBot significantly from consultation in financial fields. The extensive general information area includes such items as weather and personal information.

Data Science Cloud Service

Status of Data Science Cloud Service
1
2
3
4
Korean Women’s Development Institute – Collection and analysis of social big data regarding policies for women

Due to the development of data-related technologies, the form of research is shifting from the sampling survey, the traditional investigation method of the past, to the big data-based complete survey. This is spreading to all industrial fields, including public institutions, local governments and companies, and various policies for providing big data-based administration services and establishing an evidence-based policy are being promoted. Therefore, it was intended to establish basic data in order to determine the demands for policies for women and prepare countermeasures by analyzing public opinion in a society that changes minute by minute in the field of research for policies for women.

cloud4

< Korean Women’s Development Institute – Analysis process >

① Main Content
  1. This research has been promoted with the purpose of analyzing various social phenomena and issues regarding gender based on social media by collecting and analyzing social data. The words weighted value analysis, emotion analysis and concurrently appeared in word analyses that were internal solutions from Saltlux used for the analysis method. The analysis result using the visualized chart was evaluated as an important research result since it provided basic data for promoting policies for women.
② Applied Technology
  1. Trend Analysis
    The main interests of people were determined and whether discussion was initiated, expanded or reduced was based on the frequency of reference which was concentrated by aggregating the frequency of analysis target references.
  2. Word weighted value and emotion analysis
    Words that appeared frequently in all documents were filtered and important words were extracted using the frequency of word appearance within a document and across documents. Emotions within each sentence were classified using the emotional word dictionary and map machine learning. Based on the findings, the comprehensive emotions of the analysis target or word were determined.
  3. Concurrently appeared in word and original text (VOC) analysis
    Any important words that appeared frequently were sorted and only sentences that included the relevant word were extracted. Original text where keywords were mentioned together was extracted in order to determine correlation between such keywords. Trends and linkage contexts of an analysis target were determined by examining extracted original text.
③ Main Performance
  1. In this research, data from press media, social media and communities related to subjects such as feminism and minimum wage were collected in a comprehensive analysis of public opinion, and limited understanding of a certain sample. Also, basic data necessary for promoting policies for women were prepared by analyzing the collected data. This was a research case that extended research in media that were scarce in the past, and it is expected to promote the future use of big data by discovering the demands for policies for women and seeking countermeasures.
Hyundai Motor Company – Worldwide competing vehicles information collection

The worldwide competing vehicles information collection service was undertaken with the purpose of improving customer satisfaction and reinforcing market competitiveness through the preparation of external data in multiple languages from multiple channels and active market sensing using such data. For such purposes, an exclusive data collection infrastructure was constructed for the comprehensive understanding of customers’ needs, acquisition of information regarding trends in the competitive market, and real-time monitoring and extraction of potential risks from the collected information of competing vehicles in the world. Data were collected from all news, magazine, cafe, community and forum websites related to vehicles as well as Facebook, Twitter, Instagram, YouTube and custom websites requested by the customer. The collected data were standardized through automatic classification, filtering, purification and processing, and established and provided in a data form that could be used immediately by the customer.

cloud5

< Hyundai Motor Company – Data collection and purification service >

① Main Content
  1. This project was carried out for the purpose of improving customer satisfaction and enhancing market competitiveness through the collection of external data using multiple channels and active market sensing using the collected data. For such purposes, an exclusive data collection infrastructure that enabled comprehensive understanding of customers’ needs, acquisition of information regarding the trends of the competitive market and real-time monitoring and extraction of potential risks from the collected information was constructed. The target data for collection include data from vehicle-related detailed message boards in websites, cafe websites, keyword-based data and data from Facebook, Twitter, Instagram, YouTube and other custom websites, and the collected data are filtered, standardized and provided in a form that could be used immediately by the customer.
② Applied Technology
  1. Data collection
    Approximately 1,500 target websites for collection selected by the customer are examined and sorted. The curation center carries out meta data investigation for the target websites for collection and data filtering through consultation with the customer. After extracting search result URLs and URLs in the original text of the collected data, the frequency is analyzed. Additionally, the curator finds a website that directly related to the vehicle.
  2. Automation of data quality
    The functions including data filtering, automatic cafe schema detection, machine learning spam classification, data function standardization, creation of data to transfer and failure detection are carried out.
  3. Collection status monitoring dashboard
    Real-time monitoring of collection status in the project, verification of each project management status, verification of top websites within the degree of collection risk, verification of trends in collected data and verification of the list of data collection places are available.
③ Main performance
  1. Personalized data that could be used immediately for internal analysis were provided through a comprehensive understanding of the customers’ needs, acquisition of information regarding the trends of the competitive market and monitoring of potential risk, extraction of hidden information, such as customer’s behavior patterns, were gained from the collected information.
AIA Group Limited – Determined the needs of singles aged between 35 and 45 regarding insurance products and planned insurance products targeting these singles

It was necessary to improve market competitiveness by planning a product based on data rather than using the previous method which depended on intuitive experience related to the past. It was intended to develop an insurance product that could attract the attention of singles aged between 35 and 45 by securing target analysis data related to single people and carrying out convergence analysis and cognitive analysis of secured structured and unstructured data.

cloud6

< AIA Group Limited – Planning of an insurance product targeting singles aged between 35 and 45 >

① Main Content
  1. This project was a case of big data collection and analysis by age bracket and gender for the purpose of “determining the needs of singles aged between 35 and 45 regarding insurance products and planning an insurance product targeting these singles.” This was a case where insight was gained and used for the establishment of a product strategy by AIA Group Limited through the targeting of singles aged between 35 and 45 and the verification of issues and hypotheses.
② Application Function
  1. Hypothesis verification scenario
    Consumers’ interests and reputation are analyzed by identifying subject keywords and grouping the identified keywords in order to develop an insurance product that can attract the attention of singles aged between 35 and 45.
  2. Analysis of trends of frequency of reference
    Determines the trends and tendency of customers’ interests and determines and examines detailed issues through trend comparisons for each property.
  3. Word Cloud and Keyword Analysis
    Provides insight by analyzing consumers’ interests and reputation for each related issue and identifying the relationships between related keywords.
  4. Sentiment analysis
    Determines consumers’ responses in comparison to the frequency of appearance through emotion analysis result of detailed property keywords for each property and a bubble chart analysis.
③ Main Performance
  1. Customer satisfaction can be improved and market competitiveness can be enhanced through active market sensing using external data from multiple channels. It is possible to secure a verified platform that enables the development of a customized product simply by changing the target.
PwC Consulting – Real-time company information and monitoring service

The basis for a company to promptly respond to rapid social change by enabling the determination of company status in real time was necessary. Especially, the collection of huge social media and SNS data, analysis and visualization of such data are necessary means and goal for gaining insight regarding a specific company.

cloud7

< PwC Consulting – Purpose of company information and monitoring service >

① Main Content
  1. Provides real-time company information and competitors monitoring service based on social data from more than 15 billion cases The analysis functions, including trend analysis, keyword comparison and analysis, related keyword analysis, emotion analysis and dictionary management were provided, and an important result in the company analysis and monitoring was provided based on such functions.
② Application Function
  1. Social Data Search
    Extracts and provides a document that contains a query from each search source of news, blogs and Twitter. Provides the total number of searched documents and the number of documents after removing redundant documents.
  2. Trend Analysis
    Provides the statistics of documents registered in news, blogs and Twitter by period for a given query. The result is provided in the form of json, and the results of document statistics corresponding to the query, and the results of entire document statistics for the same period are provided.
  3. Automatic related subject analysis
    Related subjects are extracted from data from news, blogs and Twitter for a given query and a ranking is provided based on the frequency of appearance, and the connection information is provided. The unit duration can be set for a selected duration.
  4. Today’s Topic Analysis
    N number of today’s topics for an entered news category is extracted and the ranking of such topics is provided.
③ Main Performance
  1. The entire process including the collection, saving, analysis and expression of data regarding a certain company and competitors can be processed integrally. In this project, data related to companies from the press media, social media and communities were collected. A variety of information including the status, technologies and forms of service by a company is collected and insight is provided through the analysis of the collected data, and it can be also used as customized information.
Korea Institute of Science and Technology (KIST) – Data-based R &D environment establishment service

Under the circumstances that the percentage of digital data produced is increasing gradually, the introduction of a research support system using big data in a research institute could improve overall productivity of the research institute. In this product, global-scale research data analysis environment that could improve the research productivity of KIST was built by establishing a structured and unstructured data collection and management system in order for KIST to maintain competitiveness in R &D.

cloud8

< Korea Institute of Science and Technology – Information analysis platform >

① Main Content
  1. The system for sharing internal research information in real time and linking with information regarding external (overseas) research papers that helped setting research direction was established in order to assist R &D productivity of the research institute. Such a system consisted of KiRI DataBank that could collect, purify, store and analyze materials, such as domestic and foreign papers or patents that were necessary for catalyst field-related research, and the KiRI Note platform for recording and managing research data.
② Applied Technology
  1. Inquiry for the recording and status of research activities
    Research activities and the status of projects in progress can be managed efficiently through the research node function.
  2. Basic paper search and analysis, similar papers, trends and keyword analysis
    This function searches for and analyzes content related to main keywords. Various analyses, including basic time series analysis, trend comparison, keyword analysis, paper-author, or material-paper network diagrams are available.
  3. Curation tool that can extract paper information in detail
    The basic information of a paper crawled from the web, as well as the curation tool, are provided together. Automatic DB creation is carried out using basic abstract information and detailed and technical contents are structured, stored and managed through the curation process.
  4. Digitizing that extracts the analysis result of a paper
    The digitizing tool that can convert the research result in a graph included in a paper into data is provided. When you select an image you wish to extract, set the axis and target. You can convert a graph image into digital data and the result is stored and managed in the storage for each paper.
③ Main Performance
  1. Research support was enabled through the sharing of research information and the qualitative improvement of research productivity and communication, and the latest information sensing and paper use in the catalyst field were improved by establishing DB using information of domestic and overseas research papers and patents. The experiment notes created through various experiments are converted into data and applied to data science, such as data mining and AI.