Entity-Relation Extraction as Multi-turn Question Answering

• 發現領域內的問題

  • task formalization level

    • 三元組本身的知識表達能力有限，比如，Musk case裡的hierarchical dependency，時間、地點、職位、人物需要在一個更高維度的空間表達

  • algorithm level

    • 輸入：a raw sentence with two marked mentions
      輸出：whether a relation holds between the two mentions
    • hard for neural models to capture all the lexical, semantic and syntactic cues in this formalization
      • (1) entities are far away
        ;
        (2) one entity is involved in** multiple triplets;**
        or (3) relation spans have overlaps
• related work
  • Extracting Entities and Relations
    • pipelined approach
      • 優點：**flexibility **of integrating different data sources and learning algorithms
      • 缺點：suffer significantly from error propagation
    • joint approach
      • through various dependencies
        • constraints solved by integer linear programming
      • card-pyramid parsing
      • global probabilistic graphical models
      • structured perceptron with efficient beamsearch
      • table-filling approach
        • search orders in decoding and global features
      • shared parameters，end-to-end approach that extract entities and their relations using neural network models
        • neural tagging model，multi-class classification model based on tree LSTMs
      • multi-level attention CNNs
      • seq2seq models to generate entity-relation triples
    • reinforcement learning or Minimum Risk Training
      • a global loss function to jointly train the two models under the framework work of Minimum Risk Training
      • hierarchical reinforcement learning
  • Machine Reading Comprehension, predicting answer spans given context
    • 主要做的是 extract text spans in passages given queries
    • 一種思路可以簡化成 two multi-class classification tasks
    • 另一種思路，對於multi-passage MRC, directly concatenating passages, first rank the passages and then run single-passage MRC on the selected passage
    • 其他有用的：Pretraining methods like BERT or Elmo
    • 趨勢：a tendency of casting non-QA NLP tasks as QA tasks
    • 具體的比如
      • BiDAF、QANet
• 本論文的工作

  • inspiration 來源

    • identifying the relation between two predefined entities and the authors formalize the task of relation extraction as a single-turn QA task
      Levy et al. (2017). Levy et al. (2017) and McCann et al. (2018)

  • Idea

    • model hierarchical tag dependency in multi-turn QA， identifying answer spans from the context
      each entity type and relation type is characterized by a question answering template, and entities and relations are extracted by answering template questions
    • question query encodes
    • jointly modeling entity and relation
    • exploit the well developed machine reading comprehension (MRC) models
    • multi-step reasoning to construct entity dependencies

  • advantages

    • capture the** hierarchical dependency of tags**, progressively obtain the entities we need for the next turn , closely akin to** the multi-turn slot filling dialogue system**
    • the question query encodes important** prior information** for the relation class we want to identify
    • the QA framework provides a natural way to simultaneously extract entities and relations: most MRC models support outputting special NONE tokens, indicating that there is no answer to the question

  • dataset

    • ACE04, ACE05 and the CoNLL04 corpora
    • a newly developed dataset RESUME in Chinese
      

    extract biographical information of individuals from raw texts. The construction of structural knowledge base from RESUME requires four or five turns of QA
    * 最大的特點：one person can work for **different **companies during **different **periods of time and that one person can hold **different **positions in **different **periods of time for the **same **company

  • model

    • 分解成了兩個子任務：a multi-answer task for head-entity extraction + a single-answer task for joint relation and tail-entity extraction
    • 第一階段：head-entity extraction，extract this starting entity, we** transform each entity type to a question** using EntityQuesTemplates
      • 這個階段抽取到的不一定就是head entities
    • 第二階段：The relation and the tail-entity extraction，定義了relations chain用於multi-turn QA，因為一些的抽取取決於其他的抽取

    • Generating Questions using Templates

      • type-specific
      • natural language questions or pseudo-questions

    • Extracting Answer Spans via MRC

      • backbone：BERT，基於多輪問答的問題對Traditional MRC models做了調整：predict a BMEO (beginning, inside, ending and outside) label
      • Training and Test
        • L = ( 1 − λ ) L ( head-entity ) + λ L ( tail-entity,rel ) \mathcal{L} = ( 1 - \lambda ) \mathcal{L} ( \text { head-entity } ) + \lambda \mathcal{L} ( \text { tail-entity, rel} ) L=(1−λ)L(head-entity)+λL(tail-entity,rel)
        • 訓練的時候兩個共享引數，測試的時候head-entities and tail-entities are extracted separately，λ 控制兩個子任務的 tradeoff

    • Reinforcement Learning

      • 一個turn中抽取的答案還會影響downstream turns，也影響later accuracies
      • 由於multi-turn dialogue generation的結果比較好，所以打算也用reinforcement learning
        (Mrkˇsi´c et al., 2015; Li et al., 2016; Wen et al., 2016)
      • action：selecting a text span in each turn
      • policy：probability of selecting a certain span given the question and the context
        p ( y ( w 1 , … , w n ) = answer ∣ question, s ) = p ( w 1 = B ) × p ( w n = E ) ∏ i ∈ [ 2 , n − 1 ] p ( w i = M ) \left. \begin{array} { l } { p ( y ( w _ { 1 } , \ldots , w _ { n } ) = \text { answer } | \text { question, } s ) } \\ { = p ( w _ { 1 } = \mathrm{B} ) \times p ( w _ { n } = \mathrm{E} ) \prod _ { i \in [ 2 , n - 1 ] } p ( w _ { i } = \mathrm{M} ) } \end{array} \right. p(y(w1​,…,wn​)=answer∣question,s)=p(w1​=B)×p(wn​=E)∏i∈[2,n−1]​p(wi​=M)​
      • Reward：特定句子，用正確抽取的triples作為獎勵 ，maximizes the expected reward E π [ R ( w ) ] E _ { \pi } [ R ( w ) ] Eπ​[R(w)]，通過sampling from the policy π π π來近似
      • gradient的計算：likelihood ratio:
        $\nabla E ( \theta ) \approx [ R ( w ) - b ] \nabla \log \pi ( y ( w ) | \text { question } s ) ) $
        b 是 baseline value（所有之前獎勵的平均）, 每輪答對就獎勵+1，final reward是所有輪次積累的獎勵
      • policy networks initialization：pre-trained head-entity and tail-entity extraction model
      • experience replay strategy：for each batch, half of the examples are simulated and the other half is randomly selected from previously generated examples.
      • strategy of curriculum learning 用在了 RESUME dataset 上，gradually increase the number of turns from 2 to 4 at training

  • Experimental Results（ SOTA results）

    • 指標：micro-F1 scores, precision and recall
    • 自己搞的資料集RESUME上的結果
      • 先確定一個baseline：tagging+relation，entity+dependency
      • entities 部分用BERT tagging models，relations部分用 CNN to representations output by BERT transformers
      • 這個任務akin to a dependency parsing task at the tag-level rather than the word-level
      • 具體做法：通過BERT tagging model給每個word分配tagging labels, 然後調整SOTA dependency parsing model Biaffine來construct dependencies between tags（jointly trained）
    • 比較常用的ACE04, ACE05 and CoNLL04上的結果

  • Ablation Studies

    • Effect of Question Generation Strategy：natural language questions會比pseudo-questions更好，因為提供的是more fine-grained semantic information

    • Effect of Joint Training： λ 間隔0.1地測了10個數據，其中，entity-extraction並不是在 λ = 0 的時候最好（說明relation extraction的部分可以提升entity extraction的效果）

    • Case Study：和SOTA MRT model對比，可以識別出相隔比較遠的實體; 當句子裡包含兩對相同關係的時候，也可以識別出來

  • 後續發展空間

    • could easilty integrate reinforcement learning (just as in multi-turn dialog systems)