forked from zaclys/searxng
82 lines
2.2 KiB
Python
82 lines
2.2 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-or-later
|
|
# lint: pylint
|
|
"""Semantic Scholar (Science)
|
|
"""
|
|
|
|
from json import dumps, loads
|
|
from datetime import datetime
|
|
|
|
about = {
|
|
"website": 'https://www.semanticscholar.org/',
|
|
"wikidata_id": 'Q22908627',
|
|
"official_api_documentation": 'https://api.semanticscholar.org/',
|
|
"use_official_api": True,
|
|
"require_api_key": False,
|
|
"results": 'JSON',
|
|
}
|
|
|
|
paging = True
|
|
search_url = 'https://www.semanticscholar.org/api/1/search'
|
|
paper_url = 'https://www.semanticscholar.org/paper'
|
|
|
|
|
|
def request(query, params):
|
|
params['url'] = search_url
|
|
params['method'] = 'POST'
|
|
params['headers']['content-type'] = 'application/json'
|
|
params['data'] = dumps(
|
|
{
|
|
"queryString": query,
|
|
"page": params['pageno'],
|
|
"pageSize": 10,
|
|
"sort": "relevance",
|
|
"useFallbackRankerService": False,
|
|
"useFallbackSearchCluster": False,
|
|
"getQuerySuggestions": False,
|
|
"authors": [],
|
|
"coAuthors": [],
|
|
"venues": [],
|
|
"performTitleMatch": True,
|
|
}
|
|
)
|
|
return params
|
|
|
|
|
|
def response(resp):
|
|
res = loads(resp.text)
|
|
results = []
|
|
|
|
for result in res['results']:
|
|
item = {}
|
|
metadata = []
|
|
|
|
url = result.get('primaryPaperLink', {}).get('url')
|
|
if not url and result.get('links'):
|
|
url = result.get('links')[0]
|
|
if not url:
|
|
alternatePaperLinks = result.get('alternatePaperLinks')
|
|
if alternatePaperLinks:
|
|
url = alternatePaperLinks[0].get('url')
|
|
if not url:
|
|
url = paper_url + '/%s' % result['id']
|
|
|
|
item['url'] = url
|
|
|
|
item['title'] = result['title']['text']
|
|
item['content'] = result['paperAbstract']['text']
|
|
|
|
metadata = result.get('fieldsOfStudy') or []
|
|
venue = result.get('venue', {}).get('text')
|
|
if venue:
|
|
metadata.append(venue)
|
|
if metadata:
|
|
item['metadata'] = ', '.join(metadata)
|
|
|
|
pubDate = result.get('pubDate')
|
|
if pubDate:
|
|
item['publishedDate'] = datetime.strptime(pubDate, "%Y-%m-%d")
|
|
|
|
results.append(item)
|
|
|
|
return results
|