File size: 10,730 Bytes
d195d4f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
import time
import json
import math
import os
import yaml
from typing import Dict, List

from factool.scientific.tool import google_scholar
from factool.utils.base.pipeline import pipeline

class scientific_pipeline(pipeline):
    def __init__(self, foundation_model):
        super().__init__('scientific', foundation_model)

        self.tool = google_scholar()

        with open(os.path.join(self.prompts_path, "claim_extraction.yaml"), 'r') as file:
            data = yaml.load(file, Loader=yaml.FullLoader)
        self.claim_prompt = data['scientific']

        with open(os.path.join(self.prompts_path, 'agreement_verification.yaml'), 'r') as file:
            data = yaml.load(file, Loader=yaml.FullLoader)
        self.verification_prompt = data['scientific']

    async def _claim_extraction(self, responses):
        messages_list = [
            [
                {"role": "system", "content": self.claim_prompt['system']},
                {"role": "user", "content": self.claim_prompt['user'].format(input=response)},
            ]
            for response in responses
        ]
        return await self.chat.async_run(messages_list, List)
    
    async def _check_authors(self, authors):
        messages_list = [
            [
                {"role": "system", "content": self.verification_prompt['system']},
                {"role": "user", "content": self.verification_prompt['user'].format(string1=claim_author, list2=real_author)},
            ]
            for claim_author, real_author in authors
        ]

        return await self.chat.async_run(messages_list, Dict)

    async def _verification(self, claims, responses):
        authors = [(claim['paper_author(s)'], response['author']) for claim, response in zip(claims, responses)]
        check_authors_results = await self._check_authors(authors)
        final_responses = []
        for i, (claim, response) in enumerate(zip(claims, responses)):
            final_response = {
                'generated_paper_title': claim['paper_title'],
                'generated_paper_author(s)': claim['paper_author(s)'],
                'generated_paper_pub_year': claim['paper_pub_year'],
                'actual_paper_title': response['title'],
                'actual_paper_author(s)': response['author'],
                'actual_paper_pub_year': response['pub_year'],
            }

            errors = []
            if final_response['generated_paper_title'].lower() != final_response['actual_paper_title'].lower() and final_response['generated_paper_title'].lower() not in final_response['actual_paper_title'].lower() and final_response['actual_paper_title'].lower() not in final_response['generated_paper_title'].lower() :
                errors.append('wrong_paper_title')
            if check_authors_results[i]['factuality'] == False:
                errors.append('wrong_paper_author(s)')
            if final_response['generated_paper_pub_year'] != final_response['actual_paper_pub_year']:
                errors.append('wrong_paper_pub_year')

            final_response['error'] = errors
            final_response['factuality'] = len(errors) == 0

            final_responses.append(final_response)

        return final_responses

    async def run_with_tool_live(self, samples):
        claims_in_responses = await self._claim_extraction(samples)
        queries_in_responses = []
        evidences_in_responses = []
        verifications_in_responses = []
        for claims_in_response in claims_in_responses:
            queries = [claim['paper_title'] for claim in claims_in_response]
            queries_in_responses.append(queries)
            evidences = [self.tool.run(paper_title) for paper_title in queries]
            evidences_in_responses.append(evidences)
            verifications = await self._verification(claims_in_response, evidences)
            verifications_in_responses.append(verifications)

        return claims_in_responses, queries_in_responses, evidences_in_responses, verifications_in_responses
        
    async def run_with_tool_live_without_claim_extraction(self, claims):
        # claims = [{"paper_title": "A Survey of Modern Authorship Attribution Methods", "paper_author(s)": "Stamatatos, Efstathios", "paper_pub_year": "2013"}, {"paper_title": "BERT", "paper_author(s)": "John Smith", "paper_pub_year": "2020"}]
        papers_titles = [claim['paper_title'] for claim in claims]
        responses = [self.tool.run(paper_title) for paper_title in papers_titles]
        final_response = await self._verification(claims, responses)
        return final_response

    async def run_with_tool_api_call(self, prompts, responses):
        batch_size = 5
        num_batches = math.ceil(len(prompts) / batch_size)

        self.sample_list = [{"prompt": prompt, "response": response, "category": 'scientific'} for prompt, response in zip(prompts, responses)]

        for i in range(num_batches):
            print(i)
            batch_start = i * batch_size
            batch_end = min((i + 1) * batch_size, len(responses))

            claims_in_responses, queries_in_responses, evidences_in_responses, verifications_in_responses = await self.run_with_tool_live(responses[batch_start:batch_end])

            for j, (claims_in_response, queries_in_response, evidences_in_response, verifications_in_response) in enumerate(zip(claims_in_responses, queries_in_responses, evidences_in_responses, verifications_in_responses)):
                index = batch_start + j

                self.sample_list[index].update({
                    'claims': claims_in_response,
                    'queries': queries_in_response,
                    'evidences': evidences_in_response,
                    'claim_level_factuality': verifications_in_response,
                    'response_level_factuality': all([verification['factuality'] if verification != None else True for verification in verifications_in_response])
                })
        return self.sample_list

    async def run_with_tool_dataset(self, annotated_dataset_path: str, with_tool_classified_dataset_path: str, rerun: bool = False, rerun_indices: list = []):
        # Example of a line: 
        # {"paper_title": "A Survey of Modern Authorship Attribution Methods", "paper_author(s)": "Stamatatos, Efstathios", "paper_pub_year": "2013", "label": True / False}
        if rerun == False:
            with open(annotated_dataset_path, 'r') as f:
                data = [json.loads(line) for line in f]
            self.sample_list = [claim for sample in data for claim in sample['claims']]
            rerun_elements = self.sample_list
        else:
            with open(with_tool_classified_dataset_path, 'r') as f:
                data = [json.loads(line) for line in f]
            self.sample_list = data
            rerun_elements = [self.sample_list[i] for i in rerun_indices]

        batch_size = 5
        num_batches = math.ceil(len(rerun_elements) / batch_size) # 5

        for i in range(num_batches):
            print(i)
            batch_start = i * batch_size
            batch_end = (i + 1) * batch_size if (i + 1) * batch_size < len(rerun_elements) else len(rerun_elements)

            responses = await self.run_with_tool_live_without_claim_extraction(rerun_elements[batch_start:batch_end])
            for j, response in enumerate(responses):
                index = batch_start + j if rerun == False else rerun_indices[batch_start + j]
                if response == None:
                    self.sample_list[index]['with_tool_classification'] = 'None'
                    self.sample_list[index]['error'] = 'None'
                else:
                    self.sample_list[index]['with_tool_classification'] = response.get('factuality', 'None')
                    self.sample_list[index]['error'] = response.get('error', 'None')
        
            # save everything after each batch to prevent data loss
            with open(with_tool_classified_dataset_path, 'w') as f:
                for item in self.sample_list:
                    json_str = json.dumps(item)
                    f.write(json_str + '\n')

    async def run_self_check_live(self, fewshot, batch):
        user_prompt_key = 'user_3_shot_CoT' if fewshot else 'user_zero_shot_CoT'
        messages_list = [
            [
                {"role": "system", "content": self.self_check_prompt['system']},
                {"role": "user", "content": self.self_check_prompt[user_prompt_key].format(scientific_literature=response)}
            ]
            for response in batch
        ]
        return await self.chat.async_run(messages_list, Dict)

    async def run_self_check_dataset(self, annotated_dataset_path: str, self_check_classified_dataset_path: str, fewshot: bool = False, rerun: bool = False, rerun_indices: list = []):
        # Example of a line: 
        # {"paper_title": "A Survey of Modern Authorship Attribution Methods", "paper_author(s)": "Stamatatos, Efstathios", "paper_pub_year": "2013", "annotation": True / False}
        data_path = annotated_dataset_path if not rerun else self_check_classified_dataset_path
        with open(data_path, 'r') as f:
            data = [json.loads(line) for line in f]
        self.sample_list = data if rerun else [claim for sample in data for claim in sample['claims']]
        rerun_elements = self.sample_list if not rerun else [self.sample_list[i] for i in rerun_indices]

        batch_size = 5
        num_batches = math.ceil(len(rerun_elements) / batch_size)

        for i in range(num_batches):
            print(i)
            batch_start = i * batch_size
            batch_end = (i + 1) * batch_size
            batch = rerun_elements[batch_start:batch_end]
            batch = [{k:v for k,v in d.items() if k != "label"} for d in batch]

            responses = await self.run_self_check_live(fewshot, batch)
            for j, response in enumerate(responses):
                index = batch_start + j if rerun == False else rerun_indices[batch_start + j]
                if response == None:
                    self.sample_list[index]['self_check_classification'] = 'None'
                    self.sample_list[index]['self_check_reasoning'] = 'None'
                else:
                    self.sample_list[index]['self_check_classification'] = response.get('factuality', 'None')
                    self.sample_list[index]['self_check_reasoning'] = response.get('reasoning', 'None')
        
            # save everything after each batch to prevent data loss
            with open(self_check_classified_dataset_path, 'w') as f:
                for item in self.sample_list:
                    json_str = json.dumps(item)
                    f.write(json_str + '\n')