ccdv commited on
Commit
ec0eca8
·
1 Parent(s): 00de2d6

add _preprended

Browse files
Files changed (2) hide show
  1. README.md +4 -1
  2. mediasum.py +32 -8
README.md CHANGED
@@ -22,11 +22,14 @@ This dataset is compatible with the [`run_summarization.py`](https://github.com/
22
 
23
  # Configs
24
  4 possibles configs:
25
- - `roberta` will concatenate documents with "\</s\>" (default)
26
  - `newline` will concatenate documents with "\n"
27
  - `bert` will concatenate documents with "[SEP]"
28
  - `list` will return the list of documents instead of a single string
29
 
 
 
 
30
  ### Data Fields
31
 
32
  - `id`: paper id
 
22
 
23
  # Configs
24
  4 possibles configs:
25
+ - `roberta` will concatenate documents with "\</s\>"
26
  - `newline` will concatenate documents with "\n"
27
  - `bert` will concatenate documents with "[SEP]"
28
  - `list` will return the list of documents instead of a single string
29
 
30
+ Add `_prepended` to config name to prepend the speaker name before each dialogue: `speaker: text` \
31
+ Default is `roberta_prepended` (compatible with BART).
32
+
33
  ### Data Fields
34
 
35
  - `id`: paper id
mediasum.py CHANGED
@@ -62,9 +62,29 @@ class MediaSumSummarizationDataset(datasets.GeneratorBasedBuilder):
62
  version=datasets.Version("1.0.0"),
63
  description="MediaSum dataset for summarization, document",
64
  ),
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
65
  ]
66
 
67
- DEFAULT_CONFIG_NAME = "roberta"
68
 
69
  def _info(self):
70
  # Should return a datasets.DatasetInfo object
@@ -102,12 +122,12 @@ class MediaSumSummarizationDataset(datasets.GeneratorBasedBuilder):
102
 
103
  def _generate_examples(self, filepath):
104
  """Generate MediaSumSummarization examples."""
105
- if self.config.name == "newline":
106
  join_ = "\n"
107
- elif self.config.name == "roberta":
108
  join_ = "</s>"
109
- elif self.config.name == "bert":
110
- join_ = "[SEP]"
111
 
112
  with open(filepath, encoding="utf-8") as f:
113
  for id_, row in enumerate(f):
@@ -118,9 +138,13 @@ class MediaSumSummarizationDataset(datasets.GeneratorBasedBuilder):
118
  'document': List[str],
119
  """
120
 
121
- document = data["utt"]
 
 
 
 
122
 
123
  if self.config.name != "list":
124
- document = join_.join(document)
125
  summary = data["summary"]
126
- yield id_, {"document": document, "summary": summary}
 
62
  version=datasets.Version("1.0.0"),
63
  description="MediaSum dataset for summarization, document",
64
  ),
65
+ MediaSumSummarizationConfig(
66
+ name="newline_prepended",
67
+ version=datasets.Version("1.0.0"),
68
+ description="MediaSum dataset for summarization, concat sections",
69
+ ),
70
+ MediaSumSummarizationConfig(
71
+ name="roberta_prepended",
72
+ version=datasets.Version("1.0.0"),
73
+ description="MediaSum dataset for summarization, document",
74
+ ),
75
+ MediaSumSummarizationConfig(
76
+ name="bert_prepended",
77
+ version=datasets.Version("1.0.0"),
78
+ description="MediaSum dataset for summarization, document",
79
+ ),
80
+ MediaSumSummarizationConfig(
81
+ name="list_prepended",
82
+ version=datasets.Version("1.0.0"),
83
+ description="MediaSum dataset for summarization, document",
84
+ ),
85
  ]
86
 
87
+ DEFAULT_CONFIG_NAME = "roberta_prepended"
88
 
89
  def _info(self):
90
  # Should return a datasets.DatasetInfo object
 
122
 
123
  def _generate_examples(self, filepath):
124
  """Generate MediaSumSummarization examples."""
125
+ if "newline" in self.config.name:
126
  join_ = "\n"
127
+ elif "roberta" in self.config.name:
128
  join_ = "</s>"
129
+ elif "bert" in self.config.name:
130
+ join_ = " [SEP] "
131
 
132
  with open(filepath, encoding="utf-8") as f:
133
  for id_, row in enumerate(f):
 
138
  'document': List[str],
139
  """
140
 
141
+ documents = data["utt"]
142
+
143
+ if "_prepended" in self.config.name:
144
+ names = data["speaker"]
145
+ documents = [name + ": " + document for name, document in zip(names, documents)]
146
 
147
  if self.config.name != "list":
148
+ documents = join_.join(documents)
149
  summary = data["summary"]
150
+ yield id_, {"document": documents, "summary": summary}