Hulan-2014-10-17

来自cslt Wiki
2014年10月17日 (五) 08:22Lr讨论 | 贡献的版本

(差异) ←上一版本 | 最后版本 (差异) | 下一版本→ (差异)
跳转至: 导航搜索

Dialog system

plan to be done

  • Spell mistake : xingchao
  • using ngram to get candidate sentence.
  • order problem : liurong
  • add vsm and BM25 to strengthen the ability of search.
  • Synonyms word to extract information.like "办理"->"办",“办理”
  • the good IDF to suit the government domain.may the word important from Sogou search.
  • tag

plan to discuss

  • Structured Knowledge from duxingkai
  • template structure :{本体类}-{范围,限定}-{动作}:{户口}-{木里县,残疾人}-{办理,补办},办理项模板:{办理流程,如何办理}
  • knowledge map: ontology , instance and action. it is to presented to the user.
  • Structure Knowledge from teacher wang
  • Template information:
  • add the attribute to the template and answer.like 如何办理户口 -> {木里县}:具体咨询当地相关部门。
  • Build Knowledge or computer representation
  • the tree of knowledge representation. currently represention may include the ontology of knowledge in specific(like triples),the strong relation of freebase.
  • transform of static knowledge and dynamic knowledge
  • answer select using node of knowledge ,like 我该怎么办理户口呢" -> [户口] [办理] -> answer."我该怎么办理户口呢,具体流程是什么" -> [户口] [办理] [流程]-> answer.
  • some others.
  • convert customer data to our data(like nanshandata)
  • answer classification
  • question classification