对齐错误句子和正确句子， 使用编辑距离算法抽取编辑操作
| 47 | return confusion_dict |
| 48 | |
| 49 | class Alignment: |
| 50 | """ |
| 51 | 对齐错误句子和正确句子， |
| 52 | 使用编辑距离算法抽取编辑操作 |
| 53 | """ |
| 54 | |
| 55 | def __init__( |
| 56 | self, |
| 57 | semantic_dict: Dict, |
| 58 | confusion_dict: Dict, |
| 59 | granularity: str = "word", |
| 60 | ) -> None: |
| 61 | """ |
| 62 | 构造函数 |
| 63 | :param semantic_dict: 语义词典（大词林） |
| 64 | :param confusion_dict: 字符混淆集 |
| 65 | """ |
| 66 | self.insertion_cost = 1 |
| 67 | self.deletion_cost = 1 |
| 68 | self.semantic_dict = semantic_dict |
| 69 | self.confusion_dict = confusion_dict |
| 70 | # Because we use character level tokenization, this doesn't currently use POS |
| 71 | self._open_pos = {} # 如果是词级别，还可以利用词性是否相同来计算cost |
| 72 | self.granularity = granularity # word-level or character-level |
| 73 | self.align_seqs = [] |
| 74 | |
| 75 | def __call__(self, |
| 76 | src: List[Tuple], |
| 77 | tgt: List[Tuple], |
| 78 | verbose: bool = False): |
| 79 | cost_matrix, oper_matrix = self.align(src, tgt) |
| 80 | align_seq = self.get_cheapest_align_seq(oper_matrix) |
| 81 | |
| 82 | if verbose: |
| 83 | print("========== Seg. and POS: ==========") |
| 84 | print(src) |
| 85 | print(tgt) |
| 86 | print("========== Cost Matrix ==========") |
| 87 | print(cost_matrix) |
| 88 | print("========== Oper Matrix ==========") |
| 89 | print(oper_matrix) |
| 90 | print("========== Alignment ==========") |
| 91 | print(align_seq) |
| 92 | print("========== Results ==========") |
| 93 | for a in align_seq: |
| 94 | print(a[0], src[a[1]: a[2]], tgt[a[3]: a[4]]) |
| 95 | return align_seq |
| 96 | |
| 97 | def _get_semantic_class(self, word): |
| 98 | """ |
| 99 | NOTE: Based on the paper: |
| 100 | Improved-Edit-Distance Kernel for Chinese Relation Extraction |
| 101 | 获取每个词语的语义类别（基于大词林，有三个级别） |
| 102 | """ |
| 103 | if word in self.semantic_dict: |
| 104 | code = self.semantic_dict[word] |
| 105 | high, mid, low = code[0], code[1], code[2:4] |
| 106 | return high, mid, low |
no outgoing calls
no test coverage detected