åºäºä¼ ç»ç¦å·è¡¨ç¤ºçç¥è¯åºé®ç

Size: px

Start display at page:

Download "åºäºä¼ ç»ç¦å·è¡¨ç¤ºçç¥è¯åºé®ç"

碑射容
5 years ago
Views:

1 基于传统符号表示的知识库问答冯岩松北京大学 October 19, 2016 Feng et al. (PKU) Question Answering October 19, / 40

2 知识库问答任务利用知识库回答自然语言问题 Feng et al. (PKU) Question Answering October 19, / 40

3 知识库问答任务利用知识库回答自然语言问题输入 : 自然语言问句资源 : 结构化知识库输出 : 答案 Feng et al. (PKU) Question Answering October 19, / 40

4 知识库问答任务利用知识库回答自然语言问题输入 : 自然语言问句资源 : 结构化知识库, 表格, 结构化 / 半结构化记录, 输出 : 答案 Feng et al. (PKU) Question Answering October 19, / 40

5 知识库问答任务利用知识库回答自然语言问题 Feng et al. (PKU) Question Answering October 19, / 40

任务框架 What else did the director of the movie Interstellar direct? select?y [ ] fb:m.0fkf28 fb:object.type fb:film.film fb:m.0fkf28 fb:film.film.directed_by?x?x fb:film.

6 任务框架 What else did the director of the movie Interstellar direct? select?y [ ] fb:m.0fkf28 fb:object.type fb:film.film fb:m.0fkf28 fb:film.film.directed_by?x?x fb:film.director.fim?y?y fb:object.type fb:film.film Inception, The Dark Knight Rises The Dark Knight Batman Begins.. Feng et al. (PKU) Question Answering October 19, / 40

7 任务框架 What else did the director of the movie Interstellar direct? select?y [ ] fb:m.0fkf28 fb:object.type fb:film.film fb:m.0fkf28 fb:film.film.directed_by?x?x fb:film.director.fim?y?y fb:object.type fb:film.film Inception, The Dark Knight Rises The Dark Knight Batman Begins.. Feng et al. (PKU) Question Answering October 19, / 40

8 早期工作 Baseball : 有关棒球比赛的问答 [Green et al., 1961] How many games did the Yankees play in July? LUNAR: 有关科研数据的问答 [Woods, 1993] How many samples contain Titanium? 订机票地理知识, 找工作 Feng et al. (PKU) Question Answering October 19, / 40

9 早期工作 Baseball : 有关棒球比赛的问答 [Green et al., 1961] How many games did the Yankees play in July? LUNAR: 有关科研数据的问答 [Woods, 1993] How many samples contain Titanium? 订机票地理知识, 找工作一些问题面向特定领域解答所需的知识库结构较为简单 : 数百个实体关系知识库规模有限 : 数千个三元组大多可以采用人工设计的模板或规则来解析问题难以应付开放域问题 : 更多的实体关系 ; 更多的三元组 Feng et al. (PKU) Question Answering October 19, / 40

10 大规模知识库 Key 更多实体更多关系 Freebase DBpedia Feng et al. (PKU) Question Answering October 19, / 40

11 大规模知识库 Key 更多实体更多关系 Freebase Free917, WebQuestion DBpedia QALD Feng et al. (PKU) Question Answering October 19, / 40

12 数据集 Free917 [Cai and Yates, 2013] 使用 Freebase 917 个问题, 标注了逻辑表达式 WebQuestions [Berant et al., 2013] 使用 Freebase 5,810 个问题, 通过 Google Suggest API 爬取利用 Amazon Mechanical Turk 服务得到答案 ( 存在错误 ) 一个问题可能存在多个答案利用 Average F1 评价 WebQuestionsSP: WebQuestions with SPARQL annotations [Yih et al., 2016] QALD 知识库问答评测,QALD:Question Answering over Linked Data 使用 DBpedia 每年 100 个问题左右 Hybrid Track:DBpedia 不足以完整回答问题, 必须依靠文本信息 Simple Questions [Bordes et al., 2015] 108,442 个简单问题, 附带一条 Freebase 三元组作为答案 Feng et al. (PKU) Question Answering October 19, / 40

13 技术挑战 1 如何恰当地表示问题的语义 2 如何将问题的语义表示与知识库进行关联 Feng et al. (PKU) Question Answering October 19, / 40

14 技术挑战 1 如何恰当地表示问题的语义丰富的提问方式, 复杂的提问意图 2 如何将问题的语义表示与知识库进行关联 Feng et al. (PKU) Question Answering October 19, / 40

15 技术挑战 1 如何恰当地表示问题的语义丰富的提问方式, 复杂的提问意图 2 如何将问题的语义表示与知识库进行关联大规模开放域知识库 Feng et al. (PKU) Question Answering October 19, / 40

16 技术挑战 1 如何恰当地表示问题的语义丰富的提问方式, 复杂的提问意图语义分析 2 如何将问题的语义表示与知识库进行关联大规模开放域知识库知识库映射 Feng et al. (PKU) Question Answering October 19, / 40

17 技术挑战 1 如何恰当地表示问题的语义丰富的提问方式, 复杂的提问意图语义分析 2 如何将问题的语义表示与知识库进行关联大规模开放域知识库实体链接关系抽取 Feng et al. (PKU) Question Answering October 19, / 40

18 语义分析语义分析利用形式化方法表示问题语义一步到位两步方法 Feng et al. (PKU) Question Answering October 19, / 40

19 语义分析语义分析利用形式化方法表示问题语义一步到位直接获得与具体知识库相关的语义表示 [Kwiatkowski et al., 2013, Yih and Ma, 2016] Feng et al. (PKU) Question Answering October 19, / 40

20 语义分析语义分析利用形式化方法表示问题语义一步到位两步方法先通用语义表示与再实现知识库映射 Feng et al. (PKU) Question Answering October 19, / 40

21 语义分析语义分析利用形式化方法表示问题语义一步到位两步方法语义表示 λ Calculus: λx.sibling_of (justin_bieber, x) gender(x, female) Lambda Dependency-based Compositional Semantics (λ-dcs) SisterOf.Justin_Bieber 借助于现有句法语义分析技术,PCFG CCG Phrase Dependency Graph Query Graph... Feng et al. (PKU) Question Answering October 19, / 40

22 语义表示 :λ Calculus constants entities, numbers, functions logical connectors,,, quantification, additional quantifiers argmax, argmin,... [Liu, 2015] Feng et al. (PKU) Question Answering October 19, / 40

23 语义表示 :λ-dcs Lambda Dependency-based Compositional Semantics [Liang, 2011] 更简洁易用组成 : 实体关系 Join/Intersection 操作 Feng et al. (PKU) Question Answering October 19, / 40

24 语义表示 :λ-dcs Lambda Dependency-based Compositional Semantics [Liang, 2011] 更简洁易用组成 : 实体关系 Join/Intersection 操作 Feng et al. (PKU) Question Answering October 19, / 40

25 语义表示 :CCG 组合范畴语法 :Combinatory Categorial Grammar [Reddy et al., 2015] Feng et al. (PKU) Question Answering October 19, / 40

26 语义表示 :CCG [Reddy et Feng al., 2015] et al. (PKU) Question Answering October 19, / 40 组合范畴语法 :Combinatory Categorial Grammar

27 语义表示 :Query Graph 借助于知识图谱中的片段, 以及实体链接 / 关系抽取 Who first voiced Meg on Family Guy? = λx. y.cast(familyguy, y) actor(y, x) character(y, MegGriffin) 直接有效 : 疑问词实体关系及一些限制节点组成的图 [Yih et al., 2015] Feng et al. (PKU) Question Answering October 19, / 40

28 语义表示 :Phrase Dependency Graph 1 语义表示需独立于某一具体知识库 [what] else movies] did the director of] [the movie] [Interstellar] [direct] Feng et al. (PKU) Question Answering October 19, / 40

29 语义表示 :Phrase Dependency Graph 1 语义表示需独立于某一具体知识库 [what] else movies] did the director of] [the movie] [Interstellar] [direct] 2 与知识库的映射可以更灵活 [ ] select?y fb:m.0fkf28 fb:object.type fb:film.film fb:m.0fkf28 fb:film.film.directed_by?x?x fb:film.director.fim?y?y fb:object.type fb:film.film [ ] select?y ns:interstellar dbo:type dbo:film ns:interstellar?y?y dbp:director?x dbp:director?x dbo:type dbo:film [Xu et al., 2014] Feng et al. (PKU) Question Answering October 19, / 40

30 Framework what else movies did the director of the movie Interstellar direct Feng et al. (PKU) Question Answering October 19, / 40

31 Framework what else movies did the director of the movie Interstellar direct Parsing Feng et al. (PKU) Question Answering October 19, / 40

32 Framework what else movies did the director of the movie Interstellar direct Parsing [what] else movies] did the director of] [the movie] [Interstellar] variable category relation category entity [direct] relation Feng et al. (PKU) Question Answering October 19, / 40

33 Framework what else movies did the director of the movie Interstellar direct Parsing [what] else movies] did the director of] [the movie] [Interstellar] variable category relation category entity [direct] relation Instantiation Feng et al. (PKU) Question Answering October 19, / 40

34 Framework what else movies did the director of the movie Interstellar direct Parsing [what] else movies] did the director of] [the movie] [Interstellar] variable category relation category entity [direct] relation Instantiation [ ] select?y fb:m.0fkf28 fb:object.type fb:film.film fb:m.0fkf28 fb:film.film.directed_by?x?x fb:film.director.fim?y Feng et al. (PKU)?y Question fb:object.type Answering fb:film.film October 19, / 40

35 Phrase Dependency Graph [what] else movies] did the director of] [the movie] [Interstellar] variable category relation category entity [direct] relation Node 每个短语都具有语义标签 l {entity, category, variable, relation} Edge 短语之间的谓词论元结构 unary predicate binary predicate Feng et al. (PKU) Question Answering October 19, / 40

36 结构预测问题输入 : 自然语言问题输出 : Phrase Dependency Graph 流水线框架 1 Phrase Detection what else movies did the director of the movie Interstellar variable category relation category entity direct relation 2 Phrase Dependency Parsing [what] else movies] did the director of] [the movie] [Interstellar] variable category relation category entity [direct] relation Feng et al. (PKU) Question Answering October 19, / 40

37 Phrase Detection what else movies did the director of the movie Interstellar direct V-B O C-B O R-B R-I R-I C-B C-I E-B R-B Sequence labeling problem Structured perceptron with lexical features Feng et al. (PKU) Question Answering October 19, / 40

38 Phrase Dependency Parsing [what] else movies] did the director of] [the movie] [Interstellar] variable category relation category entity [direct] relation Transition-based parsing A queue of incoming phrases A stack of processed phrases Four actions (ArcLeft, ArcRight, Shift, Reduce) Multiple heads Feng et al. (PKU) Question Answering October 19, / 40

39 Parsing Example Queue [what] else movies] did the director of] [the movie] [Interstellar] variable category relation category entity [direct] relation Current Dependency Graph Stack Feng et al. (PKU) Question Answering October 19, / 40

40 Parsing Example Queue Shift [what] else movies] did the director of] [the movie] [Interstellar] variable category relation category entity [direct] relation Current Dependency Graph Stack Feng et al. (PKU) Question Answering October 19, / 40

41 Parsing Example Queue else movies] did the director of] [the movie] [Interstellar] category relation category entity [direct] relation [what] variable Current Dependency Graph Stack Feng et al. (PKU) Question Answering October 19, / 40

42 Parsing Example ArcLeft Queue [direct] relation [what] variable Current Dependency Graph [what] variable movies] category the director of] [the movie] [Interstellar] relation category entity [direct] relation Stack Feng et al. (PKU) Question Answering October 19, / 40

43 Parsing Example Shift Queue [direct] relation [what] variable Current Dependency Graph Stack [what] else movies] did the director of] [the movie] [Interstellar] variable category relation category entity [direct] relation Feng et al. (PKU) Question Answering October 19, / 40

44 Parsing Example Queue [direct] relation [what] variable Current Dependency Graph Stack [what] else movies] did the director of] [the movie] [Interstellar] variable category relation category entity [direct] relation Feng et al. (PKU) Question Answering October 19, / 40

45 解码 ArcRight shift Reduce ArcLeft incremental processing z = arg max w f (S, Q, a) a A where A = {Shift, Reduce, ArcLeft, ArcRight} 特征词汇特征结构特征语义特征 Feng et al. (PKU) Question Answering October 19, / 40

46 与知识库的关联 Qind select?y [ ] Interstellar Interstellar type the director of movies?x?x direct?y?y type movies ns:interstellar dbo:type dbo:film dbp:director?x Qd [ ] select?y ns:interstellar?y?y dbp:director?x dbo:type dbo:film 与知识库中的实体链接面向知识库的概念匹配面向知识库的关系抽取 Feng et al. (PKU) Question Answering October 19, / 40

47 与知识库关联关键实体链接实体链接费城 = 费城 ( 城市 ) 费城 ( 电影 ) 费城 ( 街道 ),... 问题通常较短缺乏足够的上下文命名实体边界模糊,... Feng et al. (PKU) Question Answering October 19, / 40

48 与知识库关联关键实体链接实体链接费城 = 费城 ( 城市 ) 费城 ( 电影 ) 费城 ( 街道 ),... 问题通常较短缺乏足够的上下文命名实体边界模糊,... 需充分利用现有实体链接工具资源 (Heng Ji s EDL tools collection) Feng et al. (PKU) Question Answering October 19, / 40

49 与知识库关联关键实体链接, 关系抽取实体链接费城 = 费城 ( 城市 ) 费城 ( 电影 ) 费城 ( 街道 ),... 问题通常较短缺乏足够的上下文命名实体边界模糊,... 需充分利用现有实体链接工具资源 (Heng Ji s EDL tools collection) 关系抽取 / 匹配缺乏足够的上下文表述方式灵活多样候选关系众多与知识库表述不匹配易受错误传递影响,... 复合关系表示, 如 Freebase 中的 CVT(compound value type) 节点球队 - 球员 - 球衣号 - 场上位置, 夫妻婚姻存续期,... 多管齐下 : 关系抽取通过实体猜测联合消解,... Feng et al. (PKU) Question Answering October 19, / 40

50 联合消解歧义将子任务的中间结果暂存, 多个子任务联合寻找最优结果克服流水线框架中错误传递 [Xu et al., 2016] Feng et al. (PKU) Question Answering October 19, / 40

51 联合消解歧义将子任务的中间结果暂存, 多个子任务联合寻找最优结果克服流水线框架中错误传递短语切分与短语依存结构抽取之间的联合消解 top-1 how many stores are in nittany mall variable relation relation entity top-2 how many stores are in nittany variable category relation entity... mall [Xu et al., 2016] Feng et al. (PKU) Question Answering October 19, / 40

52 联合消解歧义将子任务的中间结果暂存, 多个子任务联合寻找最优结果克服流水线框架中错误传递实体链接与关系抽取之间的联合消解 [what] else movies] did the director of] [the movie] [Interstellar] [direct] fb:time.month fb:time.event fb:film.film... fb:sports.sports_team.cha mpionships fb:film.direct_by... fb:time.month fb:time.event fb:film.film... fb:m.04p4lvx fb:m.04p3_v_ fb:m.01yz0x... fb:sports.sports_team.cha mpionships fb:director.film... [Xu et al., 2016] Feng et al. (PKU) Question Answering October 19, / 40

53 结构化知识库之外 : 文本的作用单独依靠知识库也许是不够的难以显示的依靠知识库解析 :Who did Shaq first play for? 偏主观的表述方式 : What is the most popular movie in the past 5 years in UK? [Zhang et al., 2015, Xu et al., 2016a, Xu et al., 2016b] Feng et al. (PKU) Question Answering October 19, / 40

54 结构化知识库之外 : 文本的作用单独依靠知识库也许是不够的 = 实体与关系的联合消解 who did shaq first play for KB-QA Entity Linking Relation Extraction shaq: m.012xdf shaq: m.05n7bp shaq: m.06_ttvh sports.pro_athlete.teams..sports.sports_team_roster.team basketball.player.statistics..basketball.player_stats.team Joint Inference m.012xdf sports.pro_athlete.teams..sports.sports_team_roster.team Los Angeles Lakers, Boston Celtics, Orlando Magic, Freebase Miami Heat Feng et al. (PKU) Question Answering October 19, / 40

55 结构化知识库之外 : 文本的作用单独依靠知识库也许是不够的 = 利用维基正文清洗候选答案 Wikipedia Dump Los Angeles Lakers, Boston Celtics, Orlando Magic, Miami Heat Freebase Answer Refinement Los Angeles Lakers Boston Celtics Orlando Magic Shaquille O'Neal Shaquille O'Neal Shaquille O'Neal O'Neal signed as a free agent with the Los Angeles Lakers O'Neal played for the Boston Celtics in the season before retiring O'Neal was drafted by the Orlando Magic with the first overall pick in the 1992 NBA draft O Neal signed as a free agent with the Los Angeles Lakers O Neal played for the Boston Celtics in the season before retiring Refinement Model O Neal was drafted by the Orlando Magic with the first overall pick in the 1992 NBA draft Orlando Magic Feng et al. (PKU) Question Answering October 19, / 40

56 在 WebQuestions 上 Question & Answers 1. what is the largest nation in europe Before: Kazakhstan, Turkey, Russia,... After: Russia 2. which country in europe has the largest land area Before: Georgia, France, Russia,... After: Russian Empire, Russia 3. what year did ray allen join the nba Before: 2007, 2003, 1996, 1993, 2012 After: who is emma stone father Before: Jeff Stone, Krista Stone After: Jeff Stone 5. where did john steinbeck go to college Before: Salinas High School, Stanford University After: Stanford University Feng et al. (PKU) Question Answering October 19, / 40

57 WebQuestions 上的结果 Method average F 1 Berant et al. (2013) 35.7 Yao et al. (2014) 33.0 Xu et al. (2014) 39.1 Berant et al. (2014) 39.9 Bao et al. (2014) 37.5 Dong et al. (2015) 40.8 Yao et al. (2015) 44.3 Bast et al. (2015) 49.4 Berant et al. (2015) 49.7 Reddy et al. (2016) 50.3 Yih et al. (2015) 52.5 EL/RE (KB) 44.1 EL/RE + Joint 47.1 EL/RE + Refine 47.0 EL/RE + Joint + Refine 53.3 Feng et al. (PKU) Question Answering October 19, / 40

58 Hybrid-QA: 基于混合资源的知识库问答采用疑问词实体关系表示问题语义同时考虑结构化知识库与维基正文来抽取规范化谓词和开放式关系实体链接利用实体关系抽取技术抽取规范化的知识库谓词关系 :X associatedband Y 利用开放信息抽取技术抽取开放式的文本关系 :X is written by Y 利用整数线性规划来选择最佳组合, 构成知识库查询 [Xu et al., 2016b] Feng et al. (PKU) Question Answering October 19, / 40

Hybrid-QA who is the front man of the band that wrote Coffee & TV Question Decomposition < ans, is the front man of, var1 > < var1, is a, band > < var1, wrote, Coffee & TV > < var1, wrote, Coffee &

59 Hybrid-QA who is the front man of the band that wrote Coffee & TV Question Decomposition < ans, is the front man of, var1 > < var1, is a, band > < var1, wrote, Coffee & TV > < var1, wrote, Coffee & TV > Entity Linking Coffee & TV Triple Triple Solver Triple Solver wrote is the front man of Textual Relation Extraction DBpedia Lookup Bitter_Coffee_(Iranian_video _series) Irish_Coffee_(TV_series) Coffee & TV KB-based Relation Extraction DBpedia Freebase Wikipedia Dump Open Information Extractor Textual KB Multi-Channel Neural Network Paraphrase Model influencedby associatedmusicalartist associatedband writer front man of is written by lead vocalist of Joint Inference Damon Albarn Feng et al. (PKU) Question Answering October 19, / 40

60 WebQuestions 上的结果 Method average F 1 Berant et al. (2013) 35.7 Yao et al. (2014) 33.0 Xu et al. (2014) 39.1 Berant et al. (2014) 39.9 Bao et al. (2014) 37.5 Dong et al. (2015) 40.8 Yao et al. (2015) 44.3 Bast et al. (2015) 49.4 Berant et al. (2015) 49.7 Reddy et al. (2016) 50.3 Yih et al. (2015) 52.5 EL/RE (KB) 44.1 EL/RE + Joint 47.1 EL/RE + Refine 47.0 EL/RE + Joint + Refine 53.3 Hybrid (KB + Text + Joint) 53.8 Feng et al. (PKU) Question Answering October 19, / 40

61 在 QALD-6 上的结果 Method WebQ QALD EL/RE (KB) KB + Joint Text Text + Joint KB + Text + Joint (Hybrid) QALD-6 What is the most common language in norway What currency do they use in switzerland What countries does queen elizabeth ii reign What is the best sandals resort in st lucia WebQuestions What is the largest city in the county in which Where was the Father of Singapore born Who is the architect of the tallest building in Japan Feng et al. (PKU) Question Answering October 19, / 40

62 小结选择适当的语义表示形式是否需要与知识库分离? 现有基础分析工具是否足够好? 利用更丰富的资源, 例如, 文本, 多知识库等但是仍然依靠很多人工介入规则模板关键自然语言与知识库之间的映射 ( 实体关系 ) 基于知识库的浅层推断利用常识 Feng et al. (PKU) Question Answering October 19, / 40

63 参考文献 -1 John Zelle and Raymond Mooney. Learning to parse database queries using inductive logic programming. In AAAI 1996 Luke S Zettlemoyer and Michael Collins. Learning to map sentences to logical form: Structured classification with probabilistic categorial grammars. In UAI 2005 Yuk Wah Wong and Raymond J Mooney. Learning synchronous grammars for semantic pars- ing with lambda calculus. In ACL 2007 Percy Liang, Michael I Jordan, and Dan Klein. Learning dependency-based compositional semantics. Computational Linguistics, Qingqing Cai and Alexander Yates. Large-scale semantic parsing via schema matching and lexicon extension. In ACL 2013 Jonathan Berant, Andrew Chou, Roy Frostig, and Percy Liang. Semantic parsing on freebase from question-answer pairs. In EMNLP 2013 Tom Kwiatkowski, Eunsol Choi, Yoav Artzi, and Luke S. Zettlemoyer. Scaling semantic parsers with on-the-fly ontology matching. In EMNLP 2013 Feng et al. (PKU) Question Answering October 19, / 40

64 参考文献 -2 Siva Reddy, Mirella Lapata, and Mark Steedman. Large-scale semantic parsing without question-answer pairs. In TACL 2014 Xuchen Yao and Benjamin Van Durme Information extraction over structured data: Question answering with freebase. In ACL 2014 Jonathan Berant and Percy Liang. Semantic parsing via paraphrasing. In ACL 2014 Junwei Bao, Nan Duan, Ming Zhou, and Tiejun Zhao. Knowledge-based question answering as ma- chine translation. In ACL 2014 Kun Xu, Sheng Zhang, Yansong Feng, and Dongyan Zhao. Answering natural language questions via phrasal semantic parsing. In NLPCC Antoine Bordes, Sumit Chopra, and Jason Weston. Question answering with subgraph embed- dings. In EMNLP 2014 Jonathan Berant and Percy Liang. Imitation learning of agenda-based semantic parsers. TACL Feng et al. (PKU) Question Answering October 19, / 40

65 参考文献 -3 Antoine Bordes, Nicolas Usunier, Sumit Chopra, and Jason Weston. Large-scale simple question answering with memory networks. CoRR, abs/ Wen-tau Yih, Ming-Wei Chang, Xiaodong He, and Jianfeng Gao. Semantic parsing via staged query graph generation: Question answering with knowledge base. In ACL-IJCNLP Li Dong, Furu Wei, Ming Zhou, and Ke Xu. Question answering over freebase with multi- column convolutional neural networks. In ACL- IJCNLP 2015 Yi Yang and Ming-Wei Chang. S-MART: Novel tree-based structured learning algorithms applied to tweet entity linking. In ACL- IJCNLP 2015 Hannah Bast and Elmar Haussmann. More accurate question answering on Freebase. In CIKM 2015 Kun Xu, Siva Reddy, Yansong Feng, Songfang Huang, and Dongyan Zhao. Question Answering on Freebase via Relation Extraction and Textual Evidence. In ACL 2016 Feng et al. (PKU) Question Answering October 19, / 40

66 参考文献 -4 Yuanzhe Zhang, Shizhu He, Kang Liu and Jun Zhao. A Joint Model for Question Answering over Multiple Knowledge Bases, In AAAI 2016 Wen-tau Yih, Matthew Richardson, Christopher Meek, Ming-Wei Chang, and Jina Suh. The Value of Semantic Parse Labeling for Knowledge Base Question Answering, In ACL 2016 Kun Xu, Yansong Feng, Songfang Huang, and Dongyan Zhao. Hybrid Question Answering over Knowledge Base and Free Text. In COLING 2016 Siva Reddy, Oscar Tackstrom, Michael Collins, Tom Kwiatkowski, Dipanjan Das, Mark Steedman, and Mirella Lapata. Transforming Dependency Structures to Logical Forms for Semantic Parsing. In TACL 2016 Wen-tau Yih and Hao Ma, Question Answering with Knowledge Bases, Web and Beyond, In NAACL 2016, Tutorial Feng et al. (PKU) Question Answering October 19, / 40

在北美的山脉返回给用户事实上, 为了得到正确的答案, 问答系统还需要根据山脉高度对候选答案进行排序, 并选择海拔最高的山脉返回给用户该过程通常需要人工编写解析规则对答案进行筛选, 费时费力此外, 由于自然语言描述的多样性, 人们也无法事先穷举所有这样的规则然而事实上,Freebase 这样的

在北美的山脉返回给用户事实上, 为了得到正确的答案, 问答系统还需要根据山脉高度对候选答案进行排序, 并选择海拔最高的山脉返回给用户该过程通常需要人工编写解析规则对答案进行筛选, 费时费力此外, 由于自然语言描述的多样性, 人们也无法事先穷举所有这样的规则然而事实上,Freebase 这样的基于知识的智能问答技术许坤, 冯岩松引言近年来, 信息抽取技术的快速发展使得快速构建大规模结构化半结构化知识库成为可能一大批结构化知识库如雨后春笋般涌现出来, 如 Google Knolwedge Graph (Freebase) Yago,DBpedia 微软 ProBase 搜狗知立方及百度等企业内部的知识图谱等同时, 这些大规模知识库也被应用于关联检索个性化推荐知识问答等任务中