Accelerating Innovation Through Analogy Mining 通过类比想法挖掘加速创新 KDD2017 Best Paper 覃山 SA18006180
通过类比想法挖掘加速创新 KDD2017 Best Paper Accelerating Innovation Through Analogy Mining 覃山 SA18006180
目录1.背景介绍 2.实验方法 3.算法评估 4.总结
1.背景介绍 2.实验方法 3.算法评估 4.总结 目录
刘 /01 背景介绍
背景介绍
类比发明示例 ● Salvador Luria(1969年诺贝尔生物医学奖)：受老虎机的中奖模式的启发， 发现细菌的突变理论。 ·Chrysippus:水波与声波。 ·一位汽车修理工：从酒瓶中取出木塞的视频->分娩辅助设备。 可以挽救数百万人的生命，尤其是在发展中国家 https://www.nytimes.com/2013/11/14/health/new-tool-to-ease-difficult-births-a-plastic-bag.html
4 类比发明示例 • Salvador Luria(1969年诺贝尔生物医学奖):受老虎机的中奖模式的启发， 发现细菌的突变理论。 • Chrysippus：水波与声波。 • 一位汽车修理工：从酒瓶中取出木塞的视频->分娩辅助设备。 • 可以挽救数百万人的生命，尤其是在发展中国家 https://www.nytimes.com/2013/11/14/health/new-tool-to-ease-difficult-births-a-plastic-bag.html
海量在线数据 ·美国专利数据库拥有1976年至今发布的900多万项专利的全文。 ·InnoCentive包含超过40,000个商业，社交，政策，科学和技术问题和解决方 案。 ·Quirky是一家帮助工程师进行开发的公司，已经提交了200多万件产品创意。 OpenIDEO为各种社会问题提供了数百种解决方案。 ·在Google?学术中可以搜索到百万篇科学论文和法律案例。 目标：自动发现大型非结构化数据集中的类比想法 5
5 海量在线数据 • 美国专利数据库拥有1976年至今发布的900多万项专利的全文。 • InnoCentive包含超过40,000个商业，社交，政策，科学和技术问题和解决方 案。 • Quirky是一家帮助工程师进行开发的公司，已经提交了200多万件产品创意。 • OpenIDEO为各种社会问题提供了数百种解决方案。 • 在Google学术中可以搜索到百万篇科学论文和法律案例。 目标：自动发现大型非结构化数据集中的类比想法
挑战 ·对人类而言 ·工作量大，认知负担沉重 。 对表面相似性非常敏感，倾向于表面属性上的相似，而不是内在关系、结 构的相似 ·对机器而言 ·大多算法集中在完全结构化的数据集 ·NLP/IR方法，例如word2vec、LSA、LDA等模型，在检测表面相似性方面 表现出色，但往往无法检测出单词分布不一致的文档之间的相似性。 6
6 挑战 • 对人类而言 • 工作量大，认知负担沉重 • 对表面相似性非常敏感，倾向于表面属性上的相似，而不是内在关系、结 构的相似 • 对机器而言 • 大多算法集中在完全结构化的数据集 • NLP/IR方法，例如word2vec、LSA、LDA等模型，在检测表面相似性方面 表现出色，但往往无法检测出单词分布不一致的文档之间的相似性
刘 /02 实验方法
实验方法
弱结构化表示 ·找到一个弱表示 ·对于类比想法的挖掘表示出足够的信息 ·可以被机器所学习 结构化数据 自然语言 ENABLETENABLE(ENABLE( Thirsty too SMALLER THAN[UNINFLATED(plastic_bag),bottle_opening), Pet water bowl/dispenser for your vehicle cup holder. INSERT INTO(person,plastic bag,bottle]), Over spill lip to catch water CAUSEICAUSE( Has optional sleeve for larger cup holders INSER INTO(person,air,plastic_bag). Optional floor base EXPAND[air,plastic_bag]) One way valve so water cant over flow from bottle ENCLOSE(plastic bag,cork))). Small reservoir CAUSEI Reservoir acts as backsplash PULL(person,plastic_bag]. Water bottle attachment ~IN{cork,bottle])) Holds water in your vehicle cupholder for pet Foldable handle to get unit out of holder Dishwasher safe Optional sleeve for larger cup holders 8
8 弱结构化表示 • 找到一个弱表示 • 对于类比想法的挖掘表示出足够的信息 • 可以被机器所学习 Thirsty too Pet water bowl/dispenser for your vehicle cup holder. Over spill lip to catch water Has optional sleeve for larger cup holders Optional floor base One way valve so water cant over flow from bottle Small reservoir Reservoir acts as backsplash Water bottle attachment Holds water in your vehicle cupholder for pet Foldable handle to get unit out of holder Dishwasher safe Optional sleeve for larger cup holders ENABLE[ENABLE(ENABLE( SMALLER_THAN[UNINFLATED(plastic_bag),bottle_opening), INSERT_INTO(person,plastic_bag,bottle]), CAUSE[CAUSE( INSER_INTO(person,air,plastic_bag), EXPAND[air,plastic_bag]) ENCLOSE(plastic_bag,cork))), CAUSE[ PULL(person,plastic_bag], ~IN{cork,bottle])) 结构化数据 ? 自然语言
用途和机制 ·认知心理学认为类比过程的核心是用途(Purpose)和机制(Mechanism)。 类比问题则归纳为 ·用途相同，机制不同 ·机制相同，用途不同 What is the product good for? How does the product work? *Amazing Pillow* *Amazing Pillow "A pillow combined with alarm clock,bluetooth,sensors and more features to improve and monitor sleep A pillow combined with alarm clock,bluetooth,sensors and more features to improve and monitor sleep. Wake up comfortably with built in alarm clock *Wake up comfortably with built in alarm clock Track sleep patterns Track sleep patterns Built in blind fold with led lights and sensors Built in blind fold with led lights and sensors Full support for any kind of sleeper "Full support for any kind of sleeper "Alarm includes led lighting,vibrations and built in headphones Alarm includes led lighting,vibrations and built in headphones for comfort for comfort
9 用途和机制 • 认知心理学认为类比过程的核心是用途(Purpose)和机制(Mechanism)。 • 类比问题归纳为 • 用途相同，机制不同 • 机制相同，用途不同
关于匹配的标注方法 ·查找界面 ·找出给定产品间的类比 记录行为痕迹 ·正例：被标记为匹配； 负例：被隐式拒绝的文档，并不一定是真负例。 Target document Search for matches 1 selected match(es) No good matches.Give me a new document Q control Submit match(es) Pattern:provide convenient interface to 487 results Not match Match external services BY integrated controls on wearable device Not match Match Connected Dehumidifier A connected dehumdifier to your home to set speeds Wall Awl tur on and off and to receive notifications to your app. A simple hand tool to produce a precise size tapered hole Easy to change out or empty water Pandora/Spotify Headphones needed for plastic wall anchors. Mildew growth sensor inside the collection bucket. Wired headphones with mic and buttons that can be used Alternative to stopper disks:sized colored set of Wall Awls Notifies owner when it needs to 'thumbs up'or'thumbs down'a Pandora/Spotify each with collar Be on wheels so it's easy to move around. song Colour coded disks with centre hoie slide on tapered tool to Sends you a message when it's full. Wifi/4g capability to stream songs Sync with smart thermostat (Norm,Nest)to sync temp control hole size A mic to listen for commands.EX'play song name' Hexagonal Shank fits any %driver/handle and humidity control Bluethooth capability so it will work across all OS A4-sided tapered tool for punching a hole in Drywall Stainless steel look ·d go wireless Pointed tip allows precise placement at desired spot on wall Waming light to indicate bucket max level of fullness help curate better playlists easier and auto shut off feature easy to use hand tool for punching precise anchor holes in Saves having to pull your phone out of your pocket drywall Long tray hose to run to drain so can run longer is high humidity situations. be notified of effectiveness and rate of de Recent search queries Not match Match control 488 stream G 10
10 关于匹配的标注方法 • 查找界面 • 找出给定产品间的类比 • 记录行为痕迹 • 正例：被标记为匹配；负例：被隐式拒绝的文档，并不一定是真负例