카테고리 협업 필터링 기반의 모바일 광고 추천 시스템 설계

Abstract

국내 모바일 광고 시장규모는 지난 해 2억 1천만 달러를 기록했고, 올해 3억 3천 달러까지 성장할 것이라고 한다.(2011년 방송통신위원회 자료) 또한, 국내 스마트폰 가입자 수가 2,000만 명을 넘어서면서 광고 미디어로서 기본적으로 가져야 하는 폭넓은 사용자 확보는 물론 낮은 비용으로 광고를 할 수 있어 그 가치가 더욱 높게 평가 되고 있다. 모바일 광고 시장 규모가 커짐에 따라 단순한 배너 광고나 키워드 광고를 벗어나서 광고의 성과를 높이기 위해 사용자 선호도가 높은 광고를 추천하는 다양한 방법들이 연구 개발되었다. 하지만, 모바일 디바이스인 스마트폰은 그 특성 상 개인정보를 알 수 없고, 네트워크상의 기기 고유번호로만 식별이 가능하기 때문에 인구통계학상(나이, 성별 등)의 추천 시스템은 사용할 수가 없다. 또한, 개인정보를 추가적으로 수집한다 하더라도 특정 어플리케이션에 한정 되어지기 때문에 사용이 제한적 일 수밖에 없다. 위치기반서비스(Location based service)의 광고 시스템 역시 위치를 기반으로 한 광고를 추천 할 수는 있지만, 사용자의 선호도는 예측할 수 없다. 본 논문에서는 개인 정보나 컨텐츠의 내용 등에 상관없이 선호도 정보만으로 정확도 높은 추천이 가능한 카테고리 협업 필터링 기반의 모바일 광고 추천 시스템을 제안한다. |The market size of mobile advertisement has reached $210 million in the past year, and is estimated to reach $300,003,000 this year(Korea Communications Commission, 2011). Since user coverage as an advertisement media has been reached - domestic smartphone subscribers having surpassed 20 million members - and costs are relatively low, the merits of this advertisement method are being highly valued. Following the growth of advertisement market size and advertisers no longer rely on simple banner or keyword advertisement, many new methods for recommending advertisement based on user preferences which increases the effectiveness of advertisement are being researched. But since personal information can’t be obtained from mobile devices such as smartphones and only the unique machine number on the network can be identified, recommendation systems based on demographics can’t be applied. Even if additional personal information is obtained, that data is limited to specific applications, so its usage is limited. Location based advertisement services can also be used for recommending location specific advertisement, but can’t be used to predict user preferences. This thesis proposes a mobile advertisement recommend system based on category collaborative filtering which can provide accurate recommendations using only preferences and not reliant on personal information or content.; The market size of mobile advertisement has reached $210 million in the past year, and is estimated to reach $300,003,000 this year(Korea Communications Commission, 2011). Since user coverage as an advertisement media has been reached - domestic smartphone subscribers having surpassed 20 million members - and costs are relatively low, the merits of this advertisement method are being highly valued. Following the growth of advertisement market size and advertisers no longer rely on simple banner or keyword advertisement, many new methods for recommending advertisement based on user preferences which increases the effectiveness of advertisement are being researched. But since personal information can’t be obtained from mobile devices such as smartphones and only the unique machine number on the network can be identified, recommendation systems based on demographics can’t be applied. Even if additional personal information is obtained, that data is limited to specific applications, so its usage is limited. Location based advertisement services can also be used for recommending location specific advertisement, but can’t be used to predict user preferences. This thesis proposes a mobile advertisement recommend system based on category collaborative filtering which can provide accurate recommendations using only preferences and not reliant on personal information or content.